I got the chance to check out Claude Code, Anthropic’s new agentic coding system in the terminal while in research preview. But it is now open for all to check out here . I’m going to give you a glimpse of what Claude Code is, what its capabilities are, my experience building something from scratch on it, the final output, and finally, who this really is for. This post is a bit more technical than usual, but bear with me.

What is Claude Code

First, what is Claude Code, and what is it good for? Claude Code is almost a Cursor-esque agent that you can access in a Terminal. You can point it to a directory where your code lies and then run the `claude` command to activate it.

Claude Code’s key capabilities include:

• Editing files and fixing bugs across your codebase

• Answering questions about your code’s architecture and logic

• Executing and fixing tests, linting, and other commands

• Searching through git history, resolving merge conflicts, and creating commits and PRs

Most of the features that you see are possible in a tool like Cursor, but some of them like searching through git history and resolving merge conflicts, is interesting, and I’m not sure Cursor does it out of the box yet.

As a non-technical PM who has learnt to vibe code, I tried out an example of creating a Roast My LinkedIn Profile app from scratch. Which involved creating new files, folders, editing code, running bash commands, and installing packages.

Let’s take a look at the key parts of Claude Code with this example.

Interface

Like I mentioned previously, the interface is nothing but a Terminal window, which may seem daunting to non-devs but is bread and butter for devs. There are no fancy UI elements like the ones that you would find in an online AI coding app or an IDE like Cursor or Windsurf. You can’t open the files directly unless you use another Terminal window and use something like vim or some other IDE.

So it’s just text prompting, which is a limitation because it doesn’t allow screenshots. You can’t take a screenshot of the error and expect Claude Code to work on it. The external capabilities like images, web search, and tagging files are limited.

Something as small as using text expansion snippets doesn't work in the terminal, so I wasn't able to use my pre-made text expansion snippets so that's another downside of Claude Code.

There are a lot of benefits to using an IDE over the Terminal. Like extensions, visual file management, and now AI chat tools. Only hardcore coders would use the terminal and Vim for coding.

So overall, not a great experience. Even if you're a dev, you would prefer an IDE like Cursor.

Output UI

Here is what the final output was like. As you can see, it clearly didn't create a great UI. It’s left aligned, the buttons are not styled, there’s no visual hierarchy except for the title, and the text box for the link is not wide enough to show the entire link. There’s no semblance of basic aesthetics.

The UI that is possible out of the box using Sonnet 3.7 in the Claude app, in Cursor, or in tools like Lovable and Bolt are far better. It's a dramatic improvement over 3.5. It could be some of the Cursor Rules that I have set or something that the tools are doing themselves, but they do a much better job. So, if I was building something completely from scratch, I wouldn’t use Claude Code to do it.

It might be a good option to update an existing codebase or just the backend code.

Get in touch with me on X/Twitter

P.S. if you're building something interesting, say hi to me on X/fka Twitter. I’m always happy to help.

Pricing

It's also relatively expensive to pay $5 to create a very simple prototype that I could have probably used an app like Lovable, Bolt and just using a free plan got it done or using cursor which would barely take a fraction of the fast request allowance for the entire month by paying $20 a month. So, from a price consideration, it is better to use Cursor.

Final Thoughts

Overall, at the moment, I cannot recommend Claude Code over Cursor or other tools for most non-technical or technical folks.

Unless you have an existing codebase that you want to experiment with and try out Claude’s agentic capabilities, it may be worthwhile checking out. But for the rest of us, probably not now.

I'm sure some of the learnings that Anthropic has with Claude Code can go into the next models and, as a result, even further improve what Cursor has to offer.

I truly believe there is a fundamental shift for AI companies like OpenAI and Claude to focus on the app layer rather than just sticking to the foundational model layer. Foundational models drive a lot of the end-user apps like Cursor, Bolt, Lovable, Replit, etc that are generating millions of dollars of revenue.

Cursor, for example, only picked up in usage and popularity after Claude Sonnet 3.5 shipped and made the most of Anthropic’s capabilities. What if Anthropic captures some of that App layer value for themselves by building a better Cursor? I think that’s where Claude Code fits in, by being the first step at the Agentic App layer. Even though a terminal may not be the best interface, even for someone who is extremely experienced at vibe coding, it's a stepping stone in the right direction.

That’s it for today!

Need product or AI expertise? Book a paid consulting call with me to discuss your challenges and find solutions. Reply to this email or email kavir @ hey dot com to schedule. Learn more about how I can help.

Smash that like button!

Before we begin, if you're a reader of the newsletter just hit the like button above, or reply to this email or comment so I know this is reaching you and you're still interested in this content. It would really help me understand my audience better. Let’s get started!

When ElevenLabs launched their AI voice agent in December, I experimented with POCs for both a travel planning helper and a sales agent. However, the initial pricing was prohibitive, leading me to put these experiments on hold.

But now, ElevenLabs reduced their costs by 50%, bringing the price down to $0.08 per minute—a much more affordable range. I've since revisited the project, and I'm excited to share my implementation experience and demo with you.

What ElevenLabs and other similar voice conversational agents allow you to deploy an AI voice assistant to interact with your customers or end users through a bunch of steps. First you define a system prompt for the interaction, allow you to associate your contextual docs, decide on a voice from options, connect tools for retrieval and storage of data from the call, and configure use cases when they are called — inbound or outbound. Let’s expound on these elements.

System Prompt Design

Drawing from my understanding of sales conversations and business context, I developed a comprehensive system prompt that covers:

• Conversational tone and flow

• Pricing and discount negotiation strategies

• Objection handling frameworks

• Data collection requirements

• System limitations and safety checks

If you want to see the full prompt, get in touch with me on X.

RAG and documents

For this straightforward use case, I opted to include all necessary information directly in the system prompt rather than implementing a full RAG (Retrieval-Augmented Generation) system. This approach helped minimize latency while maintaining functionality. However, for more complex use cases requiring extensive domain knowledge, implementing a proper document-based knowledge base would be recommended.

Voice localization features

ElevenLabs offers an extensive collection of localized voices spanning multiple regions (including UK, Australian, and Indian English) alongside standard US English. This localization capability is crucial for creating authentic customer interactions and represents a key differentiator for ElevenLabs. For companies implementing AI voice agents, voice localization should be considered a critical feature for market-specific deployments.

Tool Usage

ElevenLabs allows you to integrate function calling.

In my experience, it’s not foolproof, so I am still figuring out the right way. I am using Gemini Flash 2.0 which is recommended by ElevenLabs and I was generally getting good results in terms of communication.

I also created a tool that before the end of the call called a webhook which transcribed, used the transcript to pass the information and categorize it into different fields like name, email, and mobile.

Output

Putting it all together, I was able to deploy a widget on the website to help leads hold a natural conversation with the AI in a local accent with a structured conversation flow, objection handling, pricing negotiations, data collection, lead scoring, and data parsing and saving to CRM through a webhook.

The lead scoring is done basis what the lead has said in terms of urgency or being okay with the pricing and their enthusiasm for going ahead. This is helpful in terms of human follow-ups.

I could add another element of sending an automated payment link through either email or WhatsApp and then complete the loop but at the moment I'm just keeping it manual so that any errors can be rectified before that gets sent. But yeah this is very promising. So far one or two customers have spoken to the AI and have not either realized or implied with it.

Here is the widget on the website.

This is the summary of the conversation with the user.

And here is the webhook sent to the team in Slack with the details.

Finally, here’s the demo that I teased at the beginning of the article.

Comparison with OpenAI’s conversational agent

I also tried out the same use case with OpenAI’s conversational agent. Here is what I thought:

Advantages:

• Reduced latency through Omni model architecture (text and voice in one model - no need of a separate transcribing step)

• Superior handling of conversation interruptions

• GPT-4 provides reliable calculation accuracy (though at higher cost)

Limitations:

• GPT-4-mini exhibits calculation inaccuracies for discount computations

• Limited voice options with no international accent support

• Higher operational costs at scale

Implications

Currently, ElevenLabs is priced at $0.08 per minute or around ₹30-40 for a 5-minute call. OpenAI’s conversational agent is slightly more expensive at $0.16 per minute.

But I expect these prices to come down over time, and AI voice agents to become more sophisticated with lower latency to transform a ton of call center operations over time. AIs can work 24x7 with more script adherence, no emotional involvement, and have better customer outcomes.

This tech can replace existing use cases, as well as enable a lot of newer use cases that are programmatic and earlier were not possible for whatever reason.

These calls are an order of magnitude better than current robocalls that are just recorded intrusive messages.

Again, here’s the demo of the AI sales call.

That’s it for today!

In case you or someone you know would like to have a paid consulting call with me about product or AI, get in touch with me by replying to this email or emailing me at kavir @ hey dot com. You can check out how I can be useful here. Alright then, talk soon!

Smash that like button, don’t be shy!

Before we begin: In case you or someone you know would like to have a paid consulting call with me about product or AI, get in touch with me by replying to this email or emailing me at kavir @ hey dot com. You can check out how I can be useful here.

But first, what are agents?

Agents as defined by Anthropic, are: systems where LLMs dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks.

This is different from a workflow that just uses LLMs and AI to do a task and achieve a goal. I will cover more about agents in a subsequent piece, subscribe, like, and comment if you would like to read that.

Zapier Agents

I’ve been using Zapier Agents (earlier Zapier Central) for the last 4 months for one particular use case and it’s proving to be really useful.

In a nutshell, Zapier agent is an AI-powered automation tool that combines Zapier's integration capabilities across 7000+ apps and platforms with LLMs to perform complex and contextual tasks like drafting emails or creating content.

What did I use it for?

My use case was to draft Gmail replies contextually for a service business. The problem statement was that it takes time for the team to prepare and send replies. Some team members are not that proficient in English and hence have to rely on the lead to review the message and send it.

I decided not to send these automatically but to create drafts to keep the human in the loop who can review and send them.

But it still saves a lot of time — so if this AI bot writes 10 email drafts a day, which would've normally taken 5 minutes per email, bringing it down to 1 minute per email, it still saves 40 minutes per day - 80% time-saving. And many times replies are not sent because it's just cumbersome to think and write.

Setup

I set it up in just a few minutes and it’s saving at least an hour a day for the team.

Using LLMs and RAG, you can set up common rules, scenarios, and context and prepare response.

For e.g. I had a message template Google Sheet that had most of the scenarios and how to respond. You could have it in Google Docs, Notion, or anywhere else.

Apart from that, in the system instructions you can add a lot of detail on the type of your business, and how you want the bot to respond - for e.g. empathetically and politely.

Here are the steps:

1. Go to Zapier Agents and create an account. You get free 400 tasks a month.

2. Create a new behavior

3. Connect all your accounts

4. Add a trigger (could be a schedule, or chat, or email received) - I have kept it at multiple intervals during the day to batch process

5. Add your instructions (use ChatGPT or Claude to create a good prompt

6. Only have it run on the emails from other people and not marketing emails - so add category:primary or other filters. And you can filter out some auto-email notifications like Zoom, Tally, Google Calendar, etc And add a filter for newer_than. Because it cant run on all your emails from eternity. For e.g newer_than:12h

7. Add context docs

8. Test it out properly to ensure it is doing what it is intended to do, if not tweak the instructions and the email filters

9. And finally publish!

Other Applications

What other things you can do:

• Create a sales outreach agent

• Lead enrichment agent

• Sales prep agent

• Marketing blog draft creator

• Search the web and add to CRM

Let me know in the replies and comments — what would be your use cases for agents in your work, company, or personal life?

📘 Read of the week

Understand what agents really are, in this guide by Anthropic.

Thanks for reading till the end!

Another reminder: In case you or someone you know would like to have a paid consulting call with me about product or AI, get in touch with me by replying to this email or emailing me at kavir @ hey dot com. You can check out how I can be useful here. Alright then, talk soon!

Here is The Dispatch #2

DeepSeek

DeepSeek is a new open source reasoning model that's come from China and was apparently trained and built for a budget of less than $6M. That is several orders of magnitude less than what the other SOTA models from OpenAI, Anthropic, and Gemini are trained for. And it's matching and even exceeding benchmarks of these models.

There are a few things interesting about this model as a consumer and from a macro perspective. From a consumer point of view, it's a reasoning model that is available for free. o1 and o1 mini are only on the paid ChatGPT plus accounts. This exposes more people to the reasoning aspect of LLMs and that's a gamechanger.

Being able to see the unfiltered thought process of an LLM to consider all cases and then answer a question is eye opening. It personalizes the LLM and you can see how it has got something correct or how there are gaps in thinking. This builds trust in an LLM. Seeing the train of thought also allows you to tweak the subsequent responses to address the gaps or add more context to get an even more nuanced response.

See an example of DeepSeek:

Whereas o1 and o1 mini offer a very summarized version of this thinking.

Here is another example of the level of thinking that DeepSeek does.

But one thing is for sure, it's hard to go back to using LLMs the old way after seeing the train of thought.

I am a Claude Sonnet 3.5 super user. And am just waiting for it to launch it's reasoning model. Till then I have used custom instructions to simulate this thinking process in the output. Here is how you can set it up.

I am yet to see applications apart from math and coding, where a reasoning model like o1 or deepseek-r1 can outperform GPT4o or Claude Sonnet 3.5. For example for some writing tasks I've stuck to gpt4o for simplicity.

Deepseek r1 will be a game changer in tools like Cursor, Replit, etc. They will make coding more effective, because you can see what the LLM is considering while writing the code.

Anyway, I encourage you to try out Deepseek for yourself.

From a Macro lens, it's a fundamental shift in the market where it is now clear that new companies without much of a budget can reproduce SOTA models thanks to DeepSeek's breakthroughs at a small budget and open source tech.

There are a lot of takes on Twitter/X. but I found Yishan's take that it's not a Sputnik moment where the USSR (Soviet Union) made breakthroughs in space tech before the US and the western world could, but more like Google in 2004 when it launched their cost-effective and optimized search engine, took market share, and went public.

Read the rest of the thread here

🚀 Spotlight: AI Calorie Tracker

I’ve always struggled to count my calories and hence have a consistent diet. To solve my own problem, I've launched an AI calorie tracker app where you take a picture of your food and it automatically tracks calories. It's reasonably accurate — I tried to make it more accurate through prompt engineering. You can also add height, weight, goals, and accordingly calorie and protein targets will be calculated.

This is something that I built for myself, please try it out and give me feedback!

🍱 Check out the AI Calorie tracker here

Lastly, a quick biohacking update: I'm going to the sauna frequently, and I think it's really helping me mentally and physically. My sleep has improved tremendously - I'm actually sleeping more than I normally do, and it's giving me mental clarity and peace in the morning.

That's it for today, thanks for reading! Have you tried DeepSeek? What's your take on it? Reply or comment below.

In case you or someone you know would like to have an paid consulting call with me about product or AI, get in touch with me by replying to this email. You can check out my value add here. Alright then, talk soon!

This is a series where I cover what I find interesting in the world of tech, AI, and product — things that I think you, dear reader, are interested in. It’s not going to be a list like other curated newsletters. I’m going to talk about my personal experiences with the tools and tech that I talk about — and more importantly what you can learn from it.

I’ve been writing this newsletter since June 2020, on and off, as you might know. I needed a format that would be doable on a weekly basis. Many of the long form pieces that I write take longer than a week, but I wanted to keep a habit of writing and publishing weekly. Those longer pieces will still continue, and hopefully become more frequent.

With that out of the way, let’s get started:

Sora 🎥

Sora is the new AI video generation tool that OpenAI just shipped. Having tried out other AI video generation tools like Hailio, Kling, Runway and others, I thought I would have a good frame of references to compare Sora.

And I’m not impressed.

Sora is definitely a polished product, but the models are not as good as the other platforms I’ve tested.

Here is an example.

Either I am doing it wrong or I feel that the models are not as good as the competition. I used Claude to generate a prompt, but I feel that’s too detailed for sora.

If you look at their featured videos, those videos are with very minimalist prompts. That shows that you can’t a high level of customization. You just have to trust the models to output something decent.

And again in their featured videos, you will find more landscapes, cartoon animations, rather than people and faces. At least in the ChatGPT Plus account, It doesn’t allow to create videos with people and faces yet, even if those images are AI generated. I don’t know whether this is different in the ChatGPT Pro account.

There are other limitations with the product, that I can’t select what area of the image is cropped.

My key insight of AI video generation is to always start with an image first. so it’s more like image to video, rather than text to video. You can use AI image generation tools like Flux, or even Grok for this. This hasn’t applied well to Sora yet, because I haven’t got satisfactory output with either text to video or image to video.

There are a few interesting product takes from their launch:

• they started with a way more polished product

• Sora is not chatgpt.com/sora, it is its own thing with sora.com and no API available yet

• they started with a featured section to showcase and probably cherry pick the best examples of what is possible. also as a learning tool to understand the way the prompting is done to get decent results

• they have all bells and whistles in terms of features - more than the competition at launch time

• OpenAI is now a product company rather than a research company

Alright enough of Sora, let’s move on to the next topic:

Sales Recording Analysis tool 🤑

I was looking to evaluate one of the sales people in my team. Most of the sales conversations happen over phone calls in this business. I was thinking that it’s a time consuming task to hear out entire phone conversations and give feedback. You could do this for a small sample of calls, but if you need to do it for most, it won't happen. AI to the rescue.

What I decided was to first download the recording, then put it into a whisper tool (local or online) to get the transcription, and after that run it in Claude to get the analysis.

The Whisper tools that run locally on the Mac were not that that accurate because it was using a smaller model. The cloud ones had a limit of one free transcription and then a paid subscription.

Why should I pay for a subscription for this when I can (now) build it myself. Enter v0 and Cursor. And here is the output:

The user workflow is like this:

• upload the mp3 of the recording.

• Whisper transcribes it.

• After it transcribes the recording, it runs an analysis of the transcription that provides insight into “what was done well” and “what can be improved” along with a “likelihood of conversion”

Building with AI tools like v0 and Cursor is honestly addictive. You give plain English instructions and it writes code to build a real app and feature.

Here is my workflow on both v0 and Cursor.

Napkin AI 🗳️

Another useful tool that I came across recently is the Napkin.AI tool that creates diagrams from text.

It is helpful to create visuals for:

• presentations

• content marketing

• blog posts

It is especially helpful for content marketing when you're trying to post everyday from a company account and you want visuals to go with text. Because on all social media networks - text with images perform better than just text.

Most of them resemble McKinsey-level diagrams found in Microsoft PowerPoint, such as funnels, pie charts, and process diagrams.

I feel the color combinations could’ve definitely been better. For example, it doesn’t get contrast at all.

Currently free in beta! Check it out.

That’s a wrap. Let me know if you like this format. Or you prefer the deep dives. Would love to hear from you. I respond to each reply I get.

Talk to you soon!

Press the ♥️ button if you liked this edition.

👇🏽

✨ Special offer: If you would like some insight on how to use AI in your company or work, reply to this email, and I’ll help you out. ✨

After writing about the potential use cases of customer support in the age of LLMs, I want to write about the application of AI in sales.

The advent of Large Language Models (LLMs) is reshaping the landscape of sales, offering unprecedented tools for efficiency and personalization. While AI presents exciting opportunities, its true power lies in augmenting rather than replacing human salespeople. In this article, I argue that the most effective sales strategies in the age of LLMs will combine AI's data-processing capabilities with the irreplaceable human elements of empathy, relationship-building, and nuanced understanding of the customer. Let’s dive in!

My Experience

Throughout my entire career as a product person, I have not directly worked in sales. However, I have worked closely with business heads, founders, and sales professionals in the field at B2B startups. For example, I worked closely with the founders and the DevRel Lead at a funded startup, where we were selling access to a large developer audience to dev tools companies by way of hackathons or sponsorships. Before that, I worked directly with the founder of an MIT Labs-funded startup that was selling AI-driven HR Tech and pivoted to Customer Support quality SaaS tools to SMBs and midmarket companies.

Most B2B product companies, even if their growth is product-led, like Notion, Coda, and others — should have a sales-assist approach to them. Because eventually, you will need sales if you have to go upmarket from SMBs to midmarket to enterprise.

What is good sales?

Good sales at its core involves understanding the customer’s needs, challenges, and goals and then pitching the product as a solution to those problems. This is also known as consultative sales. And what aids this is the fact of connecting with another person and building a relationship among shared goals and interests.

Good sales is not about manipulation or pressure tactics. Instead, it's about aligning the interests of the seller and the buyer. When done right, both parties benefit: the customer gets a solution that improves their business or life, and the salesperson (and their company) gains a satisfied customer and revenue.

I think AI can help tremendously in these two aspects.

The Sales Funnel

Let’s discuss the application of AI by discussing it at each stage of the sales process into different parts of a funnel — prospecting, qualifying, proposal, and closing.

(image)

Before diving into the funnel stages, it's crucial to define the Ideal Customer Profile (ICP) according to the product vision and features. It serves as the foundation for all subsequent sales activities, ensuring that efforts throughout the funnel are targeted toward the most promising and valuable potential.

Ideal Customer Profile

Creating an ICP involves identifying specific attributes that define the most valuable customers. These typically include:

• Demographics: Industry, company size, location.

• Technographics: Technologies the company uses which may complement or compete with your product.

• Behavioral Characteristics: Their buying patterns, brand loyalty, and product usage.

• Psychographics: The company’s values, culture, and priorities.

• Budgetary Constraints: Their spending power and willingness to invest in solutions like yours.

An example of an ICP could be:

A SaaS company that offers project management tools might define its ICP as mid-sized tech companies with 100-500 employees, based in North America, that use agile methodologies and have a high adoption rate of cloud technologies. These companies are likely to value collaborative, scalable tools with robust integration capabilities.

Where AI can help:

You could give details of your current set of customers, customers that have churned, and let it define an ICP for you.

Since defining the ICP is not always a one-time task, you can set this up to be iterative and constantly iterated on with new input data being fed in.

And like with customer support, you could use a combination of prompt engineering, RAG, and function calling to support different workflows.

Prospecting

Once you have defined your ICP, the sales team enters the stage of prospecting.

Depending on the GTM motion you employ, if the outreach is completely outbound, you identify key stakeholders and prospects at your ICP using platforms like LinkedIn, proprietary systems, or your existing network through alumni, investors, and other founders.

Or if the outreach contains inbound leads through webinars, demos, contact forms, freemium, or trials of the product, you can also reach out to prospects that way.

Where AI can help:

Create content to generate inbound lead gen through blogs, videos, social media posts, etc.

Get a list of companies that match the ICP by using a mix of LLMs + Company services API like Crunchbase and others. You can get the size and industry of companies and then prompt engineer the model to rate the company on a scale of 1 to 10 based on the ICP definition.

Using web scrapers, LLMs, and tools like Hunter you can get research done on prospects and to generate outreach messages. Cold outreach should be specific and relevant to the user without being overly personal. It's important to talk about their company and latest posts while avoiding potential pitfalls like including emojis from names or misinterpreting humorous additions to company information. These rules can be implemented through good prompt engineering in language models.

Qualifying

After sending the outreach and tracking the response, organize a discovery call once you get a response.

AI can enhance this process by using live transcription of recorded calls (with disclosure) to transcribe and prompt salespeople with relevant questions, responses, or information about features, integrations, unique selling points, or similar customers. It can cater to objection handling and be tailored to a specific company or industry.

The points from the meeting can also be neatly classified and added to a CRM system.

Qualifying leads with the BANT (Budget, Authority, Need, Timeline) criteria is another process that can be done with an AI assist.

This could be done with just formulas and rules, but AI can contain the subjective part of it to ensure that it's not only numbers and can give a score.

That’s the part about calls. For email communications, AI can help with the following:

From email responses into a CRM to make it searchable and easier to analyze. Responses from calls can be categorized into reasons for accepting or rejecting the product, providing valuable insights for product, engineering, and support teams.

This approach can help normalize the quality of sales calls, making average performers significantly better while slightly improving top performers.

Proposal

Once you reach the proposal stage, it’s important to create the proposal keeping in mind the exact points mentioned in the sales discussions.

AI can help with this by auto-creating drafts of proposals from selected points from sales discussions.

I remember a friend was working at an Ed-tech company where they were creating sales proposals and there was a quality check team to ensure that there were no mistakes.

The AI can go through the proposal and see if there is something missed from the conversations or something from the calls or comments that is incorrect in the final proposal and flag it to the proposal creator.

Closing

One thing I have used AI for is to create simulations of the satisfaction levels of different stakeholders while doing PM work.

This is not too dissimilar from change management efforts while doing sales. You can estimate how convinced one part is vs the other.

And we also know that decisions are not always made by one person, but by a decision-making unit (DMU) that consists of a few important stakeholders. E.g. while buying a SaaS tool for Sales operations, the VP of Sales, CFO, and IT Lead will form a decision-making unit to make the final decision on features, pricing, and integration.

You can add stakeholders, their latest comments, your sentiment, and get the AI to give a score and what should be the next steps.

Other use cases can apply across stages include:

• Insights on the pipeline to meet revenue targets

• Combination of code and insights generator

• Sales forecasting

• Account Management

But we’ll cover them in a separate article.

Final Thoughts

AI is revolutionizing sales by enhancing efficiency and providing valuable insights across the entire funnel. However, it's crucial to remember that AI should augment, not replace, human salespeople. The core of sales remains building relationships and understanding customer needs.

While AI can handle data-intensive tasks and provide real-time assistance, successful implementation requires balancing technological capabilities with the human touch. Organizations must also navigate challenges like data privacy and the ethical use of AI.

As AI evolves, we can expect more sophisticated applications in sales. However, the fundamental principles of good sales – understanding customer needs and providing value – will remain constant. The future belongs to those who can effectively combine AI's power with strong interpersonal skills.

That's it for today, thanks for reading! What do you think of Sales with AI? Reply or comment below.

Give feedback and vote on the next topic here.

Talk to you soon! Follow me on Twitter @kavirkaycee

✨ Reminder for the special offer: If you would like some insight on how to use AI in your company or work, reply to this email, and I’ll help you out. ✨

Let’s get straight to it.

In 2019, I worked at a startup that focused on customer support quality. We used machine learning to analyze text customer support tickets of our customers (that included a $1B fitness startup) and evaluate them on quality criteria such as empathy, friendliness, helpfulness.

Before our solution, this entire process was carried out manually, if at all. Reviewers could review only 1% of tickets, leading to sampling error and bias.

In addition to the evaluation on quality criteria, we also calculated process metrics like First Contact Resolution, Time to First Response, and its impact on end quality metrics like CSAT, NPS, and Customer Effort Score and found a correlation.

The analysis led to better and more frequent feedback to the customer support team as a result improved the customer support quality, and for one of our customers, we analyzed 30K monthly support tickets increasing quality by 15% and resulting in $400K cost savings. That was the impact then, now it’s to a whole new level with GenAI.

The Role of LLMs in Customer Support

Fast forward to 2024. In the age of LLMs that are real-time, cost-efficient, and cheap, I believe most customer support should be handled almost automatically using LLMs.

The characteristics of LLMs that include large context windows, function calling, and retrieval augmented generation (RAG) together can tackle a lot of use cases. Let’s take a look at each.

Large context windows can ensure that conversation history is maintained and the customer doesn’t have to repeat themselves across sessions.

Function calling can look at the customer's ID and check the systems to determine if they have a pending order and reliably retrieve the information from the database.

Retrieval augmented generation can pass through support docs and documentation to provide an answer with citations, ensuring authenticity and limiting hallucinations.

Using a prompt-engineered version of an LLM that caters specifically to the support use case is crucial. The prompt can include things a support agent should do, such as asking clarifying questions, empathizing with the problem, being polite, and providing a resolution.

I believe we have reached a stage where a large percentage of requests can be handled by a bot. For instance, when you ask Swiggy (a food ordering app in India) where your order is, it calls the delivery database, checks the expected delivery time, and provides the result along with an option to call the rider. This can be improved by making it more empathetic and human, checking order status, and offering to ping the rider if needed.

There are many benefits to using LLMs in customer support, like they can run 24/7 without requiring breaks or payment. But let’s look at some of the challenges.

Practical Implementation and Challenges

A sufficient trial period is needed to identify jailbreaks, edge cases, and determine when the system provides reliable solutions. For example, an LLM should be prohibited from writing code or going off-topic and to focus solely on the customer support use case, which can be achieved through prompt engineering - chain of thought reasoning, and few shot examples.

Another implementation is not offering an unbounded chat interface, but preset buttons for the most common support queries. While offering some flexibility, in case their question is not in the list.

Customer Support as a Driver of Product Improvement

From a product manager's perspective, customer support is also a source of product features. It helps identify problems people are facing with the product. Customer support is not just about solving requests; it's about addressing the root causes.

The AI can also take these requests, tag them, and write crisp problem statements, write specs or acceptance criteria for the devs to work on with the PM's blessing, and execute it that way.

Build vs Buy

Depending on the size of the company, they might decide to buy or build. A large company might want create their own versions of this, especially if they’re offering some kind of service with a sizeable customer support volume. DoorDash or Swiggy will decide to build this in house.

But startups might opt for off-the-shelf products since customer support is not core to their offering. Intercom has positioned themselves as AI-first with their AI chatbot offering ‘Fin’ which solves up to 50% of support questions instantly while legacy platforms like Zendesk have also shipped AI features. There is space for a new platform that can take support requests and perform all the mentioned tasks.

Additional Considerations

Additional tasks include collecting objective and subjective feedback from the user in the form of CSAT, NPS, or other metrics and surveys. The text component can then be summarized and tagged and put into a system for it to make sense.

You can then draw a correlation between customer support quality and end business metrics like revenue, profitability, and retention. Especially for a subscription business, where churn is an issue, customer support becomes a profit driver rather than just a cost center.

Personal Reflections and Learnings

In my previous startup experience, as a product lead, I stayed close to the user by handling customer support on Twitter and Discord. Promptness was key and handling dissatisfied users involved being empathetic to their requests, understanding the root cause, and providing them with a resolution. If it was something that we previously encountered, I shared the existing support docs. If it was something new and happened a few times, I created a new support doc so that we could scale support instead of writing the same message each time.

But if it were a feature request, then I tried to understand the motivation behind the request first and then put it into the queue for consideration. Not all requests were implemented of course, but the feedback is always useful.

Final Thoughts

The adoption of Large Language Models in customer support represents a significant leap toward more efficient and empathetic customer interactions. As we have explored, LLMs offer the potential to handle a substantial portion of routine inquiries autonomously, allowing human agents to address more complex and nuanced issues.

While the transition incorporates challenges, particularly in accuracy and reliability, the benefits in cost savings and improved customer satisfaction are compelling. For businesses, staying ahead in customer support technology is no longer just an option but a necessity to thrive in a competitive landscape.

That's it for today, thanks for reading! What do you think of Customer Support with AI? Reply or comment below.

Give feedback and vote on the next topic here.

Talk to you soon! Follow me on Twitter @kavirkaycee

I’ve increasingly started using voice-to-text a lot in my life with the advent of OpenAI’s voice-to-text AI, Whisper, and apps that build on top of it. It’s the fastest way to move from thought to action. And I can imagine why secretaries and personal assistants were a thing in the past. It’s way easier to speak than to write.

I use voice-to-text extensively for writing my daily to-do lists, emails, and messages, brainstorming, doing a brain dump, vent, and much more. But for today’s newsletter, we’re going to cover the use case of writing first drafts with your voice. Let’s dive in!

Writer’s block, no more

The challenge of writing first drafts is staring at a blank page most of the time and not knowing how to start. If you're an inexperienced writer, you may feel that your opening line should be a masterpiece and that the first version will be the final version. So you're aiming for perfection, but your brain and skills are letting you down. Hence, you quit.

Enter voice-to-text as first draft.

For most people, it’s easier to talk on a topic than write. Talking is natural for humans with at least 200,000 years of history, if not more. But we only picked up writing less than 5000 years ago in ancient Mesopotamia and Egypt.

Talking is unstructured. Unless you're giving a speech, it’s unlikely that you will be completely coherent with what you're saying from A to B to C. You will jump in and out of ideas.

Which makes it perfect for a first draft. The purpose of the first draft is to get your ideas down on paper. Imperfection is fine.

Voice-to-text AI

Once, when I had a hand injury (long story) in 2014 when I was in business school, I used the text-to-speech feature of the Mac a lot to write my assignments. It was way easier to talk and write, rather than use just one hand.

However, the quality of transcription was so low that it required a lot of changes, edits, and a lot of start-stops, making the process cumbersome and not as seamless.

The quality has significantly improved with the launch of OpenAI’s Whisper. It brings enhancements to accuracy, speed, and the ability to translate text from different languages. You can also add custom words, such as proper nouns like names, to the model so that those words are transliterated correctly.

There are quality client apps on top of Whisper that do a great job. One example is AudioPen.ai. Other alternatives are Superwhisper and voicenotes.

I spoke to the creator of AudioPen, Louis Pereira, who is a good friend, and wanted to get his views on how he is using AudioPen to write. Here’s what he had to say:

I think of AI-powered voice-to-text tools as the easiest way to take a fuzzy thought and turn it into clear text.

It's akin to putting an Instagram filter on a photo. You still need some skill to take a good photograph. In this case, you need to have a thought worth sharing – but AI helps apply a filter that can sharpen and fine-tune your thoughts, making it easier for others, or your future self, to read.

Some tools like AudioPen (shameless plug) let you choose the intensity of the filter you apply. You can get an almost exact transcript of your voice note, minus the grammatical errors, or choose to get a complete rewrite in a writing style of your choice.

I use it for all sorts of stuff – writing emails, journaling, capturing meeting notes, writing essays, or even creating to-do lists for my future self. And I'm only scratching the surface. I know folks who have used AudioPen to help write their PhD thesis, and some who write newspaper columns with AudioPen's help.

Superhuman’s founder Rahul Vohra had this to say about Superwhisper.

I’m very bullish on voice itself as an input mechanism. I’ve been using this app that’s really great called Super Whisper, and it’s built on Open AI’s Whisper technology. It lets me dictate in near real time into all my Mac apps. The technology has gotten so good that in a year or two we’ll be able to actually drive apps with just our voice.

Here’s another example of Nat Eliason using voice-to-text to write the first draft.

How I use Voice-to-text AI to write a draft

I start by talking about a topic. These recordings often go on for 10-15 minutes. I’m just trying to get all my thoughts about the topic recorded.

I use tools like AudioPen, Claude Opus, or Sonnet to lightly edit the transcript for grammatical mistakes and readability.

Then I use AI to organize and categorize my points into an outline, so I have a clear structure for the article. From there, I start building out the points, add screenshots, etc.

Finally, I again use AI to help me improve my writing, editing, and give me feedback to understand what I missed using Lex. I’ll write about my writing workflow in more detail later, but that’s how it is right now.

Frees you up

Writing becomes easier with Voice-to-text as your partner. You can increase your output and quality at the same time while sharing your authentic thoughts. AI doesn’t have to write for you, you can write with AI. It frees you to be a more curious person and connect different ideas.

In the world of AI, the skills that you need are to have a point of view, be bold to talk about it, have taste, have excellent editing skills, and not procrastinate and publish the article when done.

It also unshackles you from being in front of your laptop to type, provided you're ok with speaking into your phone in public. Typing on my phone is extremely cumbersome, very boomer of me, I know.

But I love “writing” while I am taking a walk in the park or while commuting.

It’s also accessible for people who are blind or have a missing limb.

Sparring partner

Many people have mentioned that they do most of the “writing” in conversation with others. When the updated version of ChatGPT Voice with low latency and high intelligence actually launches, that’s when you can use AI as a sparring partner to discuss, argue, and cover all angles to come up with a stellar first draft.

Encoding more information

When I think of what is additional information in voice rather than text, you can think of voice is characterized by tone and emotions. Imagine when you're using these tools, it’s able to analyze your emotions as you speak about something. We’ve already seen that Hume can understand emotions, and so can ChatGPT 4 omni.

For example, here is the part that you were most passionate about. Or here is the part that you were underconfident about — try to strengthen the argument or the belief.

Final Thoughts

Writing that is not published is not useful. Voice-to-text writing will help you eliminate writer’s block, improve your output, and get more useful things out there in the world.

Your first draft is usually a bunch of points that lack structure, coherence, or a clear idea. So instead of struggling with writing the first draft, why don't you speak your first draft and then write, edit, research, get feedback, and improve?

That's it for today, thanks for reading! What do you think of using Voice-to-text AI to write? Reply or comment below.

Give feedback and vote on the next topic here.

Talk to you soon! Follow me on Twitter @kavirkaycee

As a reminder, I’m opening up a few slots in my calendar to discuss product and growth for only my newsletter subscribers. So if you're interested in speaking with me about your startup’s or company’s product or growth challenge for free, sign up here.

Before we get into it, I wanted to share a new experiment that I’m trying out:

If you have a product, design, or growth challenge that you would like my insights on, reply to this email or comment below with your problem statement and I’ll reply with my thoughts or a Loom video. Make sure to include enough context for me to provide my thoughts on, but it can be concise. I would love to help and give back to my newsletter subscribers 🙌🏽

With that out of the way, let’s dive into today’s piece.

OpenAI Spring Update: Voice AI, GPT 4 ‘Omni’, ChatGPT Mac App and more

We knew before the event that voice was touted to be the big feature to be released. The Information covered it and there were many hints including Sam Altman liking a tweet mentioning that OpenAI was launching ‘her’.

And then after the event tweeting just the word ‘her’.

Voice is a useful medium and I’ve lately found myself using it regularly for use cases including writing first drafts (including this article), listing my next day’s tasks, brainstorming, venting, and more use cases. I’ll write more on these, so subscribe if you haven't.

But with the introduction of GPT-4o, particularly its advanced voice interaction and unified Omni model, represents a step function upgrade to voice interactions and poised to redefine interactive assistants' roles and capabilities in our daily lives and professional environments.

In this piece, we’ll discuss my personal experience with ChatGPT voice, the new omni model, comparison with existing voice assistants, the new ChatGPT Mac app and the new text model.

Let’s start with voice.

Personal Experience with ChatGPT Voice

I’ve used it extensively ever since it launched, from venting out my concerns and getting empathetic and supportive responses to brainstorming ideas, and thinking through solutions — often on walks when I don’t want to type and they usually are 5-10 minute conversations.

Since I speak slowly with pauses, I end up triggering ChatGPT voice to respond before I am done. So I end up holding onto the screen and using it as a walkie-talkie rather than a normal conversation.

There have been a few things that break the experience.

Sometimes the connection breaks and you have to repeat what you say. That’s a big no-no. The latency between you finishing speaking and the response is long enough to break the flow. In yesterday’s event, they deconstructed the way it used to work with the GPT4 model. The AI had to first transcribe your voice to text, feed the model with text, get an output, and then convert the output from text to speech.

I found that the voice interaction was already more human-like than any other interaction I had before.

This changed a bit when Hume, the empathic voice agent, launched. It was real-time, could decipher emotions through a combination of text and voice tone, was able to emote, and handle interruptions.

Yesterday’s update basically shipped all of that, with a way more powerful reasoning model.

The Omni Model

The simplified version of yesterday’s update is that they have managed to create a model that natively supports text, audio, vision — all in one. It takes the input as is, processes it, and returns some intelligence, reasoning, and response based on that.

And it’s super fast.

This unlocks a lot of use cases that I’ll talk about in a bit. But before that, let’s look at the existing ecosystem of voice assistants and how they have all flattered to deceive.

Comparisons to Siri, Alexa, and Other Assistants

Voice assistants are nothing new. Siri, Apple's virtual assistant, launched on October 4, 2011, alongside the iPhone 4S. 2011! That’s a whole 12.5 years ago now.

Amazon Alexa launched on November 6, 2014, with the release of the Amazon Echo smart speaker. Google Assistant was introduced on May 18, 2016, during Google's I/O developer conference and became available on the Google Pixel smartphones later that year.

It works for simple tasks like setting a timer, setting an alarm, telling me a joke, playing a song on Spotify, asking a fact, or interacting with your home devices — like switching on and off lights. I use Alexa for playing music and have set a night-time routine for sleep sounds.

Some do some things better than others. MKBHD released this comparison video in 2023. Worth a watch.

But none of them have leveled up their intelligence to the point of it being useful. All of them are built on previous tech. Here is Siri’s response to me asking it to call me an Uber.

ChatGPT voice would change things. And if rumor has it, if it’s integrated into Siri and onto 1.3B iPhones1, it would be a game changer. We’ll know more at WWDC on June 10th, 2024.

Applications of Voice AI

Some of the applications that OpenAI demoed were interesting. Live language translation, bedtime stories for kids, jokes, homework assistants, interview prep, and what I use it for: assistant to create tasks, write emails and messages, write first drafts, act as a life coach, act as a career coach, and more.

Apart from these, I would be really interested in applications that were demoed by Google 6 years ago!

Google Duplex allowed the AI to make calls for you - to make restaurant, hair cut, etc bookings.

And also allows you to screen calls and filter out scam and spam calls. The AI could talk to these callers and screen them for me, so I only pick up the worthwhile ones.

I would love to not have to speak on the phone for these tasks ever again.

Even though Google demoed this a while ago, we’ve not seen it live yet, which has been disappointing.

Maybe another feature for OpenAI to integrate into the iPhone. Of course, Google will be forced to finally launch this publicly soon at Google IO (May 14th, 2024).

The new ChatGPT Mac App

Now, coming to the laptop, they have announced a new ChatGPT app, which I’ve got access to. It’s an improved UI with a search filter and direct access to voice conversations, with a keyboard shortcut to access ChatGPT similar to Spotlight or Raycast. You can drag and drop images onto the chat and ask questions about them. This makes the previous ChatGPT apps on your Mac obsolete.

Here is how you can get access.

In a future iteration - You can also live share your screen, and the AI will respond, but that’s not yet available. This ability to view what’s on your screen and have an ambient assistant always available in terms of voice is going to be really powerful.

The new text model GPT 4o

Impressions on GPT 4o are that it’s extremely fast, faster than 3.5 turbo, and it seems to be more intelligent. I’ve used it in some of my business APIs, which have provided mixed results because I was used to the previous version. This changes things a bit—in some ways better, some ways worse — but I think that can be fixed with better prompt engineering and giving a few examples.

This update might reduce my usage of Claude now because things are more integrated—it has memory, it’s fast, it’s smarter, and it has code interpreter. It can use my writing style to write things and do a lot of things. It can access the GPT store, and I can create more GPTs for my use cases.

Final Thoughts

Overall, it’s not the big bang update. It’s not GPT-5. It’s not a search engine. But it is incremental and good in a way. It also allows free users to access GPT4, driving further adoption and acceptance of AI. 3.5 wasn’t cutting it anymore and it was now a lousy first impression of AI for people. For example, I can’t wait to get my parents on it. Everyone can benefit from an IQ boost and an intelligent partner.

We still don’t have access to the live voice updates yet, but when that’s out, I can’t wait to test it out and improve my quality of life with the new intelligent voice assistant.

Thanks to ChatGPT 4o and Lex.page for providing feedback on early drafts of this piece.

That's it for today, thanks for reading!

What do you think of Voice AI and the new updates?

Reply or comment below, and I'll reply to you. Give feedback and vote on the next topic here.

As a reminder: If you have a product, design, or growth challenge that you would like my insights on, reply to this email or comment below with your problem statement and I’ll reply with my thoughts or a Loom video. Make sure to include enough context for me to provide my thoughts on, but it can be concise. I would love to help and give back to my newsletter subscribers 🙌🏽

Talk to you soon!

Follow me on Twitter @kavirkaycee

Press the ♥️ button if you liked this edition.

👇🏽

With that out of the way, let’s get to today’s article.

Building and managing an online community can be immensely rewarding. It creates a space for collaboration, feedback, and engagement with your product or service to form a goldmine of insights. But it’s not easy to set up and cultivate. There are a lot of things that can go wrong. Let’s look at the few things you can do right to create a thriving online community.

I've honed my community-building skills through leading communities with The Product Folks no-code community and participating in top online communities like On Deck, Lenny’s community, Foster, and several NFT projects.

I'll share insights on what distinguishes a flourishing community from a lackluster one, and offer advice on how to curate and nurture your own space successfully.

Benefits of communities

As we’ve touched upon, communities can have a lot of far-reaching benefits.

Source of feedback

It can be an unbiased source of feedback from early adopters and power users. These are likely to be champions of your product. But are also likely to have a high bar for quality, and hence tougher to impress. Yet, it’s a great way to get feedback.

Proof of concept

If you're building a product or service or a content creator, a community lets others see what your users/members have created or learned from your product or service.

For example, if you have a graphic design tool, you can let people show off what they have created using your tool.

If you're a newsletter writer, let people showcase their learnings from your posts.

In this way, members can increase their social status, and others in the community see the value of the product.

Midjourney — the AI image creator platform — did this brilliantly with Discord. In the free version, you could not create images privately. All images were public and everyone could see what people were creating with Midjourney. This created a show-and-tell effect where the proof was in the community. Only once you opt for a paid plan, you could create private images.

In the below example, I have shown the Discord of a graphic design tool called Kittl.

Distribution

A thriving community has shared goals and a sense of belonging. If you're shipping a product/content/program, you can look to your community to amplify an announcement on Social. On Deck did this phenomenally.

On major announcement days, a community call was held for all community members and the On Deck team. After the call, there was a request to amplify the official announcement on socials. This was effective in a pandemic and lockdown world, where people had more free time online.

Discussion

This is something that not a lot of startups who run communities get. They think the community should only be about their product/service. A community is not only for announcing features or for handling customer support. If it’s a community, there should be space for discussion related to the industry, while staying on topic.

That’s the only way a community is truly a community.

Look at Obsidian’s Discord, there's a section dedicated to broader discussion of personal knowledge management and other topics.

Revenue

Of course, you can use a community to increase revenue and all those other sweet metrics like engagement and retention. For instance, a member on the verge of churning, they come to the community, and get help from either the community members or the team, and that prevents churn, which increases retention and hence revenue — a good thing. You can also get people to upgrade their free subscription to pay through the help of community engagement.

With the benefits clear, let’s look at what defines the strength of a community.

Characteristics of communities

You can quickly figure out if a community is good by looking at these characteristics.

Bad Communities

A bad community is easy to spot. It's filled with irrelevant content, relentless link-dropping, low-quality interactions, spam, unresolved support issues, and lacks active participation resembling a ghost town.

Good Communities

On the other hand, a good community pulsates with life. It remains on-topic, fostering polite and valuable interactions. It's a place for learning, sharing, and support, a give-first mindset, often extending relationships from the digital world into the physical world.

How to build a good community?

Let's get down to brass tacks and look at tactical ways to create and manage a community.

What’s in it for the community members?

Before creating the community, understand the benefits for members to join, stay, and engage. Maybe it’s to learn, get help, connect with others, and improve their social status, belonging, or achieve a sense of purpose.

Curate, curate, curate

The first step is to be highly selective of who you allow into your community. This sets the tone and the expected behaviors of the community.

Yash Roongta, who runs a community called Alt Investors, said: “I run an active community on alternative investments, my learning is that you strictly need to control the quality of people that you let in because that drives everything.”

Not all companies can do this, but you can set the tone with the first users.

Set values, not just rules

Define a set of values you want the community to live by and the actions and behaviors can be the outcome of those values. It can be anything you choose in your community to live by. For example, On Deck has a spirit of service, confidentiality, gratitude, and family as its core values.

And with each opportunity, the core values were reinforced.

Set rules

Values set the vision, but rules govern the day-to-day. Lay down rules for posting, DMs, harassment, and others. Vishwadeep Tehlan comments: “In my opinion, it's important to set some community rules to not let people digress from relevant topics.”

Lenny’s community has a no-spam link policy in most channels, except one (which we’ll get into in a bit), and the channels are closely moderated to ensure that conversations are happening in the right channels. You can also set and maintain rules against discussing politics or divisive topics.

Onboarding form

For any community, you should have an onboarding form to identify the right people. Start with profile-related questions and social media links, then ask about the value they seek and will add to the community.

Introductions

This is an important part of the mix. You don’t want people to join the community and not introduce themselves. Create and share a common format for new members to introduce themselves.

You can set up reminder bots or onboarding checklists that include this as an important part of the onboarding process.

Welcome

When new introductions come in, they should be welcomed and replied to by the team or community ambassadors to the extent possible. Imagine going to a house party and no one welcoming you. You’d feel unwanted and unlikely to stay.

You can record a Loom video from the founders and send it to each new member through automated DM (and add to the welcome section) — adds a personal touch, but at scale.

Seed discussions

It's important to seed discussion initially. Don’t expect people to come in and start engaging directly. Instruct all employees and ambassadors to start discussions and keep people engaged.

Identify team members who would take turns engaging and ambassadors and influencers to engage.

Fewer channels than needed

At the early stage, it’s tempting to keep creating new channels for each new topic. But in a community's early stage, you want to funnel most engagement where most people can see.

The reality is that many of the topic channels will get less engagement and die out. At that point, you should periodically remove dead channels.

Of course, if the community is super popular and engaged, then this shouldn't be an issue. But at the early stage, it's better to have fewer channels.

Dedicated promote yourself channel

To reduce link dropping in each channel, keep only one channel where people are allowed to share their links to avoid making each channel a link dump. For example: Lennys’ paid Slack community and Level’s Telegram community.

Community moderators - ensuring things are on track

Use AI to moderate most interactions to reduce spam. There are LLMs that allow you to do this now. Otherwise, create tools and SOPs for moderation by firstly the team and then chosen community members. Always allow people to contest moderation to correct any false positives.

Engage

Community managers should have the skill and knowledge to connect people with others in the community.

For e.g. hey you write on mental health, and there is another writer who also writes on mental health, why don’t you two chat? Here is their Discord/Twitter/LinkedIn.

Have weekly prompts to encourage discussion and improve engagement.

Reward the community members Organize community-only events/giveaways/perks.

Organize weekly or monthly community calls on Discord would be another fun way to engage the community.

Rituals

NFT communities did this well, almost as a cult — where they would have their rituals — like gm, gn, and other things specific to communities. Creating that shared memes/lore/lingo will go a long way.

AMAs

Call on external experts to do text or audio/video AMAs for the community. It can be within the platform or a Zoom/Meets call with only community members getting access.

I’ve taken part as an expert in one of the AMAs hosted by The Product Folks.

Final Thoughts

In conclusion, the backbone of a successful online community lies in the delicate balance of vibrant interaction, shared values, and strategic management. A community thrives when its members find consistent value in engaging with one another, and when they truly feel part of something greater.

By embracing the tactical advice shared here—from careful curation to active engagement and rule-setting—you can cultivate a space that resonates with its users and stands the test of time. Remember, the most vibrant communities are those where every member's voice adds to a harmonious chorus, echoing the collective purpose and passion.

That's it for today, thanks for reading! What do you think of creating and managing online communities? Reply or comment below.

Give feedback and vote on the next topic here.

Talk to you soon! Follow me on Twitter @kavirkaycee

As a reminder, I’m opening up a few slots in my calendar to discuss product and growth for only my newsletter subscribers. So if you're interested in speaking with me about your startup’s or company’s product or growth challenge for free, sign up here.

I’m opening up a few slots in my calendar to discuss product and growth for only my newsletter subscribers. So if you're interested in speaking with me about your startup’s or company’s product or growth challenge for free, sign up here.

With that out of the way, let’s get to today’s article.

Subscribe now

After attending several meetups in Bangalore, Toronto, and SF, I wanted to try my hand at hosting a meetup and was keen to get to know a few tech/product people from my home city of Bombay. So I decided to host a community event of ’s product community.

And in true PM style, I wanted to ensure the experience was good. So I thought through what would make for a good experience, executed some of it during the meetup, and reflected on how it could be improved to create this guide on how to host IRL events.

Before the meetup

Theme

Before you get started, it’s helpful to describe a theme for the meetup.

Most of the meetups I’ve attended were agenda-less and focused on getting to know each other, almost like speed networking. I think this format works well in most cases. Organic conversations can result in good connections.

While some others center on a specific theme or sector, like Fintech or AI. And some come with specific questions like “one challenge that you're facing at work”. This could potentially lead to tangible outcomes from the meetup but requires some time investment from all participants beforehand. The decision lies with you on what you want from the meetup.

Application Form

Have an application form for event details (time, venue, description) using Luma. You can Luma to send updates via posts to all registrations and use automated emails for key updates and reminders.

I recommend keeping an approval process by asking questions like "How did you hear about this meetup?" or "What do you hope to gain from this meetup?" You can use this question and look at their Twitter/LinkedIn profiles to apply a filter and keep a high quality/relevance bar. For example, in one of the meetups I hosted, there were folks from a completely unrelated field. Those people were not able to contribute equally to the conversation.

So you could, for example, exclude non-PMs in a product manager meetup.

There is typically a 50% turnout between those who fill out the form and people who attend. There could be a higher turnout for a form with an approval process, as it signifies higher value and exclusivity.

Venue

Pick a central and accessible venue— a place that doesn’t play really loud music. I prefer a standing meetup to a sitting one because it allows more movement and interaction. A high table is better than a sit-down table. Have a few seats available for those tired of standing.

Space dynamics can shape the number and quality of interactions. When I hosted the event, the narrow walking area made it difficult to pass by and move around. More space would help.

You can decide outdoors vs indoors depending on the weather. I did an outdoor event in Jan in Mumbai, which was tolerable. Any other month it wouldn’t have been.

At the meetup

Welcome

At the restaurant/bar, inform the front desk about the name of the meetup and name of the host. They can then direct attendees to the right space. This will ensure that attendees can reach the host without needing to contact the host. I’ve heard from others hosting meetups that they made the mistake of not informing the front desk that it was for a specific community meetup and instead just gave their name. This caused confusion and didn’t lead to a good first user experience.

At the entrance of the meetup space, put up a sign board/placard to indicate it's the XYZ meetup. For example, I put up a placard for Lenny’s newsletter meetup with color matching name tags.

For the first 5 attendees, the host should welcome them until others can also join in on the welcoming. You can guide them to put on nametags and get a drink. It’s important for people to feel welcomed in the beginning of the meetup for them to become comfortable.

Name tags

Don’t do a meetup without name tags. I’ll tell you why.

Print name tags and keep them near the entrance visibly with at least 2 markers. And another detail that you would normally overlook — the nametags should easily peel off, or it’s a hassle. (This happened to me!)

Attendees should write their role and company along with their name. It made a huge difference in my experience. I ended up recalling and associating people's names with their role and company, making it easy to initiate and reinitiate conversations.

Beverages/Food

Ask people to order their beverages at the beginning before any major conversations. For that, the Menu/QR code should be easily accessible and instructions given to call the waiter or go to the bar and order a drink.

The host should take care of ordering snacks for the table within the budget. And you should remind the attendees to snack once the food arrives.

Conversation

I’ve realized that “What you're working on” is good enough to start a conversation and then people can take it forward.

What stood out for me from attending meetups in the US is that people were extremely curious. They asked a lot of questions and came to learn rather than just talk. That was something new to me. The focus is usually on putting yourself out there. But by asking good, curious questions you learn and it makes for a good conversation when reciprocated. Try to prevent one or two people from dominating the conversation, especially in a group of 5+ people. Go to some lengths to encourage conversations from different people.

As a good host, it's important to connect people. For example, if two people are starting up and looking for a co-founder, you can connect them to chat and see what happens. This is 101 of community building and can work virtually in Discord/Slack communities too.

Future Connection

You don’t want a one-time meetup. You want to build relationships for the future. A lines vs dots approach.

The easiest way to connect is on LinkedIn or WhatsApp with the QR code instead of jotting down names and numbers. You can lead this initiative and others will follow. It's convenient for everyone.

Meetups can be fruitful when you build this long term relationship. I ended up meeting a bunch of VC associates in Bangalore once that has led to a long lasting relationship and a bunch of collaboration, introductions, and even more value created down the road.

Final Thoughts

In closing, remember that hosting a meetup goes beyond the logistics of place and time; it's about creating an ecosystem of engagement where each attendee's experience is valued and every interaction is meaningful. By setting the stage with thoughtful preparation and genuine curiosity, you enable a space where connections are not just made but nurtured. In doing so, you don’t merely host an event; you plant the seeds for a thriving community that can grow well beyond the confines of a single gathering.

Thanks to for co-hosting the event with me, for building such a great community, Orb for sponsoring the meetup, and Lex for editing this article with me.

That's it for today, thanks for reading! What do you think of hosting meetups? Reply or comment below.

Give feedback and vote on the next topic here.

Talk to you soon! Follow me on Twitter @kavirkaycee

As a reminder, I’m opening up a few slots in my calendar to discuss product and growth for only my newsletter subscribers. So if you're interested in speaking with me about your startup’s or company’s product or growth challenge for free, sign up here.

I have a small ask before we begin: In case you or someone you know would like to get me on board for product management roles, product consulting, or startup advisory roles, check out my work and book an exploratory call.

With that self-promotion out of the way, let’s get into the article:

The much-awaited GPT Store has just been launched! It was announced last week via a discreet email to only GPT creators. It’s been delayed by almost a month, thanks to the ousting of Sam Altman as CEO and the drama that ensued, in which the entire existence of OpenAI was put into doubt, let alone the GPT Store. But here we are.

The GPT Store was originally announced on OpenAI’s DevDay on November 6th. In many ways, it felt like a Steve Jobs presentation with the audience (mostly devs) being thrilled with all the updates being launched.

We’re here to discuss the GPT Store and what parallels we can draw with another store that was launched in July 2008 — the Apple App Store.

Before we dive into that, let’s talk about what are GPTs.

What are GPTs?

GPTs are standalone versions of ChatGPT that have been highly customized by the creator through instructions, documents, and actions. And you can allow the user to access browsing, code interpreter, and DALL-E. They’re only accessible to ChatGPT Plus users for now with a dedicated URL. Below is an example of a custom Productivity Assistant GPT that I have created.

I kinda predicted it back in March, or at least requested a feature like this.

Instructions are essentially a set of prompts that set expertise, context, goals, and actions for the chat conversation. Think of prompt engineering. Telling the AI what their role is, what their expertise is, and how to guide the user through the conversation.

I’ll take an example of a personal productivity GPT that I open-sourced.

You are an expert productivity assistant. You have knowledge of concepts like timeblocking, pomodoro and its applications. You help out with: - Make the following tasks more actionable by breaking them down into individual actions. Write the subtasks in numbered bullet points. - Prioritizing tasks according to impact and effort - Unblock the user by reducing procrastination caused by anxiety. Use scientifically proven techniques to reduce anxiety about a task, and break it down into actionable steps You can add/update tasks in Todoist via an action. You can use natural language and add due dates to tasks via an action. Always add main tasks as tasks, and then the bullet points as subtasks. Don’t use sections in Todoist. You use motivational language to prompt the user to achieve their goals.

The documents or RAG (Retrieval augmented generation) are a set of documents that you can add to the model that provide the AI with extra factual information that the AI can then use in the conversation.

For example, you can add readme txt files to the document list that has basic instructions on the GPT. Or release_notes txt file that tells the user what the latest updates in the GPT are. You can also export your blog archive into a PDF and have the GPT search through that knowledge base instead of its pre-trained knowledge base. The goal is to supplement the pre-trained knowledge of GPT4 with extra information that can help the user. The possibilities are endless.

I didn’t use any documents in my example for the personal productivity GPT, but you can easily add documents that supplement the pre-trained knowledge.

Actions are API calls that the GPT can make to external services. If you want to get real-time weather information from a weather service API, you can. The only downside is that you can’t have the user add their API, it’s going to be a single universal one in the GPT. This works better for retrieving common information, rather than reading and writing specific customer-related information.

In this personal productivity GPT example, I connected the GPT to Todoist, a popular to-do app with an API. Again, the problem with making this public was that it was using my API key and everyone’s tasks would get added to my list.

Examples

1. UX-Copywriting

2. Product Management Bot

3. No-code News

4. Tweet Optimizer

5. Scholoor

6. and more

Various roles that can benefit from GPTs (Product Manager, Writer, UX Copywriter, Marketing, Relationship Coach, Therapist, etc.)

My top picks:

• Grimoire (Coding Assistant)

• DesignerGPT (Design and Frontend Assistant)

• Consensus (Research Assistant)

• Trey Ratcliff's Fun Photo Critique GPT (Photography Assistant)

Since the launch of GPTs and the announcement of an upcoming store, there have been several collections of GPTs to solve the discoverability problem.

• Awesome GPTs GitHub repo

• All GPTs

• GPT to search other GPTs

Of course, having a native store will greatly solve discoverability. And that’s where the GPT Store fits in.

What is the GPT Store?

OpenAI is launching a store in which people would be able to discover GPTs created by individual users and teams.

This will provide users access to

• Leaderboards and distribution

• Search

The last stat of monthly active users of ChatGPT was 180M in Nov 2023 with 100M weekly active users. However, it is not sure what the split between free users and paid users is, with some estimates saying it’s 250K paid users.

If you had the chance to put your creation in front of 180M monthly active users, you would in a heartbeat. But what we’re realistically targeting at launch is the paid users which is 250K. Since GPTs access the paid GPT4 model with advanced capabilities like web browsing, DALL-E, and code interpreter. This gives a smaller user base to target. So for the GPT Store to be truly successful, they would have to convert a lot of free users to paid. Or introduce 4 to free users, but with lower caps in the form of previews. And allow 4.5 Turbo for the paid with higher caps.

So you're building for the pro user who has a ChatGPT Plus subscription. Here are the different roles that are likely to upgrade, according to ChatGPT (how meta).

At the outset, I compared this launch to the App Store launch in 2009. Let’s see what the similarities and differences between the two are.

Apple App Store comparisons

In 2007, Apple had just launched the iPhone to resounding success. In the first year, they sold 6.12M iPhones. And launched the App Store in 2008 to allow third-party apps on the iPhone.

2008 was a completely different time. Distribution was with bigger companies like Apple. Individuals or smaller teams were dependent on Apple for distribution. Social media was just nascent. The only other way you had online distribution is if you had a large mailing list.

Apple took advantage of this disparity and with its captive audience and no competition, launched the App Store. The sections of New and Noteworthy, Featured, and Trending — were the best ways to discover apps.

However, it has since waned. I don’t remember the last time I went to this page on the App Store to discover apps anymore. It’s either word of mouth or Twitter.

Anyway back to the App Store launch. When the Apple App Store first launched, it didn't take long for independent developers to start releasing unique and innovative apps. One of the first well-known indie apps was a novelty app called "iBeer" by Hottrix, which turned the iPhone screen into a virtual pint of beer.

The app utilized the iPhone's accelerometer and touchscreen to give the illusion of drinking a beer when the user tilted the phone. It quickly became a novelty hit, showcasing the creativity of independent developers and the potential of the new platform. The app’s creator revealed that he was earning $20K a day at the peak of popularity.

Apart from novel apps, there were useful apps for social media and music like Pandora, etc. The real wave of useful apps came in 2009. Uber and later social mobile-only apps like Instagram/Snapchat.

The difference with GPTs is that they are both novel and useful at the same time. It is not just a party trick. You can get serious value with the output of GPTs.

But as I mentioned before, the 250K users of ChatGPT Plus don’t compare with the 6M+ iPhone users at the time of the App Store launch in 2008. There’s a lot that needs to be done for GPTs to compete because of the paid plan restriction.

Future of GPTs

What I think the future of GPTs would be is:

• Standalone app with their history and search and not part of the same ChatGPT interface

• Relying on the same no-code ways of building GPTs

• Get the same distribution through the GPT store (this time open to all 180M ChatGPT users)

• The creator of the app gets to decide whether they are free or paid and pays the API costs

• OpenAI can take a share of revenue as a discovery and creation fee

Final Thoughts

While we can’t compare the two launches and can’t predict the success of GPTs on the OpenAI GPT Store.

What the current wave of GenAI and in particular ChatGPT and GPTs have done is make creation fun again, and take us back to a time of the early days of mobile, when there was a lot of fervor for innovation, some stupid apps, but some apps that genuinely brought utility, entertainment, and joy to people’s lives.

That's it for today, thanks for reading! What do you think of the new GPT Store? Reply or comment below.

Give feedback and vote on the next topic here.

Talk to you soon! Follow me on Twitter @kavirkaycee

In case you or someone you know would like to get me on board for product management roles, product consulting, or startup advisory roles, check out my work and book an exploratory call.

Refer a friend

Hey there folks! It’s been a while since I’ve last published a post here. I’ve been heads down working my full time role as a product manager at Hashnode and shipped a lot of product features.

I have had multiple start stops to this serendipity engine that is my newsletter, but now it’s time to restart it again.

Welcome to a bunch of new subscribers who have joined us after downloading one of my Notion templates. For those who are new here, I’m Kavir, I write on product management, startups, and AI. Check out some of my best work here.

With that out of the way, let’s dive into this week’s newsletter on ChatGPT’s custom instructions and how you can increase GPT4’s IQ points.

Today’s edition is brought to you by Bubble.

Bubble is a No-code app development framework that lets you design, develop, host and scale applications without writing any code.
Bubble was named one of Fast Company’s Most Innovative Small and Mighty Companies of 2021.

ChatGPT has really changed the game for knowledge work. I last wrote about it in Jan 2023 (ChatGPT for PMs) and since then the space has grown in leaps and bounds including GPT4, Custom Instructions, GPT-4V, DALL-E3 and voice conversations.

In today’s newsletter I will talk about custom instructions in ChatGPT. Because I think they’re an underrated facet of ChatGPT and not enough people know about it or how to use it. You can improve GPT4’s IQ points with just a few lines of text. Let’s dive in:

Custom Instructions

Custom Instructions is a relatively new addition to ChatGPT launched first for beta users and now for all users.

It allows the user to configure information related to the user and how they would like to interact with the AI. Here are two questions it asks:

What would you like ChatGPT to know about you to provide better responses?

and

How would you like ChatGPT to respond?

Both are equally interesting and can have profound effects. Let’s start with the first one:

What would you like ChatGPT to know about you to provide better responses?

One common complaint of AI has been that the responses can be extremely generic. LLMs are generalized AIs with no specialization. Unless you fine-tune the model or prompt engineer it a fair bit, it will spew out generic responses that may or may not be relevant or add value.

This setup aims to reduce that.

You can give the AI a lot of context to who you are:

what do you do, what do you write about, what are your inspirations, what are your interests, what are your goals, what are your motivations, what are your challenges.

For example, I have written that:

My name is Kavir Kaycee, I'm a product manager and writer. I write The Discourse (https://thediscourse.co)—a newsletter about startups, tech, product. And nocodeshots.com, a newsletter on no-code tools and news.

I’m looking to be the best product manager I can be. 

My personal goals are to be consistent with my fitness and mental health goals. 

I love books, writing, technology, weight lifting, soccer, traveling, standup comedy. 

My writing heroes are: Paul Graham, PG Wodehouse, Douglas Adams. 

I want to get better at time management and effectively plan my day in advance and keep some space for thinking and planning the future. I need reassurance but need to focus on growth.

This covers what am I am working on, what do I write about, what are my goals, my interests, my inspirations, and what are my challenges.

Now this comes in handy with chat, especially when you ask general questions. The AI already knows about you, so it can make the response more specific.

For example here, it added the point about my interest in startups, tech, and no-code tools.

And when you need a definitive recommendation. It’ll take your goals and constraints in mind and give you an answer one way or the other. 

But where this truly shines is during voice conversations. It can really act as your assistant, therapist, counselor, advisor and it seems so natural. More on this in a later piece, subscribe to be the first to receive that piece.

The second one is:

How would you like ChatGPT to respond

There are two key problems that ChatGPT and LLMs face:

One is that the LLM jumps to conclusions. The second is that the AI hallucinates a lot.

Here is an example of the AI getting a simple arithmetic answer wrong and then later correcting itself.

Let’s fix both of these with custom instructions.

Here’s the custom instruction prompt that I am using that is derived from Jeremy Howard’s tweet.

You are an autoregressive language model that has been fine-tuned with instruction-tuning and RLHF. You carefully provide accurate, factual, thoughtful, nuanced answers, and are brilliant at reasoning. If you think there might not be a correct answer, you say so.

Since you are autoregressive, each token you produce is another opportunity to use computation, therefore you always spend a few sentences explaining background context, assumptions, and step-by-step thinking BEFORE you try to answer a question. Take a deep breath and think step by step. 

Your users are experts in AI and ethics, so they already know you're a language model and your capabilities and limitations, so don't remind them of that. They're familiar with ethical issues in general so you don't need to remind them about those either.

Don't be verbose in your answers, but do provide details and examples where it might help the explanation.

It’s a lot of fancy words, so let me break it down for you in simple words.

Autoregressive in simple terms is similar to the autocorrect in our phones, where depending on what was written earlier, it tries to predict the next word in the sequence. Of course to a much more sophisticated level.

Fine-tuned with instruction tuning just means that the models are not just pre-trained GPT3 ones, but that are further fine tuned with specific datasets and combined with RLHF (Reinforcement Learning from Human Feedback) where the AI becomes smarter with human feedback.

Prompting the AI to give factual and accurate answers and giving an ‘out’ if they don’t know the answer. This cuts down hallucination or creatively making the answer to a great deal.

The next para prevents the LLM from jumping to an answer and attempts a chain of thought answer each time without prompting each time. As evidenced by this paper. This makes the output longer but it’s worth it to get things right and not hallucinate.

The next two are just to prevent some auto responses from the AI and not to have the AI be too verbose.  

After adding these instructions, here is the answer to the same question that it got wrong earlier.

All this gives instant IQ points increase for GPT4.

Final Thoughts

The people who find ChatGPT responses to be generic or not useful haven’t taken the time to set ChatGPT up for success.

Through GPT4's custom instructions, we saw how AI responses can fit specific user contexts rather than being broadly applicable. We learned that providing AI with details about ourselves enhances its ability to personalize responses. We also discussed how instructing AIs on response methods helps curb hasty conclusions and unfounded answers.

And this is just the start, the real value of part of this setup will come with the voice conversations — which truly takes ChatGPT into a sci-fi world of robot assistants.

Further Reading:

1. Every’s post on Using ChatGPT Custom Instructions for Fun and Profit by

2. ’s tweet on custom instructions

That's it for today, thanks for reading! What do you think of custom instructions in ChatGPT? Reply or comment below, and I'll reply to you.

Give feedback and vote on the next topic here.

Talk to you soon! Follow me on Twitter @kavirkaycee

In case you or someone you know would like to hire me for product consulting or startup advisory roles, check out my work and get in touch. 

Refer a friend

In the first edition of this series, we’ll cover one of the most popular ways in which no-code platforms are used — to build landing pages.

Landing page builders are a gateway drug to more robust no-code use cases. While you would previously need a web developer who knew HTML, CSS to create one, you can now create and maintain a fully functioning, dynamic website yourself. 

This is true if you're building at a startup or for your side project. 

When choosing a landing page builder, you’ll have requirements of cost, complexity, and jobs-to-be-done. I’ll be comparing different no-code platforms with this lens. Let’s go!

Want experts to build your app?

Got a website or app idea? 💡— Let's make it happen! 🚀

The awesome No-Code Shots team is partnering with the best agencies and freelancers to turn your vision into reality. 🌟

We’re only taking on 10 new projects this month. 🔟

🔗 Secure your spot now!

Carrd

Kicking us off is Carrd. It’s a one-page site builder, created by indie maker @ajkln.

The drag and drop interface makes it extremely simple to get started. It’s best used for simple landing pages - coming soon, one-page portfolio, or newsletter page.

To build any of the above, you will need features like forms, email capture, galleries, and embeds. Carrd has all of this. You can capture emails through form integration with Mailchimp or Mailerlite and integrate payments with Stripe or embed a Gumroad product.

Since it’s a one-page builder, it’s not ideal for websites planned to have a lot of pages and content. You can still extend the functionality with sections, but that’s mostly a workaround.

While you edit the website, the mobile website is automatically generated and doesn’t need much customization.

What stands out for Carrd is its pricing plan. You can pay $9 a year for three websites or $15 a year for 10 sites. That’s ridiculous value. 

Speed optimization: Average

Ease of use: Very easy

Cost: Very low (starting $9 for a year)

Example: showw.work

Notion

Notion is essentially a knowledge management app. However, due to its versatility,  we’ve seen people using it in novel ways: job posts, landing pages, blogs, and so on.

There are two ways you can use Notion to create landing pages: 

One is to use Notion’s inbuilt share to web option where it creates a link on a custom subdomain. For example, superteam.notion.site. This is perfect for sharing things like job postings or something similar. 

Another way is to use a hosting provider like Super or Potion, which takes your Notion workspace and generates static pages from it. The key benefits of this approach are speed, better SEO, more customizability, and a custom domain. I’ve done that for my personal website kavirkaycee.com 

If you're anything like me, Notion has become an integral part of my workflow and I use it for almost everything including writing the first draft of this post! That means that it’s constantly available. This makes it very easy to add or update text and publish it to the world! 

Compare that with something like WordPress or other options on this list, where you have to login to a site that you don’t use frequently, and then change something. This adds friction, which might mean you update your website less frequently. 

Speed optimization: Low if Notion hosting, high if Potion/Super hosting

Ease of use: Very easy

Cost: Low-Medium (Free for Notion hosting, $8-12 per month for Potion or Super)

Example: kavirkaycee.com

Softr

Softr is a unique addition to this list. It doubles up as a landing page builder and a web app that can create dynamic curated sites, internal membership, using Airtable as a backend.

It does the common things well enough, like section blocks, forms, and mobile layouts. 

I would recommend using Softr if you have dynamic data and filters for the data. Otherwise, as a pure landing page builder, there are better and more aesthetic options available.

We’ll get back to Softr in the editions on curated and membership sites, stay tuned for that! 

Speed optimization: Moderate

Ease of use: Moderate

Cost: Free plan, Moderate

Example: Ycprep.app

Webflow

Webflow is the most powerful and robust out of all the ones in the list. It’s a pro tool where designers become visual developers. It has a learning curve, but once you learn it, it’s super easy to build on it.

If you're looking to build a fluid and fast site with slick animations, Webflow is your best bet. 

I chose Webflow for a site that required the best SEO performance and had 20+ pages that were similar to each other. Symbols or reusable blocks powers building sites with the same blocks. Updating the site becomes a breeze, instead of updating manually on all pages, you just need to update the main symbol.

It also has a content management system called Collections that you can use for not only blogs but other collections as well, for e.g. job listings. There are ways in which you can import data into these collections using a combination of Airtable, Zapier, and Powerimporter. 

On mobile responsiveness, you will need to work on it manually, which takes time — something I found out the hard way.

The good part is that you don’t have to start from scratch. The template ecosystem is strong and you’ll be able to find something you can base your site off.

Go for Webflow if the above criteria matches yours or if the website is very critical to your business and you want a no-nonsense option. 

Speed optimization: High

Ease of use: Difficult

Cost: High ($15 a month)

Example: makemyresume.co.in

Typedream

I only recently used Typedream and it’s slowly becoming my favorite option. 

Typedream offers the simplicity of Notion’s building blocks in the form of a website builder.

You can use it for marketing landing pages, personal websites, or guides.

The slash commands are similar to Notion and Roam. So if you're familiar with those, creating a website becomes really easy. For example, /button to add a button. It makes it really easy to add elements, which makes a lot of sense for someone who is building a lot of websites or needs to update the websites frequently.

Editing is inline vs menus on other platforms, meaning that editing and preview almost merges together. This feels more fluid and intuitive.It also has auto mobile responsiveness. 

The collections in Typedream are mostly static. But you can embed content into a collection, like in the example of NFTs.

Typedream offers a good balance between customizability and simplicity, and slowly becoming a platform that I can recommend more and more.

Speed optimization: Moderate

Ease of use: Easy

Cost: Expensive ($15/month)

Example: hera.so

Unicorn

The next on the list is Unicorn Platform. One thing that it excels at is its collection of section templates for a landing page. For example, hero section, pricing, etc. It offers a lot of options to choose from. 

Besides that, it has easy-to-use design options. Like theming with suggested color combinations to keep design sanity. But you can select your own custom colors.

The common features include form integrations with Mailchimp, Mailerlite, Airtable, Zapier, Slack, etc.

One drawback is that if you select one section template - each element of that template has to be there. For example, a section template with a mockup of the app had to contain the mockup.

There is also an option for HTML export in all paid plans. This feature exists only in the higher plans on other platforms. 

Speed optimization: High

Ease of use: Easy

Cost: Medium ($8)

Example: fiticapp.com

Framer

Rounding off the list is Framer. Framer is now the tool that I recommend the most when it comes to building landing pages at an affordable price.

It also has a CMS, animations, and an AI to-website builder feature.

The design is easier than Webflow with a bit less customizability.

Speed optimization: High

Ease of use: Easy

Cost: Low ($5/m) when paid annually, otherwise $10/m. But it also has PPP pricing for developing countries.

Example: resumecraftersai.com

Honorable mentions:

• Dorik

• Squarespace

Final Thoughts

There you go! Recommendations often have a lot of context involved. I hope I’ve been able to add enough context to the recommendations to help you choose which platform to use for your specific use case. 

In any case I missed any tool you love, please reply or leave a comment!

The space is quickly evolving and I’ll keep updating this list to keep up with the advancements. Share this with someone who you think would help.

📘 Read of the week: How Memes Control Everything - Nathan Baschez (7 min)

Want experts to build your app?

Got a website or app idea? 💡— Let's make it happen! 🚀

The awesome No-Code Shots team is partnering with the best agencies and freelancers to turn your vision into reality. 🌟

We’re only taking on 10 new projects this month. 🔟

🔗 Secure your spot now!

That's it for today, thanks for reading!

Share this newsletter with someone who might enjoy it!

Give feedback and vote on the next topic here.

Talk to you soon!

— Kavir

Press the ♥️ button if you liked this edition.

👇🏽

Just kicked off the third cohort for The Product Folks no-code program. Check out the kick-off event tweet here. It’s been a blast so far and I’ll continue sharing updates on the products built during the cohort.

If you’re new here, please subscribe and get insights about product, no-code, and startups delivered to your inbox every other week.

Subscribe now

ChatGPT has taken the world by storm. You’ve probably come across screenshots of what is possible with ChatGPT on Twitter or LinkedIn.

It’s no longer a toy, it’s being used for actual work. For example, I’m actively using GPT3 in a project that requires my team to create leadership profiles for top employees in a company. This project will fetch me 5-figures $ revenue. 

With any kind of new technology, there is often pessimism of how we would lose our jobs. Sure, advancements in technology means that some jobs are made redundant. Before the invention of alarm clocks, there were literally people whose job it was to go door to door and wake people up. They were called “knocker-uppers”. But these same advancements in technology creates new kinds of jobs. Try explaining any of the modern jobs like software developer to people a few decades ago. Now it is estimated that over 24.7 million people will be employed in this job by 2024.

CGP Grey, who makes YouTube explainer videos, published this fantastic video called Humans Need Not Apply eight years ago. Even though it was before all the latest advancements of GPT-3, DALL-E2 and others, he posits that even though there have been advancements in tech throughout human history we’ve always created new jobs. But this time, with AI, things are different. 

My take on it is not as gloomy. Now, you can be in the camp of ‘My job is going to be automated.’ Or you could be in the camp of - how can I make the most use of this new tool to provide more value in my work. 

My thesis is that with every advancement of technology, you need to embrace it and use it as a tool for your own advancement. 

What is ChatGPT and what can it be used for 

For those who don’t know, ChatGPT is an LLM (Large Language Model) by OpenAI. In simple words, it’s an AI tool in the form of an interactive Chatbot that can give you answers on various topics, spark ideas, create content, and solve problems. 

It’s a tool that can make you more creative and efficient. Let me explain how:

ChatGPT as a Product Manager's Assistant

defines the PM role as:

“Your job as a PM is to deliver business impact by marshaling the resources of your team to identify and solve the most impactful customer problems.”

If you boil this down to first principles, the Product Manager role at its core is about thinking and communicating influentially. 

Thinking involves drawing on the breadth of your experience, what you’ve read, and the context of the current problem. Some problems can easily be solved with best practices. Other problems require unique creativity to find a solution.

As far as team communication is concerned, most of it nowadays is written. It happens in written docs, in Slack messages, and in emails. Communication needs to be influential, to achieve outcomes from people who don’t report to you directly. 

The way ChatGPT can help you in thinking and communicating is that you can expand rough notes or questions into something tangible. This creates a headstart to your work rather than trying to start from scratch. Then you, as the editor of the work, can refine the output based on your judgment.

Which brings us to the mental model of ChatGPT being your assistant rather than being your replacement.

When you delegate work to an assistant or junior, you have to give training, then instructions, review the work, and then give feedback. Similarly, with ChatGPT, you give instructions, get the output, and then can refine the work. And the key differentiator of ChatGPT over previous forms of AI is that you can have a conversation with it. You can offer corrections or more context and improve the output through that exchange. Similar to how you would interact with a human. 

So we’ve seen how the interactive nature of ChatGPT simulates how you would work with an assistant. Let’s go back to the point about creativity and efficiency. 

Creativity

In terms of creativity, and given the right context, ChatGPT can help with ideation. For e.g create startup ideas.

Some of these ideas are generic, but some may just spark unique ideas such as Tweet Scheduling, a feature that is missing from Twitter’s core experience, and done much better by the likes of Typefully. 

Efficiency 

The efficiency gains are tremendous. You take rough meeting notes and the AI constructs detailed meeting notes from them, saving you time. You can write a prompt for an email to a customer, and ChatGPT writes a better email than most could.

The benefit is that you can spend more time on decisions rather than on creating documents, writing emails, and release notes from scratch.

If you’re a leader of a team, you know that you are leveraged and that what matters is your decision making ability. You can delegate all the grunt work of writing first drafts of various PM deliverables to the AI. While you still have to review what’s created, give feedback, and refine it. The overall productivity gains are immense.

Use Cases for ChatGPT

Let’s look at individual use cases of where you can use ChatGPT in your work.

- Simple and detailed product specs

- User interview questions

- Roadmap creation

- Customer journey mapping

- Job story development

- Onboarding experience design

- API integration questions

- Competitor analysis

- Meeting preparation (agenda, notes)

- Increasing engagement and making an app faster

- Slack reminder message generation

- Meeting notes 

Let’s take the example of creating product specs for a feature or product. I have written extensively about my framework for writing product specs. Here is the same format, being used by ChatGPT to write a fictional product spec for an esports registration page.

Prompt:

💡 Write a product specs for an esports registration page for a sports media company based in India targeting gamers in the age group of 18-25 in this format:

- Summary

- User Need/Problem

- Business Metrics

- Primary Research (User Interviews, User Testing)

- Secondary Research (Competitor Analysis, Benchmarks, Surveys)

- Features in the form of Jobs-to-be-done

- Post Mortem

Answer:

🤖 Summary:

Our esports registration page is designed for a sports media company based in India targeting gamers in the age group of 18-25. The page will allow gamers to sign up for various esports tournaments and competitions hosted by our company.

User Need/Problem:

Many young gamers in India face difficulty finding legitimate and organized esports tournaments and competitions to participate in. Our registration page aims to address this problem by providing a one-stop platform for gamers to discover and sign up for various tournaments.

Business Metrics:

Our main metric for measuring the success of the esports registration page will be the number of users signing up for tournaments through the page. We will also track the number of tournaments and competitions hosted on the page, as well as the number of unique visitors to the page.

Primary Research:

We conducted user interviews with young gamers in India to gather insights on their needs and preferences when it comes to discovering and signing up for esports tournaments. We also conducted user testing to ensure that the registration process is smooth and user-friendly.

Secondary Research:

We conducted a competitor analysis to understand the offerings of similar platforms in the market and identified areas where we can differentiate ourselves. We also benchmarked against industry standards to ensure that our esports registration page meets the expectations of our target audience. Additionally, we conducted surveys to gather feedback from users on their experience using the page.

Features in the form of Jobs-to-be-done:

- Discover and browse various esports tournaments and competitions

- Easily sign up for tournaments through a simple and secure registration process

- Stay updated on tournament schedules and details through notifications

- View and track your tournament participation history

Post Mortem:

After launching the esports registration page, we will conduct a post mortem analysis to understand the performance of the page and gather feedback from users. Based on this analysis, we will identify areas for improvement and implement necessary changes to enhance the user experience.

Here is the full list of ChatGPT prompts for PMs, downloaded by more than 800 people.

Limitations and Considerations

One limitation of ChatGPT that I would think of is that the output responses may seem generic. ChatGPT can only draw on existing information and inputs, and hence can’t create something unique. However, make no mistake that the responses are specialized. For example, outputting marketing copy and code are specialized outputs requiring real domain expertise and skill.

Despite all this, it can still provide important ideas or information that you might have overlooked. So it's still a net positive.

Another limitation is that the quality of outputs also depends on the quality of the prompt. Some amount of learning is required to write detailed prompts that result in higher quality outputs.

If you think about it, there is a real possibility that ChatGPT can become a crutch. Just how you rely on Google Maps for finding places, when you don’t have Google Maps, it becomes a problem. In the same way, are you even getting smarter if you use an AI to solve your problems?

The process of learning is through feedback cycles. ChatGPT can speed up the process of these feedback cycles, by providing creativity and efficiency gains. 

Final Thoughts

What’s very apparent is that ChatGPT is likely to make us editors rather than creators. You’re not going to be able to get away with only copy pasting ChatGPT for your work. You will need to be a world class editor of ChatGPT output. 

Editing skills comes with taste, skill, and experience working in that role.

In conclusion, ChatGPT is a powerful tool that can be used in various ways to enhance the work of a Product Manager. It can help with tasks such as creating product specs, roadmaps, customer journey maps, and more. The key to leveraging ChatGPT is to view it as an assistant, rather than a replacement. 

By giving it feedback and guiding it in the right direction, you can improve the quality of its output and free up more time for decision making. The use cases for ChatGPT are vast and can help with many aspects of a PM's work. Embrace this technology and use it as a tool for your own advancement.

Thanks to Kym Ellis () , , Steven Ovadia for providing feedback on early drafts of this piece.

📘 Read of the week: The End of Organizing - from Every (7 min)

That's it for today, thanks for reading! What did you think of this topic, let me know in the comments or reply to the email!

Share this newsletter with someone who might enjoy it!

Give feedback and vote on the next topic here.

Talk to you soon!

— Kavir

Refer a friend

If you’re new here, please subscribe and get insights about product, no-code, and startups delivered to your inbox every other week.

The future is here. ChatGPT is a step function improvement in AI that can change our relationship with work. With its advanced natural language processing capabilities, ChatGPT has the potential to revolutionize a wide range of applications, such as chatbots, language translation, content creation, and broadly knowledge work. 

I’ve already been using it for a variety of use cases in product management, writing, general problem solving, and honestly every part of my work. I will discuss more on the different applications of ChatGPT in next week’s edition.

Today, however, we are here to talk about how you can build a content library with AI and no-code tools.

AI-powered Content Library in No-Code

A content library can consist of any kind of content — blog articles, landing pages for different use cases and personas, marketing pages comparing your product with competitors, social media posts, pages focused on long tail keywords. This entire subset of articles can come under what’s known as Programmatic SEO. Earlier you would use different APIs to get this information, now you would use just one, which is OpenAI. 

I’ll show you how to set this on auto-pilot using AI and no-code tools.

Get Started

To get started you will need to create accounts on the following platforms:

Airtable: A cloud-based database platform that allows you to organize, manage, and manipulate data in a variety of formats and layouts. You can use Airtable to structure your content library, storing individual pieces of content as different columns or fields.

OpenAI: A research institute that is at the forefront of AI development, with a particular focus on natural language processing. You can use OpenAI's API to access and utilize its advanced AI capabilities, such as summarizing or generating text.

Zapier: A platform that enables you to connect and automate different apps and services, allowing you to create "Zaps" that perform specific tasks or actions based on triggers or events. You can use Zapier to push data from Airtable to WebFlow, as well as format the text properly for markdown.

Webflow: A website design and development platform that provides a visual drag-and-drop interface for building and styling websites. You can use Webflow's CMS (content management system) to manage and publish your content library.

Alternatively, you could use other platforms or tools that serve similar functions, such as Google Sheets, Make, or any other CMS. The choice will depend on your specific needs, preferences, and budget.

Structure the data in Airtable

Once you’ve set up your accounts and platforms, the next step is to structure your Airtable. This will involve deciding on the types of content you want to include in your library and how to organize them. For example, you might want to keep individual pieces of content as different columns in Airtable, so that you can generate them separately. Or you could add everything in one field. It's up to you.

You should also set up a status field in Airtable that you can change to trigger the workflows of generating content and publishing the article. This will allow you to control and automate the process of adding new content to your library.

You can duplicate this Airtable template and tinker around.

Set up the Webflow CMS

The next step is to set up your Webflow CMS collection.  

This will first involve deciding the type of fields that you require. You should ensure that your CMS reflects the types of content you want to include, such as articles, blog posts, or social media updates.

Then designing the structure and layout of the collection page template by adding sections to the page like a header, footer, intro section, main content, CTA, etc. You should also consider creating different fields or sections for different types of information, such as headlines, body text, images, or tags.

You’ll need some Webflow knowledge to set this up. Suggest check out Webflow’s YouTube channel for more info.

Enable Zapier connection

Once your Airtable and Webflow CMS are set up, you can enable the Zapier connection between the two platforms. This will allow you to transfer data from Airtable to Webflow and format it properly for markdown. When a new item is added to Airtable, Zapier will automatically create a corresponding entry in the Webflow CMS and add it to a collection on the website. You can either have the item directly published or staged for later. 

You can copy the Zapier connection here and tinker around.

Create AI content

At this point, you are ready to generate AI content for your content library. There are two ways to do this:

Semi-automated

In this approach, you will need to manually execute the script code to extract relevant information from Airtable and pass it to OpenAI's API for processing and repopulating into Airtable and then finally to Webflow using Zapier.

Original script code

// Change this to the name of a table in your base
let table = base.getTable('Documents');

let openaiUrl = "https://api.openai.com/v1/engines/text-davinci-002/completions";

// Prompt the user to pick a record 
// If this script is run from a button field, this will use the button's record instead.
let record = await input.recordAsync('Select a record to use', table);

if (record) {

    // Construct the body
    // Change the field name to match yours.
    let body = {
        "prompt": record.getCellValue("Text") + "\n\nTl;dr",
        "temperature": 0.7,
        "max_tokens": 60,
        "top_p": 1.0,
        "frequency_penalty": 0.0,
        "presence_penalty": 0.0
        }

    // Make a request to OpenAI API
    // Add your own API key in the place of Xses.
    let response = await fetch(openaiUrl, {
        method: 'POST',
        body: JSON.stringify(body),
        headers: {
            'Content-Type': 'application/json',
            'Authorization': 'Bearer XXXXXXXXXXXXXXXXXXXXXXXXXXXXX',
        },
    });
    let data = await response.json();

    // Comment this out if you don't want to see any outputs
    console.log(data)

    let str = /[^A-Za-z0-9](.+)/.exec(data.choices[0].text)[1];

    // Write to a cell, change field name to match your table.
    table.updateRecordAsync(record, 
        {
            'TLDR': str
        });
}

H/t Tommi_holmgren from the Airtable community

I was initially doing it manually by pressing a button for each record. But a viewer on YouTube suggested a way in which, you could just trigger it once and it will run for all records. Here is the updated script code. 

Updated script code

// Change this to the name of a table in your base

let table = base.getTable('Resume Keywords');

let view = table.getView('Generate Content');

let openaiUrl = "https://api.openai.com/v1/engines/text-davinci-003/completions";

let pageName = table.getField("Page Name");

console.log(pageName,"Page Name")

let results = await view.selectRecordsAsync({fields: table.fields});

console.log(results,"Results")

for (let record of results.records) {

    let pageName = record.getCellValue("Page Name");

    console.log(pageName,"Page Name 1")

    // Construct the body

    // Change the field name to match yours.

    let keywords = {

        "prompt": "List down 10 " + pageName + "\n\nskill keywords separated by commas",

        // "prompt": "List down 10 " + record.("Page Name") + "\n\nskill keywords separated by commas",

        "temperature": 0.7,

        "max_tokens": 512,

        "top_p": 1.0,

        "frequency_penalty": 0.0,

        "presence_penalty": 0.0

        }

        console.log(pageName,"Page Name")

    // Make a request to OpenAI API

    // Add your own API key in the place of Xses.

    let response = await fetch(openaiUrl, {

        method: 'POST',

        body: JSON.stringify(keywords),

        headers: {

            'Content-Type': 'application/json',

            'Authorization': 'Bearer xx',

        },

    });

    let keyData = await response.json();

    // Comment this out if you don't want to see any outputs

    console.log(keyData)

    let str = data.choices[0].text.trim();


    // Write to a cell, change field name to match your table.

await table.updateRecordAsync(record,{

        'Resume Keywords': str,});

}

Fully automated

In this approach, you will use script code to fully automate the process of generating and publishing content. This will require you to upgrade to the Pro Plan of Airtable, which allows you to use the "Run as script" function. You can use the following script code to set up your fully automated workflow:

Script code for "Run as script"

// Change this to the name of a table in your base

let table = base.getTable('Resume Keywords');

let view = table.getView('Generate Content');

let openaiUrl = "https://api.openai.com/v1/engines/text-davinci-003/completions";

let inputConfig = input.config();

let pageName = table.getField("Page Name");

console.log(pageName,"Page Name")

let results = await view.selectRecordsAsync({fields: table.fields});

console.log(results,"Results")

for (let record of results.records) {

    let pageName = record.getCellValue("Page Name");

    console.log(pageName,"Page Name 1")

    // Construct the body

    // Change the field name to match yours.

    let keywords = {

        "prompt": "List down 10 " + pageName + "\n\nskill keywords separated by commas",

        // "prompt": "List down 10 " + record.("Page Name") + "\n\nskill keywords separated by commas",

        "temperature": 0.7,

        "max_tokens": 512,

        "top_p": 1.0,

        "frequency_penalty": 0.0,

        "presence_penalty": 0.0

        }

        console.log(pageName,"Page Name")

    // Make a request to OpenAI API

    // Add your own API key in the place of Xses.

    let response = await fetch(openaiUrl, {

        method: 'POST',

        body: JSON.stringify(keywords),

        headers: {

            'Content-Type': 'application/json',

            'Authorization': 'Bearer xx',

        },

    });

    let keyData = await response.json();

    // Comment this out if you don't want to see any outputs

    console.log(keyData)

   let str = data.choices[0].text.trim();

    // Write to a cell, change field name to match your table.

await table.updateRecordAsync(record,{

        'Resume Keywords': str,});

}

Once you have set up your workflows and scripts, you can test them out to see if they are working correctly. If everything is functioning as expected, you should be able to create a fully functional content library with no-code and AI.

Final Thoughts

In conclusion, building a content library with no-code and AI is a straightforward and efficient way to generate high-quality, relevant, and engaging content for your website or platform. 

By leveraging the power of platforms like Airtable, OpenAI, Zapier, and Webflow, you can create a content library that is scalable, flexible, and dynamic. 

Don't forget to check out the YouTube video that explains everything that was discussed here for more detailed instructions and examples.

📘 Read of the week: The power of defaults - Julian Lehr (21 min)

That's it for today, thanks for reading!

Share this newsletter with someone who might enjoy it!

Give feedback and vote on the next topic here.

Talk to you soon!

— Kavir

Refer a friend

As we look back on 2022, We can't help but feel a sense of pride For all the challenges we faced And the obstacles we overcame, side by side

The year was full of ups and downs But through it all, we stood our ground We laughed and cried, we danced and sang And together, we learned and grew, as a team

We made new friends, we found new love We built new dreams, and chased them above We learned from our mistakes, and tried again And in the end, we reached our goals, and then some

So as we say goodbye to 2022, We look back on all that we've been through And we are grateful, for every moment For all the memories we've made, and all that we'll do.

That’s a poem about 2022 , written by our dear friend ChatGPT. What a time to be alive, eh?

2022 was the first year that was without major Covid restrictions across the world, so we spent more time outdoors. A lot of people started newsletters in 2020-2021, but haven’t continued them into 2022. I wanted to take a moment to thank you— each one of you — for motivating me to continue this newsletter into 2023 and beyond.

I want to share some of the most popular editions of The Discourse with you, but before that, a few notable events from the year:

1. Published 12 editions of The Discourse and passed 1000 subscribers for The Discourse this year, currently at 1693!

2. Kickstarted a new no-code focused newsletter called

3. Ran the second no-code cohort program and currently launching the third cohort. Apply here if you're interested in a special ₹5K discount for The Discourse subscribers. I’m doing an AMA tomorrow at 7 pm IST, so join the AMA if you're interested and have questions.

4. Became an official startup advisor (In case you or someone you know are looking out for a product or growth consultant/advisor, check out my work and hit me up)

With that out of the way, let’s move on to the top editions of 2022.

If you’re new here, please subscribe and get insights about product, design, no-code delivered to your inbox every week.

Subscribe now

Here are 5 of the most popular posts of the year:

1. 16 Best Product Management Tools to Help You Get the Job Done

This was my first curated list article. I usually don’t do them, but this resonated well with people and got a glimpse into what all tools a PM uses. You might as well replace all these tools with ChatGPT now.

The Discourse

16 Best Product Management Tools to Help You Get the Job Done

Thanks for reading The Discourse! Subscribe for free to receive new posts and support my work. As a product person, your responsibilities involve so many different things — right from discovery to execution. And for most of the responsibilities, you need to use tools to make your work efficient…

Read more

4 years ago · 4 likes · 2 comments · Kavir Kaycee

2. How Social Accountability Can Help You Achieve Your Fitness Goals

In 2022 I got the Apple Watch for the first time and I was curious to explore how the social accountability features improved my fitness levels.

The Discourse

How Social Accountability Can Help You Achieve Your Fitness Goals

Hey folks! 👋🏽 Kavir here. Welcome back to another edition of The Discourse. I’ve just got back from a workation in Goa and it was lovely. My first proper vacation since the pandemic started. Some pics here. I’m now all recharged to write more amazing content for The Discourse…

Read more

4 years ago · 4 likes · Kavir Kaycee

3. Why Figma Won: The Rationale Behind Adobe's Blockbuster Acquisition

Figma was the biggest tech outcome in 2022 (if you don’t include Elon’s hostile takeover of Twitter) I enjoyed exploring the nuances that made Figma what it was.

The Discourse

Why Figma Won: The Rationale Behind Adobe's Blockbuster Acquisition

Hey folks! 👋🏽 Kavir here. Welcome back to another edition of The Discourse. This time we’re discussing Figma. Few updates before we start: I conducted a session on building a Notion portfolio for product managers, for The Product Folk’s Insurjo program. You can check out the recording…

Read more

4 years ago · 4 likes · Kavir Kaycee

4. I Rewrote 58 Newsletter Titles with AI, Here’s What I Learnt

AI is taking over the world. I explored how AI was able to successfully rewrite 69% of my newsletter titles and made them more attractive to the reader.

The Discourse

I Rewrote 58 Newsletter Titles with AI, Here’s What I Learnt

Hey folks! 👋🏽 Kavir here. Welcome back to another edition of The Discourse. If you’re new here, please subscribe and get insights about product, no-code, web3 delivered to your inbox every week…

Read more

3 years ago · 5 likes · 2 comments · Kavir Kaycee

5. Going viral

I explored what it meant to go viral on Twitter, for a shitpost. And what were the elements that aided it going viral. This was a fun one!

The Discourse

Going viral

Hey folks! 👋🏽 Kavir here. Welcome back to another edition of The Discourse. If you’re new here, please subscribe and get insights about product, no-code, and startups delivered to your inbox every other week…

Read more

3 years ago · 3 likes · Kavir Kaycee

What to expect in 2023:

• More strategy pieces covering landscapes of apps and technology

• More tactical pieces on product thinking and product management

• More video content on AI, no-code and interviews with founders, product people, founders, VCs and more

Here’s a glimpse of the trajectory that The Discourse has had in the last few weeks. Thanks to everyone who has subscribed! You're the real MVP!

Appreciate you.

Thanks for reading, as always. Talk to you soon and Happy new year!

— Kavir

P.S. Hit the subscribe button if you liked it! You’ll get insightful posts like this directly in your email inbox every week.

Subscribe now

Today I’m announcing a brand new addition to my Substack publication: the The Discourse subscriber chat.

This is a conversation space in the Substack app that I set up exclusively for my subscribers — kind of like a group chat or live hangout. I’ll post short prompts, thoughts, and updates that come my way, and you can jump into the discussion.

To join our chat, you’ll need to download the Substack app, now available for both iOS and Android. Chats are sent via the app, not email, so turn on push notifications so you don’t miss conversation as it happens.

How to get started

1. Download the app by clicking this link or the button below. Substack Chat is now available on both iOS and Android.

Get app

2. Open the app and tap the Chat icon. It looks like two bubbles in the bottom bar, and you’ll see a row for my chat inside.

3. That’s it! Jump into my thread to say hi, and if you have any issues, check out Substack’s FAQ.

Join chat

If you’re new here, please subscribe and get insights about product, no-code, and startups delivered to your inbox every other week.

Subscribe now

Last week, I published a tweet that went like this: 

People who are from India might get the joke. For those outside of India, let me explain. 

I took what’s a popular tweet format of “the best city has the combination of A from X, B from Y, and C from Z”.  

In India, it’s universally accepted that Bangalore has the best weather, Bombay has really good people and is known for the spirit of Mumbai, and Delhi as the capital has the best infrastructure and an interconnected metro system.

If you don’t believe me, here’s a tweet from someone 10 years ago.

Now I decided to play with people’s minds and break their patterns. I tweeted out a combination that was, let’s just say, less than ideal. I don’t want to offend my Delhi friends and readers to say that I picked the worst out of all the cities, but you get the drift.    

So I flipped it around and voilà, that was the tweet.

The tweet hit 212K impressions, with ~1700 likes, 219 replies, and 239 retweets (most of which were quote tweets dunking on me). While it might not fully classify as viral by established patterns, it was pretty outsized compared to my 3.2K followers. 

Here’s what I learnt:

Sarcasm, not everyone gets it

Some people might say that sarcasm is a higher form of humor because it requires a certain level of intelligence and quick thinking. 

It’s about identifying an established pattern and breaking it. That’s also the premise of standup comedy. 

If you post a sarcastic tweet, 50% of people will get it, 50% will be outraged. Shantanu in a reply to one of my tweets summed it up best:

And this combination works best for virality on a platform like Twitter. Because it prompts people to react, leading to the algorithm thinking that this is something more people need to see. 

The sarcasm has to have a tinge of believability. If it is too direct, people won't buy into it.

From the beginning, it started getting replies which went on either end of the spectrum (I don’t get it, and you're wrong).

The few that got it, really enjoyed!

People just love correcting!

If you spend time on Twitter, you may have noticed the ‘bad math’ subgenre of tweets, where someone will intentionally miscalculate something to a large order of magnitude. Here is an example of one such tweet: 

People see this folly and want to chime in and correct it. The people who get that the bad math is intentional step in to correct those who don’t get it. 

In my example, many were like no, this is the correct response. 

Examples: 

People are overly attached to their cities

Moreso than other topics,  the city you choose to live in (or are born in) is very integral to your identity. It’s in your bio, it’s what you tell people when they ask you where you live. 

I don’t know what other topics (apart from sports and politics) are as divisive. Maybe the phone device camp you're in (iOS vs Android) or the desktop (PC vs Mac)

Final Thoughts

From this experiment, I have learned that sarcasm can be used to engage people on Twitter, as long as it has a tinge of believability. It’s fun to see people’s reactions, as long as you don’t take it personally. 

The replies to the tweet showed that while some people did not get it, there were others who understood the sarcasm and even corrected those who didn't.

It all boils down to identifying a familiar pattern, and then breaking it.

Thanks to Yihui, Sena Gürdoğan, Dhrumi Savla, and Danver for providing feedback on early drafts of this piece.

📘 Read of the week: My Philosophy of Product Building - Nathan Baschez (11 min)

That's it for today, thanks for reading!

Share this newsletter with someone who might enjoy it!

Give feedback and vote on the next topic here.

Talk to you soon!

— Kavir

Press the ♥️ button if you liked this edition.

👇🏽

If you’re new here, please subscribe and get insights about product, no-code, web3 delivered to your inbox every week.

Subscribe now

The AI space is heating up particularly with generative AI art like DALL-E, Midjourney, Stable Diffusion, and several applications built on top of them like InteriorAI (Interior Design) and AvatarAI (Professional Avatars). Along with this increase in interest, there’s been a resurgence of text-based generative AI.

When GPT3 from OpenAI first came out, we got the first wave of AI applications like Copy.ai and others. Now we’re seeing different applications come to the surface. With that in mind, I just got access to Lex, an AI writing tool powered by GPT3 created by Every (a newsletter publication collective). 

This seemed to be the first AI tool to be built for me — a newsletter writer — whereas earlier options were catering towards copywriters and social media managers

Lex has two main features currently. You can generate paragraphs of text — by supplying a prompt, an outline, or bullet points — or you can use the AI to create title options. While I’ve used both features, I was curious to evaluate the Title rewriting feature, and so I did. 

I used Lex to provide alternative suggestions to the 58 published editions of The Discourse and here’s what I learnt:

Why titles matter 

I’ll be the first to admit that I’m not the best at writing catchy titles. tend to go for subtlety with my titles, but a title needs to be anything but. It needs to grab readers and pull them in.

Veritasium has this great video in which he introduces the concept of Legitbait as a credible alternative to the word clickbait. He goes on to define clickbait as content that doesn’t live up to the title. 

Content that follows up a catchy title with real substance is legitbait.

In a world where YouTube videos have outrageous titles and thumbnails, Twitter threads have formulaic hooks – all in a war for our attention — you have to play the game or you will end up creating content for a tiny subsection of your audience. 

What was lacking in my previous titles

After playing around with Lex for rewriting the titles, I realized there was a common theme.

My titles were short

The average title length for past issues of The Discourse was 32 characters. I think I leaned on shorter titles, because I subconsciously thought that it would look better in the email subject.  

They weren't descriptive enough

I noticed that my titles didn’t actually describe what the edition was about. They were very subtle. For example, for an article about the meditation app landscape, I came up with this title: Take a deep breath. It could generate curiosity, but in most cases, it might not draw the reader in because it was a bit vague. 

I was more descriptive in the subtitle, but I think the order needs to be reversed.

Titles work differently for blogs vs. newsletters 

Titles matter much more for blogs, because SEO is highly weighted towards titles and then a combination of the URL, meta description, and the actual content. 

But what matters for newsletters is the combination of Title + Subtitle. Because in an email, you get both the subject and the snippet.

On a mobile device, subject and description can only be 40-70 characters together, depending on whether your reader is using Apple Mail or Gmail (the two most popular email apps for mobile devices). 

So with that in mind, how did the AI fare?

(Drumroll please) 🥁

Rewrites by the AI

This was just a sample, the full list is here.

Which ones do you like best? Let me know in the comments. 

You can probably spot a trend here — the AI-generated titles have the right keywords and are more descriptive. However, there are a few downsides that show up when you use this enough times.

Reuse of common words

AI tends to reuse common words in titles such as:

• Power, Advantage, Benefits, Importance, Matters

• How to, Why

Presumably because the training data was trained on many SEO-focused blog articles.

Longer titles

The titles are substantially longer than the ones that I wrote (53 characters vs 32 characters). While this would be a good thing for blog articles, it’s not so great for newsletters. Like we discussed earlier, a  title this long might not fit in an email client on mobile.

SEO focused

These titles would do much better for SEO blog posts than for newsletters. They also look more SEO’d. That’s not a ding on Lex, just a nature of the dataset of GPT3.

Results

Here were the results of the rewriting process. In effect, I ended up modifying 69% of the titles after generating them with AI. Sometimes I used the exact phrasing the AI gave me, and in some I modified them slightly. 

The 30% of titles that I retained were closer to my heart. Titles like “Speed as a Feature” or “Whiteboarding in a Remote World”.  

Verdict

Some of the AI rewrites were significantly more descriptive than what I wrote. Taking a few examples: Take a deep breath is ambiguous whereas Business of Meditation Apps is a dramatic improvement. Similarly Secret to Musically's success is better than Product Growth Lessons from Musically. 

And finally, when it comes to this article, the AI generated the following titles: ​​​​

• The Benefits of Using AI to Generate Titles for Your Newsletter

• How AI Can Help You Write Better Titles for Your Newsletter

• Why You Should Use AI to Generate Titles for Your Newsletter

• How AI Generated Titles Can Help Your Newsletter Stand Out

• The Advantages of Using AI to Generate Titles for Your Newsletter

Clearly, I didn't use any of these. but these were pretty good by themselves. 

So what did I learn? You won’t always use the titles the AI suggests, but since it creates multiple alternatives to choose from — you can take the best pick or the options might spark some totally different variation. 

I know that I plan to use this tool to create title options for each edition. 

I don’t need to blindly go with what the AI gives me, but I know that’s a good starting point to work from.

Shout out of the week 📢

Category Surfers

Discover the next big thing in SaaS before it explodes. A weekly breakdown of fastest-growing software categories. Subscribe for free here.

Thanks to Dhrumi Savla, Russell Smith, Danver Chandler, and Nicholas Bouchard for providing feedback on early drafts of this piece.

That's it for today, thanks for reading!

What do you think about AI Writing Tools? Reply or comment below, and I'll reply to you. Give feedback and vote on the next topic here.

In case you or someone you know would like to hire me for product consulting or startup advisory roles, check out my work and get in touch.

Talk to you soon!

Follow me on Twitter @kavirkaycee