Check out the conversation on Apple, Spotify, and YouTube.
Is the product builder role real or hype (01:40)
Aakash: Maybe my most popular podcast guest ever is Ankit Shukla, former senior product manager turned founder of Hello PM. He messaged me with this new trend. After studying tons of job postings, he’s found there is a new role out there, the product builder.
I, for one, am a little bit skeptical. It might be hype. It might not be real. So we’re going to go through with Ankit today. Is this a real role? How much does this role pay? Who can become this role? He’s going to give you the real case studies that have worked for his students to land this role at companies like Meta, Amazon, Google, even Sarvam AI, one of the leading AI companies in India.
And we’re not just going to cover how you can land this product builder role. We’re going to cover the fundamentals of being a product builder. We’re going to go through the power framework to think through how to build things, and we’re going to go through a real case study of building a product.
So this is an action packed episode. Not a single minute is wasted. And with that, let’s get right into it.
Ankit, thanks for being back on the podcast.
Ankit: Thanks a lot. Thanks for having me here. I’m always excited to have a conversation with you, Aakash.
Aakash: Ankit, a year ago we created the blockbuster video, how to become an AI product manager. It feels like the product builder is becoming the new thing. Here we’re looking at LinkedIn. They got rid of their associate product manager program. They created this associate product builder program. So what I want to know is, is this product builder role real, or is it hype?
Ankit: Aakash, that is actually a billion dollar question, because I get this question from a lot of people. The forums are filled with this kind of question.
So I thought that rather than giving my opinion about what I think about this role, whether it is hype or it is value or something else out there, I have actually taken the help of Claude and GPT in order to go through actual job descriptions. More than 10,000 of them, in order to understand what is happening with the real jobs. Not what the influencers are saying, not what people are mentioning on YouTube or on LinkedIn, but what the exact job descriptions of some of the top companies of the world say.
And I was able to find that yes, for more than 30% of all the PM jobs, they are asking for some good AI skills. And among these AI skills, they are not only asking for just the tools, you knowing n8n or you knowing some kind of RAG or agentic AI thing out there. They also want you to understand the core of what problems should be solved by AI and what should not be solved by AI. So that judgment is what people are paying for.
If you combine that product judgment along with your understanding of AI and the tools that you can use, I think all the PMs, or people who want to become PMs, are sitting on a very big opportunity.
The three stages every PM has always worked in (04:39)
Aakash: Okay. So this role is real. What is a product builder? What makes a product builder different from an FDE, from an AI PM? What I don’t understand specifically is who is this person. What’s their background? What do they do?
Ankit: I’ll tell you. Before even talking about any of these terms out there, which are mostly jargon, let us talk about the exact core of the problem.
We have seen for the longest period of time that as a product manager, or as anyone who is a problem solver in the company, you are operating in one or more of three stages.
The first is discovery. Discovery basically means you are understanding the customers, doing the market research, understanding the data, in order to build a roadmap. A roadmap of what to build. This is what you are going to build, or what your company is going to build. This is the roadmap of the product.
After you have built the roadmap, you are going to detail the instructions and then work with your development team and design team in order to do the delivery. What is delivery? You will give them the specs and then they are going to build the products. And after they are building it, you have to make sure that you are following up, because things can go wrong. They can get delayed, and maybe people will not ship them.
So after the delivery, you are also going to do the third phase, which is distribution. Distribution basically means you taking them to the customer, so that they actually pay for the product and adopt the product.
These are the three things that any product manager has been doing for a longer period of time. Even if you are working in any other role, your role is directly or indirectly related to one or more of these three fields.
Now, what used to happen in the industry is that a product manager will invest a lot of time in discovery, and in discovery they are doing a lot of research, looking at a lot of data, talking to a lot of people. AI has now made them a promise that most of these things you can now do more efficiently.
For example, rather than manually browsing the sites of your competitors, you can give them to Claude or to ChatGPT, or you can use this tool called NotebookLM to give it all the content. It is going to give you an almost perfect research report that you can use to make the decisions. That is one small use case, and it is going to not only make you productive but make the quality of your decisions even better.
So that is the first reason why it is evolving.
Why delivery was always the bottleneck (10:12)
Ankit: The bigger reason is the block of the delivery.
What used to happen was that the major block in a product company was mostly the delivery, because of the nature of coding and building products. It used to take a lot of time for developers to build something. And because of that, there had to be a lot of gatekeeping here, which is that every idea you discover, you are not going to put it into delivery, because your teams cannot build it.
And because of that, what used to happen was, although it was a good move that you are going to play on your judgment, it also led to a different thing. All the successful product managers in the world will agree to the fact that it is almost impossible for you to say with 100% confidence that this idea is going to work. And because of that, what people used to do was, because they have to be confident about their ideas, they need to get clarity before they are building something, they will take only some very safe products to the delivery. And because of that they will take some mediocre decisions, or maybe some covert decisions.
So now the promise of AI is that you can do delivery faster. Your engineers, who are AI natives, can use tools such as Cursor, such as Claude Code, in order to do things faster. They can test it, they can deploy, and maybe create things at scale. So now this has been reducing.
And now understand, delivery is not only a separate part. As a product manager, when you are doing testing of your product, or when you are doing validation of your idea, or whenever you are doing experimentation, all of these are maybe interchangeable terms. You can also leverage AI in order to do the experiments by yourself. You do not need to always go to the engineering team and get their bandwidth in order to do the experiments. You can do it by yourself, and I’m going to show you in this video later on how you can actually do all of these things. No matter how complex the product is, you can at least get started in order to test it with the customers.
What the product builder role pays (12:03)
Aakash: So how much do these product builder roles pay? Should people really be pursuing these?
Ankit: I think that is going to be an interesting question. So I actually did the research on the same as well, and we were able to find that AI PMs and AI related roles in India and across the world are willing to pay a premium of more than 20% as compared to the traditional counterparts. So if you are a PM, you might be paid X, but if you get the AI skills, you might get maybe 1.2X, and maybe up to 1.8X or 2X as well.
Aakash: What do those numbers actually look like in absolute terms in the US and India?
Ankit: To do the comparison, for example in the US, we have seen that $195,000 is the median salary that is crawled across the 12,300 jobs. And then we have been able to see that if you are senior, for example if you have some years, no matter whether as a traditional PM or as an AI PM, if you are able to get these AI skills, you have the potential to make as much as $250,000 to $560,000, depending on the companies that you are working at, your own background, plus how good you are in terms of delivering the products.
You can also see that across the levels. If you’re just starting out, maybe you’ll start from a $120,000 base salary, which is not bad if you’re just acquiring the AI skills. And once you are in the senior roles, that is where your judgment and your AI skills truly give the leverage to the company and to your career. So it can reach maybe up to $340,000 or $350,000, and total compensation at the top labs can maybe reach $800,000. And we have seen the recent news that Netflix has advertised an AI product manager job of maybe up to $900,000.
Aakash: And in India?
Ankit: In India, these have been the salaries. If you are a fresher looking to get into product management, you know some bit about AI, and you have some good projects, you can make a starting salary of maybe 12 to 16 LPA. Similarly, for two years’ experience, it is 22 LPA. And for people who have already been a PM for more than five years, if you just add those right AI skills and build a strong portfolio in order to showcase your work, I’m sure you can get these kinds of salaries with the right kind of approach.
Why the power framework exists (14:26)
Aakash: So there’s a real premium to learning this product builder skill set. What’s the right framework for thinking about how to be a product builder?
Ankit: Now, everyone in the world knows that if they are able to learn AI, they should be able to add this much needed leverage and seize this opportunity. But the issue is that there is so much information everywhere that whenever people start, they either get anxious, or they get too excited and start going into the depth of the wrong things.
So in order to solve that problem, I am giving you a crystal clear framework that will reduce all of this anxiety, that will eliminate all of this confusion. You just have to follow this. I call this the power framework. If you practice it in your own company, I’m sure you should be able to confidently talk about AI projects. You should be able to build them. And then you should be able to mark it down very clearly and very confidently in your resumes, and get that leverage that we have just shown you on the screen in terms of the real salary numbers.
Now, understand, I am not talking about the adoption of AI. Many companies are still struggling with the adoption, where their people are not using AI, they’re not able to find the use cases, or maybe they’re apprehensive about their jobs. But I’m not going to talk about adoption. I’m exactly talking about the advantage, which is companies who are willing to adopt AI, or who have already adopted AI, and how they can actually get the ROI from the same. That is what this framework is all about.
This framework can be utilized even if you are an entrepreneur, if you are a small company, a bigger company, anyone, if you want to truly leverage the power of AI. For example, your CEO might be saying that everything can be done with AI, but there’s a disconnect between what you think as an employee and what your CEO thinks. So this is going to bridge that particular gap, which is that this is going to actually give you the measurable ROI.
P is for possibilities (16:19)
Ankit: This is the framework that I have built. The first P, it stands for the possibilities.
Now, traditionally in product management we always started with the problems, but AI is different. AI is different in the case that there were some problems that we were not even thinking about as problems. We were thinking of them as harsh realities, or we were not even thinking about solving them, because we had not seen a technology like AI.
So starting with the problem is good, but if you always start with the problem in the case of AI, you might not be even looking at the whole picture. So the first step in the power framework stands for possibilities, which means you should do a very detailed research for your company in order to understand what the possibilities with AI are, which is what the things that AI can do are.
For example, AI can right now understand, transform and generate text. It can look at a lot of meeting transcripts and generate the summaries. It can generate code. It can generate content. So UTG is a framework that I use, which is understand, transform and generate. Those are the three things that AI can do.
Now, I would also recommend everyone to go to their competitors, look at their industry, maybe look at the nearby industry, and try to understand what different kinds of people are doing with AI. So that should open up your mind that yes, these are the possible things that I can do with AI. So this is step number one. Understand the possibilities. And every company who is serious about AI should create their own database of possibilities, by researching across and getting that horizontal exposure into how different companies are leveraging AI. Make sense?
Aakash: Got it.
Ankit: One other proxy to look at how companies are leveraging AI is that you can go to the customer stories or testimonial pages of these frontier AI labs, such as Anthropic or DeepSeek or Moonshot or OpenAI, and you can understand their customer stories or testimonials. Then you’ll understand how different people in the world are using AI, and that should also give you a lot of other possibilities as well.
So that is the first part. And after you understand the possibilities, when you look at the problems in your company, you will start thinking that yes, this is how I can approach this problem with AI. So this is the first P, which stands for possibilities.
O is for opportunities (18:43)
Ankit: After possibilities, we have the O, which stands for opportunities.
Now, in possibilities we were looking at the world. What is possible with AI? In the opportunities, we look at a company. What are the different opportunities or problems in the company where we can actually use AI, or maybe even not use AI, in order to solve that particular problem?
For example, if I look at a product manager’s job, I can break it down into multiple parts. Whenever I do discovery as a product manager, I have to do multiple things. In discovery, first of all, I have to do the research. It could be a customer research or a competitive research. Whenever I’m doing the user research, I have to draft interview questions. And many product managers I have seen make a mistake that they ask very leading questions in the interviews.
Similarly, you are going to build a lot of artifacts as a product manager, and these things used to take a lot of time. Not the thinking part, but the documentation part. So now AI can help you. So these are the opportunities for you.
You think that right now, in order to build a product, a product manager has to do months’ worth of research, and in those months they are researching multiple companies, talking to multiple people. Can AI reduce that? So this is the opportunity part for you.
So what you need to do is talk to the people in your team, understand the whole business structure, and try to understand what the ways are, what the places are, where if you implement AI you should be able to get some or the other kind of ROI.
And now, never start with only one use case. Many people make a mistake that they only think about one thing and they start implementing it rather than looking at the whole picture. So for you as a product manager, one skill that is very important is about judgment of what problem to pick in order to solve with AI. If you are just taking some small redundant problem which is not very frequently occurring, you can build it with AI, but you’ll not be able to get the ROI out there. Make sense?
Aakash: Yeah.
W is for workflows (20:44)
Ankit: And after the opportunity part, we have the third part. Now this is where you have to put a lot of attention as a product manager. The W stands for the workflow.
Now, because you really want to solve the problems, you cannot afford to be very high level. You cannot just say that I can improve discovery or user interviews with my AI. What you need to do is go deeper. You need to understand the workflows.
For example, when I talk about research, I know that in order to take better decisions, I should always make sure that my customer support tickets are being considered. I am reading them, and I’m making sure that they are also giving me some inputs about my roadmap. I’m looking at all my reviews which are coming on Google Play, or maybe some website like g2.com if I’m a B2B company. I should be able to understand them. Whenever my sprint reviews are happening, whenever my leaders are conducting meetings about the product, I should be able to get all of these inputs in order to decide what to build.
So now you have to understand the complete workflow. If this is the opportunity, what are people doing right now in order to do their work? And not only for a PM’s job. You can look at any job in your company which people are doing, in order to make them productive. So the third part is the workflow. And here you have to conduct maybe a lot of internal interviews with your stakeholders to understand what it is that you can improve with the help of AI.
Here I have two approaches to think about. In the workflows, you have to look at two parts. One is you should look at opportunities in the workflow which are optimization opportunities, and second is the innovation opportunities. By optimization I mean initially you used to spend two or three hours every week in order to do this. Now maybe you will take one or two hours. So it is optimizing on time. But innovation is something that initially you were not even thinking about this process.
So I’ll give you an example. At Hello PM we keep doing a lot of experimentation with our content. We do a lot of experimentation with how we are delivering the content to the students, and we are also doing a lot of experimentation with how the LMS is going to look and how the website and the landing page are going to look.
Initially what used to happen was that I’ll take the decision, or my people will take the decision, mostly with the limited empathy that they have, because we have a lot of people in the program. It is almost impossible to have empathy with everyone.
So what we have done is, in the first call when anyone joins the program, we ask them to introduce themselves. We have taken about 200 of these calls with all the people out there, and then we have fed it into Claude, and then we created a user persona document from the same. I created a new folder, put all the transcripts over there, and then I asked, can you create some personas which are representative of the people who are introducing themselves, because they have given all their information?
After that, whenever I take a decision, I ask my Claude that I’m looking to take this decision, can you walk that decision through all the personas that are mentioned in this folder? And then Claude is able to tell me which persona it is going to do better with, and where I should take that particular decision. So now I can create empathy at scale with the help of this small system that we have created.
So that was the innovation use case.
Aakash: Basically, creating personas from your real transcripts enables you to scale your empathy.
E is for engineering, and why it comes last (24:09)
Ankit: You are able to observe that so far we have talked about possibilities, opportunities and workflows. But when you talk about AI automation, or AI workflow leverage, or anything about AI, you will observe that a lot of people actually start with either n8n or Claude or something else out there.
But if you observe here, we have not even taken the name of the tool till here. And that is something that should be the biggest takeaway from this particular masterclass, which is, if you really want to create ROI, please do not start with engineering.
So E here starts with engineering, which is how can I create these workflows in n8n, in Make, in Zapier, Google Gems, custom GPTs, or maybe using agentic AI solutions such as Claude and all.
Many people make a mistake that they always start with this engineering part and then they try to retrofit everything over there. But my framework is, in the power framework, always remember POW comes before the engineering part.
If you do this, there are very high chances that you’ll not make the mistake of putting your AI slop everywhere. If the engineering is coming after the POW, there are very high chances that you will not produce the AI slop, because you are actually starting with the right kind of use cases.
And across the 12,000 job descriptions that we have seen, we have also observed that RAG, agentic AI and prompt engineering, they were not the number one skill. The number one skill was identifying the right use case for AI. That is the thing that people are willing to pay you the most for, because the engineering part can actually be taken care of by AI.
Aakash: Exactly. That part’s a lot easier now. And so the real alpha is in the POW.
The levels of AI engineering, zero and one (26:48)
Ankit: And I’m going to show you, in just a moment after we cover the engineering and the R part, one product that we have just developed just a few days before, and that’s a production grade product, and how easy it was to develop that.
So now, talking about the engineering part. This is a very interesting part, everyone is interested in the same.
The engineering means that now that you have certain kinds of use cases, you understand the problem, you understand the gaps in the workflows, you know what you are going to build. Now the part is, how are you going to build this? And many people are confused about the right set of tools that they need to use. Somebody talks about Claude Code, somebody talks about Codex, somebody talks about Google Gems, somebody talks about anything. So what I’m going to do is try to give a simpler framework to understand all of this, and maybe put a method to this madness.
I start with the levels. For me, level zero is where you are just prompting your way through. You have gone to chatgpt.com or claude.com and then you are using their chatbot as it is. It will give you certain leverage, and for many use cases that is good, but it will not be able to solve a recurring use case very efficiently for you. You are wasting your time out there by prompting it again and again. So the first level is prompting. Almost 100% of people are doing that.
The level one that I give is for creating some reusable prompts, which is you can either use something called Google Gems, and I believe you have a fantastic detailed video about Google Gems on your channel, so people can check that out. And second is you can also create these custom GPTs.
For example, if you want to create a PRD, you can create a Google Gem with all the templates that you have, and then whenever you want to create a PRD you can go to that Gem, you can give your idea, and then it should be able to generate the PRD for you. Similarly, you can also create a resume optimizer. So for many people we suggest that you can take a job description, you can take your resume, create a Google Gem, and then you should be able to iterate your resume as per any job description at scale.
So the second part is you are taking that recurring prompt, you are improving it, and you are putting it into a reusable structure or a shell, such as Google Gems or a custom GPT.
Level two, skills and progressive disclosure (28:28)
Ankit: But there is an issue with Gems or custom GPTs, which is that you always have to switch between the windows. The context is always switching. If you want to create a PRD with a Gem, you have to go to that particular window. In order to do that, you have to click on that Gem.
So for that we have level two, which is the skills part and the connector part. So here Codex and Claude Code come in, where you can create the skills.
The problem with Gems was that if you try to put a lot of instructions in your Gems, it is going to run out of context. If you are having a Gem to write a PRD and also to write your resume, there can be very long instructions and it is going to run out of context. Your model will either hallucinate or it is going to give you the bad output.
Now, Anthropic has solved this very beautifully with the skills. What they do is they use something called progressive disclosure. What it actually means is that you can keep on chatting in the same window. Skills are nothing but instructions to do a particular task. If you have added those instructions, Claude will identify from your chat intelligently whether it needs this skill or not. So your context is saved, and the skill is only loaded when it thinks that yes, for this use case I should use a particular skill. So that is the second level.
Now, skills are not rocket science. This is a trick that I tell everyone who wants to create the skills. Don’t get overwhelmed by them. Open your Claude, do the task that you want to do. And when you are done with the task, when you’re happy with the results, just ask Claude to please create a skill out of the same. And then you don’t have to think much about that workflow, and Claude is going to tell you.
And whenever you are using that skill, if you learn something, if you think that you have prompted more than the skill, you can ask Claude to again update the skill based on what additional prompts you have given. That creates a scalable system, and that will reduce the fear that you need all the skills ready before you can do something.
Aakash: Yeah. And this level is really when I think you start to see some huge ROI. Gems and custom GPTs, I feel like they’re almost basic at this point. This is like the real unlock.
Connectors and the 10am product brief (30:38)
Ankit: So when the skills connect with the connector. If I want to show you guys what skills are and how you can add them, you can just click on your customize option in your Claude Code. And understand, skills are not only limited to Claude right now. Every AI tool is supporting this. That has become a de facto standard in the industry.
So click on customize, you can add the skills. As I’ve told you, the best way to create the skills for your work is do the work with Claude and in the end ask it to create a skill. That is going to make sure that next time you don’t have to repeat your instructions.
And the second part is for the connectors. Now this is very important. For example, I can connect my Gmail, my Slack and Google Calendar. For a product manager, a very interesting use case is that I can connect my analytics such as PostHog, Mixpanel, Amplitude, and I can set up an automation that every day in the morning at 10:00 a.m. tell me what has been happening with my product. And I can give it certain metrics and it is going to automatically send me either the messages, or it can also trigger an email for you.
Understand, this is a big productivity unlock, because I have seen many product managers struggling with their dashboard every day in the morning in order to log in and check what is happening. And it is going to give you a lot of data in natural language. So you can just share it with your team out there before this break.
Aakash: Yes. GitHub integration, Gmail, Google Calendar, Slack, Amplitude, whatever your analytics is, Tableau. Those all are pretty much essential connectors.
Level three is vibe coding, and the workflow tools above it (32:01)
Ankit: Now that we have understood level two, after level two, let’s say if you want more granular control, then you can go to level three. Now for me, level three is, I would say, vibe coding.
If there are a few things that you want to build and you are not able to get an out of the box solution with connectors and skills, or if you want to give an interface to your users out there where you do not want to log into Claude or something, you can do vibe coding. And my favorite platform to do vibe coding is actually a free one, which is Google AI Studio, where you can build almost all the simple products that you can think about. And this is going to give you more autonomy and more control over what you are going to do.
But you don’t have to stop at level number three.
If you are looking to automate the workflows in your company and these things are not suiting you, or maybe there are some tasks that are going to happen at the back end. For example, whenever a lead comes to your website or your product, you want to make sure that you are doing the research on that, you are trying to understand how important that lead is, and you want to assign it to the right kind of salespeople or the telecaller out there. So for this, maybe the skills connection and everything, you do not want to give access to this to your salesperson.
What you can do is you can simply use an AI workflow tool such as n8n or Make. And guys, understand, these days things have become so simple that you do not even have to understand what the hundred kinds of nodes that n8n or Make has are, because many people get overwhelmed by the same. These platforms have something called an AI assistant, where you just give it the prompt and it is going to create the whole workflow for you. You are going to ask it to test, so that you are able to build something from scratch even without knowing what exactly these nodes are. But this gives you more control. You can check the whole process, which unfortunately vibe coding does not.
Aakash: Is there a level five?
Level five, and why the Claude Code versus Codex debate is over (34:06)
Ankit: Yes. And this is what I’m going to show you as well. So level five is a level where you take the most control. This is where the production grade applications are built, and this is where the future is going to go. This is what the future is. And this is actually using tools such as Cursor, such as MCPs, which are able to connect you with maybe AWS or Vercel of the world, and then using that in parallel with your tool of choice, which is Claude Code or Codex.
And also, guys, one very important thing. Many people ask me whether Claude Code is good or Codex is good. Let me tell you very honestly, it does not matter for your use case. Unless you are solving a PhD level mathematical problem, you have to think about whether I should do this with Codex or Claude Code.
We have gone ahead, because it is our job to teach people and give them some kind of benchmarks. We have taken all the common use cases, which is building an e-commerce website, building a social commerce website, building a social network website, building a gaming platform. We have tested these models, both the platforms Codex and Claude Code, and Cursor, plus the DeepSeek harness. And all of them were able to behave almost similarly. There were some token issues, there were some cases where we had to give some more prompts to some AI, but eventually everyone was able to build a functional website.
So I think we should end this debate. For 90% of the use cases it does not matter whether you are using Claude Code or Codex out there. Even the very simple model Composer, from Cursor, actually does a better job.
Aakash: But I assume there’s even a level six, right? Within this, you could go into loop and graph engineering, and you could just keep going down and down this rabbit hole.
Ankit: Yes, but I’ll tell you, you should actually stop here. After this, you should understand the power of AI.
So now, what is the power of AI? Understand that you have to understand a mental model. In order to build any product, you need a certain kind of code. And if AI is able to code that, it is able to run that code, it is able to understand the results. So why can’t you give all the permission to the AI in order to do that? Why don’t you leverage the mind of AI in order not just to plan, not just to execute, but also to iterate?
That is where you get into harness design, and that is where you get into the agentic. Or maybe a better word would be loop engineering, where the AI decides all three parts, which is making sure that it is able to plan, it is able to execute, and then it is able to evaluate whatever it has written. And if it is right, it is okay. Otherwise, it is going to plan again and repeat this.
The case study, replacing a $550 a month email platform (36:39)
Ankit: Let me get out of theory. Let me show you a real product that we have built.
Let me start with the problem statement. At Hello PM we have an email list of almost 120,000 subscribers that we have built in the last three years. Now, in order to send these emails, we use a service from Amazon, which is a popular service, which is Amazon Simple Email Service, SES. And for sending maybe a million emails every month, we have to pay them just $100 or $120.
But the problem is that SES is actually an API. They don’t have an interface where we can create the campaigns, send the email and do a lot of things out there. They do not give us the control. That is only an API. So we have taken a third party platform which enables us to create the campaigns, manage the users, and make sure that we are able to track everything.
And then from the day we started sending a million emails, or maybe ten emails per month, or five emails per month to our users, you’re able to understand that our costs are rising. So last month only we were able to see that we have paid almost $550 or $700 to that particular platform, over and above the Amazon SES.
So then I thought, why don’t we try to build something of our own? And I took a challenge that although I know about maybe Python, I know about using PHP, I know about using Java, I’m not going to use these languages, because I want to try if I can build it in something that I don’t know and I’m able to manage it at a production level.
So I took a challenge that I’m going to use a completely modern stack, which is Node.js and TypeScript, and maybe I’m going to deploy it on AWS using all the native services that they have. And I’m not going to touch the dashboard of AWS, because in AWS you have to figure out which thing is where, and then how you are going to deploy it, and then how you are going to take the backups and all. So that is complex. So I thought that maybe I know some of it, but now I’m going to assume that I do not know anything.
Starting with a layman’s spec, not a PRD (38:39)
Ankit: So what I did was, this is the exact process that I have done. I started with a very basic document. Understand, although I have created the most famous video around how to create PRDs, I did not follow that. I just did it like a common man.
So I understood what the features are that I need. So I have written this very simple, very layman term specs. Specs is add the leads, add the fields of the lead, add text to the leads, and then import the mail from Mailbuster or the tools that we are using. We can export the data from there and then import into a new system. And I have mentioned things in a very simplistic way. A/B testing, segments and all of these things.
After that I have taken this at a very basic level and then I gave it to ChatGPT. You can also use Claude. I don’t think we have to debate about what model to use. All the models, this is very basic work. The model should be able to do this.
So I just created a simple prompt. I have not used any prompt engineering there. I wanted it to be as much for the common man as possible. So, help me create a concise but complete, straight to the point specs document from the features of an emailing system. I’m trying to replace it with an in-house tool and I will import leads from the export. I’ll be using Claude Code to develop this completely. Give me solid specs and tech recommendations which I can give to Claude Code to build, deploy and manage this.
And then I copied this all. After that it created a doc for me, but I did not like the doc, so I asked it to please write it into a markdown file and explain the features more. And then it created a document for me.
So now I have a spec document which is well written by this. Understand, initially what we used to do was write complete specs that we are sure of and then give it to Claude. But right now the level is that you can write at a high level. You can ask Claude or GPT to improve it, and then you can read it and maybe improve it. It allows you to comment everywhere that you want to. So this is what it has created. I have read all of this.
The Claude Code build, from plan mode to sleeping on it (40:34)
Ankit: After that what I have done is, I have gone to my computer and created a new folder called postbox. I am naming this tool postbox. I pasted this complete spec file there, and this is what I have done.
Now I’m going to show you my whole Claude chat for this particular product. And understand, this is just from yesterday, which is, I’m not sure it’s even 24 hours ago.
So look at this, what the exact prompts are that I have used in Claude in order to build something. But before I show you this, let me show you what we were able to build. So we wanted to build an email marketing platform that is built on top of Amazon’s SES, or you can plug and play any kind of email sending API out there.
Now, I had some features that I had in mind, so I’ve documented them in a very simplistic term. Then I asked GPT in order to expand into the specs, and then I created a folder in my computer. Claude Code was already installed. I opened Claude Code. I opened that folder with the specs already there. Then I asked this to Claude Code. This is what is written in the project. Understand this project and tell me what’s the best way for you to implement it end to end. Feel free to make changes to the specs if you like.
And then I have put it into the plan mode, and then it has done everything and created a plan for me. It asked me a few questions. After that it has created.
Then I also asked it that I do not know about AWS, so help me, how can I do this? So it asked me that you can just pull in the MCP out there, and then I’ll be able to take care of AWS as well. So I do not have to even go to AWS. It just pulled the MCP out there, and then it told me how to get the credentials. I just copied them. How to get the credentials? Just go to AWS, click on identity, it will give you the API keys. That’s it. And then I’ll maintain them, not in the chat, but maybe I’ll create a folder out there.
If I’m being a bit quick here, understand, everything that I’ve been doing has been instructed by Claude Code. So if you try the same prompts, I’m sure you should be able to figure out your way out there.
So I have created a file and put all the details out there, and then it started doing this. So it has created this plan. You don’t need to read this plan. If you want to, you can. But it has divided the whole plan into nine stages. I gave it approval stage after stage, after it has tested. But after some period of time it has tested everything. I was also testing along the way.
And eventually what we have done is, I have done phase one, phase two, and eventually I also asked it to go with slice three, and then it has gone slice three, go with slice four. And then eventually, look at this. Now I was fed up. It was almost like 20 hours from now, so it was already 12 or 1 in the night. So I asked it that I’m sleeping now for about six hours. Continue doing one slice after the other without waiting for more. Make sure you test every slice before going to the next one. See you in the morning with the completed work. I can just come and configure all the environment variables.
Aakash: And for people who don’t know environment variables, these are basically like secret API keys.
Ankit: Yes, correct. So you cannot just give your secrets away in this chat. So Claude is going to tell you how to give you that. So if someone does not know about the .env file, my recommendation is just ask Claude. It is going to tell you what it is.
Because I have approached this project almost as a layman, so that I can teach it out there. But understand that I tried to make myself a layman, but I had the experience in software development already, so I had to put a lot of work over there.
Waking up to a finished build (44:24)
Ankit: So after this it worked for a few hours maybe, and then after some hours it gave me a message. Templates were created, tools failed, everything is happening, retry, retry, retry, nine send commands, commands nine done, now writing this.
And then I asked it, now this is important. The biggest problem with people who are not able to try AI is that they are generally anxious. What is AI going to do? How am I going to deploy it? How am I going to manage this?
So I asked it that once everything is done, and this I did before I sleep, once everything is done, create a detailed document for me at deploy-checklist.md with all the instructions to deploy this on production AWS, with step-by-step instructions. So I can just read it, and I can deploy this.
After that it has done everything, and then this is the message. Good morning, all slices are implemented, tested and committed. The local stack is running, and everything.
So now it has done everything. I woke up in the morning. I tested everything. Now everything was done locally. So I asked it that I have installed AWS MCP, I have given my credentials, why don’t you deploy this live? And then it has done multiple things, took 15 or 20 minutes, asked me to put all the credentials in the respective file.
And then, as a layman, I did not know how to get these credentials. So the only word that I knew was that I need to use AWS, Amazon Web Services, which is where I’m going to deploy. For everything else, I became like a person who does not know anything. So it gave me all the instructions step by step. Tell me step by step how do I get these credentials. And then it asked me to log in there and put all the information there.
After that I copy pasted the information that I told, and it was able to give me the information. And after this, once this was working and live, then what I did was.
Auditing the cost the AI proposed (46:19)
Ankit: So now I asked it what my costs were going to be. So it also told me that yes, for your new stack, this is going to be your cost.
So now this is where you need to pay some attention, that already I was spending $550 and someone else was managing the pain for me. Now it came up with a stack which is almost $420. So now this is where you need to pay some attention and understand if this is actually a correct decision.
Because I knew already about AWS, I understood that some decisions that Claude has taken are not appropriate. For example, you do not need this large machine for this kind of use case. So I asked it to update this. Can’t we do with a smaller EC2? Also what’s the purpose of Fargate? We can do it within EC2 as well.
And after that it has given me a new stack at $110, and then I was able to deploy it.
And after that, understand, because this is software, there are going to be some errors. Some errors came, it was able to correct it, and then I was able to test, and eventually we were able to build this.
What the finished product looks like (47:14)
Ankit: And once we have built this, this is a better interface than the product that we were using, not only in terms of how it looks but also how it functions. So we have already exported almost all of our leads to this database now, and we have tested a few campaigns as well, and we have flushed some data as well.
Now you can see it has almost all the functionalities, which is leads, I can create different kinds of segments, the templates, the campaigns, the automations like n8n and Zapier, and reports and documentation and settings and everything.
A very important part is the documentation. So after I have built the product, I also asked Claude that let’s say if my team wants to operate this, can you understand all the product once again and create the manual for my team? I can just hand it over to my team, and the team should be able to understand how to operate this.
Similarly, I have created a documentation for the technical guide. Let’s say tomorrow I want to hire an engineer who’s going to dedicatedly look at the same. I can just give them the documentation, which mentions all the decisions, all the tech and everything. Even if I want to port it from maybe Claude Code to Codex, I can just give it this information in the codebase and then it should be able to understand.
So now this is what we have built. I did not want to show you how do you use AWS, or how do you use Claude Code, or how do you use prompts, these kinds of things. The purpose of this whole exercise was to prove that even if you know how to give just the instructions in the common way, as you talk to the developers or talk to maybe even non-technical people out there, if you talk to Claude in that way or Codex in that way, you should be able to create some very good applications.
Just one thing that I have done is some security testing before I could trust what Claude has built. But in your case, if you’re building something for production for your customers, make sure that you are having some clauses for protecting your data, and that you are also having a final green flag from your engineering team before you deploy this onto production.
So I hope this whole setup is going to give a lot of confidence to people, that even if you do not know anything about it, you are only maybe a few prompts away, or you are away by a few instructions that you want to give it to Claude. And then Claude or Codex, it is going to surprise you by the output that it is going to give you, as I was surprised by doing all of these things.
Where the real work now sits, the last 20% (49:35)
Aakash: And I think the most critical thing is that after you got this initial, what I’d call really a prototype, there was a lot of bug fixing, going through steps, finding errors. That last 20%, that’s where the real work is coming now, because Claude can take you 80% of the way there.
And it basically did most of that overnight while you were sleeping, but you need to apply the last 20%. And in many companies, the PM won’t be the one doing that last 20%, right? The PM will create this type of prototype, because as we’ve shown it’s pretty easy to create. Then an engineer or developer will pick it up, and that’s really the role split I’ve been seeing most commonly. Is that what you’ve been seeing as well?
Ankit: Yes, correct. So I think that’s a very good insight, that the last few percent, it could be 10 or it could be 20%, has become more critical now. But understand, the last 20% you only have to think about when you know that you are building something which is substantial, where the 80% is more important.
So many people, what they used to do was they would spend a lot of time in order to get something out in the market, just to find that maybe it will not work. But now you can build something maybe with some rough edges here and there, and then you can show it to people, and then you’ll get some kind of feedback that will help you iterate faster in the market. Because I don’t think there is any better moat than the speed of your execution.
Does every PM need this skill set (51:02)
Aakash: And so does every PM need to learn this skill set, or are there certain types of PMs where this doesn’t apply?
Ankit: I look at the use cases of AI in a PM’s life in three ways.
Either you can make your team more effective, so internal use cases, optimizing the workflows for them. For example, we were working with a company called Kissht, which is IPO bound in India, and ClearTax and a couple of other fintechs in India. We were able to understand that when a fintech gives a loan to someone, their salespeople or their underwriting team has to understand their CIBIL score and do a lot of manual checks in order to get the loan approved, and it used to take maybe four or five hours from every individual.
Now, as a PM who understands that bottleneck and understands how AI operates, you can make your internal team more efficient. So now they have built some kind of small tools where AI can help them with maybe 80% of the work. Now if they used to take three or four hours in order to do credit underwriting, maybe they are going to take maybe 20 or 30 minutes for one file out there.
So that is the first use case, which is if you know about executing things, or you know about the potential of AI and maybe using some of the tools, you can first give a leverage to your internal team, which is a lot of productivity for your team and achieving the outcomes.
Second is you can do a lot of things for you. You can increase your own productivity by doing better research, running multiple simulations. So I believe that simulation is a very important thing for a product manager to do, which is you running your decisions across multiple simulations even before sending it to the users, so that you’re able to understand what is going to happen, what the second order effects of this particular decision are going to be. So that is the second part. It can make you a lot more productive.
And the third part, which is the ultimate part, and I do not think every product manager has a use case for the same, is how do you utilize AI in order to serve your customers better?
Now, there are two branches to this. One is you are building an AI native product, like a Cursor or a Gamma or a Granola of the world, where you understand how the LLMs work. You are going to build products on top of the same, and then your engineers are going to help you out.
The second way is that you can do some common things for your users. For example, every company in the world, I believe that they can have a customer support chatbot which is powered by AI. Or you can have a RAG based model where you can make sure that yes, every question that the customer is asking, they are being answered from your grounded database out there, like maybe an Amazon Rufus. Similarly, there are other use cases that you can implement if you are someone who knows how AI works and you are able to leverage that.
So I do not think it is about whether you should or you should not learn AI. If you are able to understand AI as a tool, I’m sure if you learn this tool well, and it does not take a lot of time. We teach everything, whatever I speak and all the tools that we have mentioned. We promise that you’ll be able to learn these tools just within four weeks if you just spend the weekends. It is not very difficult. The main promise of AI is that it has made all of these things democratized and available for everyone. So you just have to put in some effort, and I’m sure you should be able to unlock that next level of power.
So the question is not whether you should learn it or not. The question is whether you are able to spend some amount of your time and effort in order to get maybe an infinite leverage for the future.
The roadmap, step zero is align (54:18)
Aakash: So Ankit, people now understand how to build an AI product. They understand the power framework. They understand the opportunity. What’s the roadmap to becoming a product builder and landing a product builder job?
Ankit: Now I’m going to give you a set of steps that I have seen work with almost all of my students. And very recently someone was able to get a job at the frontier AI lab in India out there called Sarvam AI. They were coming from a very traditional background, they were able to follow this framework and get that exact job of a senior product manager.
Now I’m going to give you a step by step roadmap that I have seen working with a lot of my students and a lot of successful AI PMs out there. Now, this is step by step, but it does not mean that it is easy to implement. But if you really want that pie share from the AI opportunity, then you should better follow this.
So I’ll start with step number zero. Step number zero is align. Align means if you do not know what you are getting yourself into, if you’re not aligned with the role, you’ll not be able to get it. So by alignment I mean go and look at about 25 to 30 job listings for either AI PM roles, or look at job listings for PM in AI native companies.
This should give you the leverage. And also understand one more part, that even the traditional companies such as banks and the fintech organizations and health tech organizations, they are also having the PM roles where they need people to learn about AI. So make a list of about 25 to 30 companies. Understand what kind of people they want to hire. Look at their job descriptions very carefully.
After this you need to understand and create a skills map. Now don’t use AI for this. Look at the job description very carefully. This is very important. This is where humans have a leverage right now. Understand this carefully. Create the skill map and understand where you can add the leverage.
For example, if you are already a business analyst, or if you are a product owner or something, then you already know how to do the stakeholder management. You are aware about the software development life cycle because you are part of the same. This is your leverage. Other skills you have to also learn.
Second is, if you are a marketing person or a salesperson who is looking to move into an AI PM or PM job, your skills are understanding the customers better. You understand the distribution channels, or maybe you also understand the domain. So that is your advantage, your domain or your customer empathy skills.
Similarly, if you are an engineer, you know about solutions really well, you know about technology really well. You can brainstorm well, plus you also know about the SDLC.
So map down what you already know, and then try to understand what can become your strength. There is almost no one in this world who has at least a couple of years’ experience and does not have any strength. In two years at least you would have built something. So this is the first part, which is align yourself and gain some confidence.
Step one, acquire the skills (57:24)
Ankit: Step one is, I would say, acquire the skills.
Now for acquiring the skills, I generally ask people to list the skills that you have found from step number zero. Go to ChatGPT, try to understand the brief of the skills. Watch some YouTube videos so that you are able to acquire some information.
But understanding only information is not enough. Information actually gives you a false sense of confidence that is shattered the moment that you enter the interview. So you don’t need information. You actually need knowledge.
But how do you convert this information into knowledge? The first part is build something, which is on my YouTube channel, on Aakash’s YouTube channel. You should be able to find a lot of videos where you should be able to learn these exact tools in order to build something. And in this video also, in between, I have shown you how to build a product. So implement that particular framework.
The second part is talk to people who are already in that role. So go to LinkedIn, search for AI product managers on the search page, click on the filter of people. You should be able to find a lot of people. Click on their profile and send them this message. If you want to take an InMail, if you want to take a premium LinkedIn, go ahead and take that, and send this exact message, which is, hey, this is who I am and I’m willing to learn about AI product management, and I find your profile really inspiring, if you really found it. And then ask them that I want to maybe chat with you for maybe 10 minutes. I am also willing to compensate for your time and it is going to really help me.
When people understand that you are respecting their time in order to compensate them, most of the time people will not take money, but they’ll be willing to help you out. Everyone wants to give advice and mentorship to the people who deserve it.
And after you have done this, you will have a good understanding of these skills. And this part alone, learning, building and doing things and reaching out to people, can take about one to four months, depending upon your speed and where you are coming from.
Step two, reach out, and step three, build for them first (59:25)
Ankit: After this is step number two, which is reach out.
Reaching out means whenever you see the job descriptions, make sure that you are not only finding the jobs on LinkedIn. There are other websites as well. At least you should be active on five job listing websites. If you are from India, then Instahyre, Naukri, Indeed, Wellfound and LinkedIn are your places out there. Reach out to multiple companies.
And if you really want to take it to the next level, apply step three. So in the reach out, make sure that you’re also adding the portfolio, the things that you have built, because people don’t owe you trust. You have to make them trust you with the help of the portfolio that you have built.
And now the ultimate step is step three. Create a list of 25 or 30 companies. These are the midsize companies. You are inspired by their work, or you want to work in that domain. Understand those companies well. Ask yourself the question, if I were the product manager, if I were the PM at this company, what would I do in my first six months?
From all the knowledge that you have acquired from step number one and step number two, build a deck, build a prototype, do the research, take the help of AI, and send it to the decision makers of the company, at least two people in a company. You can use a platform such as Apollo or RocketReach in order to get their email IDs. They all have some free credits.
Reach out to them and then follow up at least three times in a week before calling it quits. Initially four or five people are not going to reply to you, not because they are busy or because they don’t want to reply, but your work is going to be of that level.
But as you go ahead and do, after four and five, always remember learning is course correction. As you keep doing and delivering more work and more slides and more prototypes, you’ll be able to understand that you are getting that thing. And when you get that momentum, after your fifth, sixth, seventh reach out, you should be able to get responses from these people. And that is how you are going to build and get into interviews.
And with a few iterations in the interviews, you should also be able to crack the interviews, because when you are going with your portfolio, your work, in an interview, people actually talk in these interviews about your projects, which you should be confident in talking about.
So if you follow these frameworks, I’m again saying this is not easy, this is not simple. I’ve just laid down the steps, that does not mean it is going to be easy. But if you follow this, I’m sure you should be able to get that particular AI PM role, or any role that you deserve.
The three interview rounds to prepare for (1:01:59)
Aakash: And people keep talking about these interviews. I think there’s a lot of misinformation out there. What interview rounds should you be preparing for?
Ankit: So I’ll divide this into three rounds. Three kinds of questions that you should be expecting.
The number one is the fitment part, which is like the most necessary part out there. Even if you are an outstanding candidate out there, but if they do not find the fitment, they’re not going to hire you.
The fitment means you should look at the job description very carefully. Understand what shouts in the job description that this candidate is appropriate for that. That could be your domain matter understanding, your past experience, some projects that you have built. So for example, if you are into fintech, if you are already working in fintech and the next company is fintech, then they are more likely to take you because you have the relevant experience.
Maybe the second kind of people are, let’s say I’m working in education tech but I want to work for a fintech company. What do I show? So I can build some projects in my portfolio that will show to that company that yes, I’m actually interested and I actually know about that particular domain. These days, with the help of AI, understanding the domain does not take years. You can work on some projects maybe for a few months. You should be able to gain maybe 60 or 70% of the confidence.
So the first question is your fitment with the company. That is super important. Nothing happens if you are not able to get that fitment.
The second kind of questions are the general product management questions. They are also very important, because people want to understand whether you have that judgment, that product sense, that customer empathy, that problem solving ability as a product manager or not. So there they are going to test your product sense and product design skills by asking normal questions, as in, how would you improve our product? How would you improve your favorite product? How are you going to measure the success of this product out there? For this also I have created a detailed playlist on my channel. You can go through all the kinds of things.
And the third kind of questions, which are most particularly important right now, are the AI related, which is they are going to ask you that let’s say if you have built something with AI, what is the complete end to end process? And they are not going to ask you for the happy cases. They are going to ask you for the edge cases. For example, if you have built a RAG system as a pet project, what are the ways in which it can fail? If I give maybe 100 million documents, how is it going to work? What if I use a different kind of model? How are you going to evaluate that?
So whatever project that you have built, make sure that you are feeling confident on the same, you have tried multiple edge cases, and you are confident enough to answer it.
So first, fitment or the behavioral questions. Second, the product sense questions. And third is going to be your project related questions or your AI related questions. If you do these three things, it is going to take you a few weeks, but I’m sure you’ll be feeling a lot more confident.
And maybe one more tip I would share. Please leverage GPT or Claude in order to prepare and simulate for these interview conversations. That is also going to help you out a lot. Create a project, put the job description over there, mention about your projects, mention your resume, and then practice by speaking, so that you are not just putting everything into that interview round. You are practicing it before.
The biggest mistake people make (1:04:58)
Aakash: What are the biggest mistakes people make as they’re trying to break into a product builder role?
Ankit: I’ll tell you that they only think about the builder part. They always start with the tools.
So let’s say I want to become a builder, so I want to start by learning n8n. So n8n has maybe more than 100 connectors, so I’m confused how to do what. Either I’m going to lose the motivation, or after learning n8n I’m going to jump to some other tool, then to some other tool.
They will get that, I would say, false sense of progress. They will get that false sense that they are making progress by learning the tools one after the other, but eventually they are only learning the tools and techniques, not the right part of product management.
So if you really want to become a product builder, you need to first understand about the product. So work on your fundamentals. Remember the power framework, that the possibilities, opportunities and workflow, which is the problem space, always comes before building a solution or only knowing about the tools. Tools are easy, you can learn them. Right now it is a leverage, but in the long term it is not going to be leverage.
Where to find Ankit (1:06:01)
Aakash: If people enjoyed this video, where should they find you online for more?
Ankit: So two things. You can either find me on LinkedIn, you can just search for Ankit Shukla, I’m sure you’ll be able to find me. You can also go to our YouTube channel. We have a lot of free content available to you that you’ll seldom get in the paid courses out there.
And the third part is that we have recently launched our full-fledged gen AI program for the busy professionals. It is particularly for people who do not have a lot of time to learn about gen AI and still want to leverage that in order to stay relevant, in order to get ahead in their career. That’s a very advanced four week program where you are going to learn only on the weekends, and every week you’re actually going to build something. So you can also check it out on our website.
Aakash: All right, Ankit, thank you for being so generous with your knowledge. You guys remember that power framework, take a look at his case study, and now go build something yourself. We’ll see you in the next episode.
Ankit: Thank you, Aakash. Take care.