I build AI products at scale. I share what it takes to be an AI PM. Only practical, real-world advice. No-fluff, no theory.
Start here ➡
Joined November 2014
- Tweets69.7K
- Following3.8K
- Followers29.7K
- Likes38.2K
Here is how you become an AI PM before 2027.
(3 months left in the year. Bookmark this post. You will know more about AI than 90% of PMs)
Why AI gives different answers
↳ AI predicts the next word, which is why the same question can get two different answers and why it sometimes makes things up (hallucinates). Then ask any AI the same question twice and compare.
↳ Course: AI Capabilities and Limitations course (anthropic.skilljar.com/ai-ca…)
How to test it (evals)
↳ An eval is a test for your AI. You write questions with the right answers, run them, and count how many it gets right. Then write 20 test questions for something you use AI for and ask a second AI to mark the answers.
↳ Course: Building with the Claude API, Prompt Evaluation section (anthropic.skilljar.com/claud…)
What this means for AI PMs
↳ I translated these learnings for AI PMs in a free email course. Sign up here to see how this works in the real world (newsletter.justanotherpm.com…)
Building in public #1
With this build, our goal is to:
1. Make a simple system
2. Make it scalable
3. Make it easy to maintain
But ensure, every piece of news is interesting, meaningful, accurate and verfiable, explained in the simplest possible way
Stack:
▪️Brave API: search for news
▪️Apify: search reddit and X
▪️Claude: scoring, ranking, selection of top x items, writing the full edition
▪️WhatsApp and Beehiive: distribution
My family and I struggle to keep up with what is happening around us. We live in Gurgaon.
So I started finding and sharing the most meaningful updates in our family WhatsApp group. Everyone loved it. They shared it with friends, who shared it with their friends.
And now @ShirinTandan and I are starting a local newsletter called "Dear Gurgaon"
Here is the second edition. Full version linked below.
I plan to build this in public, even though I suck at building in public.
Please like, RT and share with others in Gurgaon. Please let me know what you think.
Product managers if you want to stay relevant, there is only one thing you should do. 𝗕𝗲𝗰𝗼𝗺𝗲 𝗔𝗜 𝗡𝗮𝘁𝗶𝘃𝗲. Here is how.
(Before you read, save this post. Then, read till the end. I have something to make 2026 the best year for your career.)
𝗪𝗵𝗮𝘁 𝗱𝗼𝗲𝘀 𝗔𝗜 𝗻𝗮𝘁𝗶𝘃𝗲 𝗺𝗲𝗮𝗻?
Two things. 1) You should know how to use AI to help you solve most problems you face. 2) You should know how to build AI products at scale.
𝗛𝗼𝘄 𝗱𝗼 𝘆𝗼𝘂 𝗯𝗲𝗰𝗼𝗺𝗲 𝗔𝗜 𝗻𝗮𝘁𝗶𝘃𝗲: 𝗕𝗨𝗜𝗟𝗗
1. Pick 1 tool
2. Pick 1 "𝘳𝘦𝘢𝘭" problem
3. Build a solution
4. Keep improving till it's launch ready.
5. Launch
6. Improve
If you manage an AI product at work, understand each step in depth. If you don't work in an AI product team, do the same steps for a passion project.
The goal is simple: Learn what it takes to build world class AI products.
𝗕𝗮𝗱 𝗻𝗲𝘄𝘀: you can't learn without 𝘥𝘰𝘪𝘯𝘨 it.
𝗚𝗼𝗼𝗱 𝗻𝗲𝘄𝘀: every person in the world can 𝘥𝘰 𝘪𝘵 now.
If you don't know where to start, join the 𝗔𝗜 𝗣𝗠 𝗔𝗰𝗰𝗲𝗹𝗲𝗿𝗮𝘁𝗼𝗿.
It is a 6 week live course, starting on 31st October. It is built on 4 pillars, and is the only course that will truly take you from AI-aware to AI-native.
(Don't believe me. Believe the 100+ alumni who loved the course. Video below.)
Still don't believe it? Sign up. See for yourself. Get a 100% refund, if you feel you're not AI native after completing the course.
✦ 𝗣𝗶𝗹𝗹𝗮𝗿 𝟭: 𝗧𝗵𝗲 𝗔𝗜 𝗣𝗠 𝗥𝗼𝗹𝗲
Learn how AI has transformed the role, and what it takes to succeed despite the uncertainty that AI adds to our products.
✦ 𝗣𝗶𝗹𝗹𝗮𝗿 𝟮: 𝗛𝗼𝘄 𝗔𝗜 𝗿𝗲𝗮𝗹𝗹𝘆 𝘄𝗼𝗿𝗸𝘀
Learn how AI works. Understand the fundamentals behind LLMs, RAG, Agents, Context, everything that powers the AI products you use every day. See in action what makes or breaks these products at scale.
✦ 𝗣𝗶𝗹𝗹𝗮𝗿 𝟯: 𝗘𝘃𝗮𝗹𝘀 𝗮𝗻𝗱 𝗮𝘂𝘁𝗼𝗺𝗮𝘁𝗲𝗱 𝘁𝗲𝘀𝘁𝗶𝗻𝗴
Learn how to write evals, understand what it means to "test AI", and how to automate testing using LLMs. PMs who can't do this will be the first ones to be replaced.
✦ 𝗣𝗶𝗹𝗹𝗮𝗿 𝟰: 𝗟𝗮𝘂𝗻𝗰𝗵 𝗿𝗲𝗮𝗱𝗶𝗻𝗲𝘀𝘀
Building prototypes is easy, because it lives on your laptop, has 1 user, and you know how to fix it when it breaks. But making it ready for the world is another beast. Learn what it takes to get there.
In addition, you build 3 AI products live during the course. Because you can't learn BUILDING without actually building.
1. AI powered second brain (week 1)
2. Multi agent system (week 3)
3. Your passion project (Week 4-6)
I am an AI PM. I've built AI products used by millions of users. I promise you this is the only structured way to learn and succeed as an AI PM
Interested? Do this in this order
1. Read full course details (linked below)
2. Still interested, fill the interest form
3. Hear back in 24 hours
Course curriculum, logistics, all details here: docs.google.com/document/d/1…
Interest form: forms.gle/QkWb1M5xnhA1FDVx5
My family and I struggle to keep up with what is happening around us. We live in Gurgaon.
So I started finding and sharing the most meaningful updates in our family WhatsApp group. Everyone loved it. They shared it with friends, who shared it with their friends.
And now @ShirinTandan and I are starting a local newsletter called "Dear Gurgaon"
Here is the second edition. Full version linked below.
I plan to build this in public, even though I suck at building in public.
Please like, RT and share with others in Gurgaon. Please let me know what you think.
Something weird is happening. I am working more hours even when AI is doing most of my work.
If you're in the same boat, read on...
To use AI 𝗲𝗳𝗳𝗲𝗰𝘁𝗶𝘃𝗲𝗹𝘆, you need to 1) clearly define the task 2) identify the tool that is most suited to achieve it.
PMs are good at defining tasks and requirements, so that is not the problem. The problem is two fold:
#1 𝗧𝗼𝗼 𝗺𝗮𝗻𝘆 𝗼𝗽𝘁𝗶𝗼𝗻𝘀
Too many tools claim to be best at any given task. The only way you can select one is by trying all of them, which takes a lot of time. To try a new tool, you need to sign up, on board, add connectors, get a hang of it. Only then you start getting value.
#2 𝗧𝗼𝗼 𝗺𝗮𝗻𝘆 (𝗶𝗿𝗿𝗲𝗹𝗲𝘃𝗮𝗻𝘁) 𝗶𝗱𝗲𝗮𝘀
When you choose a tool to do task A, your builder mindset is already thinking about task B, C, D. So instead of focusing on delivering A with high quality, you paralelly work on A, B, C, D. You assume you can deliver all without impacting your timeline. But, the reality is different: 1) you can't deliver all of them in the same time 2) even if you can, they will be sub-par quality.
That is why, I've simplified my AI stack. It contains one tool = Claude. I use it for everything. I rarely try new tools. I am not saying Claude is the answer. I am saying sticking to one tool, instead of fomo-drooling over every new option, is the way to go.
Secondly, I prioritise quality instead of quantity. I do not start the next task until I deliver the first with high quality. There is no need for 𝙙𝙚𝙡𝙞𝙫𝙚𝙧𝙞𝙣𝙜 𝙢𝙤𝙧𝙚, but there is big need for 𝙙𝙚𝙡𝙞𝙫𝙚𝙧𝙞𝙣𝙜 𝙦𝙪𝙖𝙡𝙞𝙩𝙮.
So I use AI to improve quality, not quantity, of work.
Have you faced the same challenges? How did you solve them? Let me know in comments.
Starting an online community of builders
▪️1 x demo day per month
▪️1 x live buidl day per month
▪️ private slack group to share progress
Interested? Reply "Build" I'll DM you
Why do you read posts from people you don't know? That is what I asked myself while browsing. So I thought of introducing myself.
I am Sid. I've been a PM for 12+ years. I love building. I love decoding why some solutions work, and others don't. I like solving hard problems which force me to learn and push me outside of my comfort zone. I spend a lot of time learning new things, and sharing what I learn.
My current focus and interest lies in 1) staying abreast with what is happening with AI 2) how AI is transforming product management 3) how AI can make my job easier, better, more efficient.
I vibe code products, solutions, systems. I have built 50+ products, had some successes, and many failures. More importantly, I've learned what it takes to build AI products and to build products with AI (both are very different and important to learn)
Over the next 30 days, I will write on the following topics:
• Understanding AI
• Decoding AI Agents
• Tools I use everyday
• Resources to excel as an AI PM
• Lessons from building (and failing)
• Personal stories of what I'm building
• What does it take to excel as an AI PM
• Things I am learning and struggling with
• Tools, guides, resources to stay up to date
What is next: if you're also learning about the same things, please introduce yourself in the comments. I would love to know more about you.
You will not lose your job to AI. You will lose it to the PM sitting next to you who ships four times more than you.
I say this with urgency, not fear.
Most PMs still spend their week on work that stopped being valuable eighteen months ago.
Here are 6 habits to drop, and what to do instead
❌ Writing the PRD from scratch
✅ Explain what you want to an LLM, and get it to create a first draft. Use that as the starting point.
❌ Reading 40 user interviews one at a time.
✅ Convert into a CSV, upload into an LLM, and ask it to categorise into most relevant themes, highlight the most important feedback, find the two users who said something nobody else said. Go talk to them. (That is where you unlock real value)
❌ Waiting two sprints for a prototype.
✅ Learn Claude Code, Codex, etc. Build a working prototype in a few hours. (If you still have never built one yourself, this is the biggest skill gap which will make career growth tough as you go along.)
❌ Created a Data Science ticket for a "deep dive"
✅ Use an LLM to write the query. You will get it wrong the first time. That is part of the learning. In the long run, you will learn how to self-serve every need as a PM. (That will do more for your career than anything else.)
❌ Sitting in a meeting to take notes.
✅ Record it, ask an LLM (or a meeting app) to summarise critical decisions and action items. Read that summary, and spend the hour on work that matters.
❌ Building the deck slide by slide.
✅ Provide data, context, to an LLM, Gamma, etc. Ask it to create slides as per you design methods. Speond time in building the right narrative, driving alignment, because that is where you can add real value/
None of this makes you a better product manager on its own.
It buys you back about a day a week. The PM who wins is the one who spends that day on the things that are mission ciritical.
Everything else is typing.
If I were building a startup today, I’d cheat.
The romantic era of "building everything from scratch" is dead.
Speed is commoditised now. Everyone can ship fast. Quality is the only thing that keeps users around
If you try to focus on both, you will lose to the founder who knows how to leverage infrastructure that already exists.
My rule is simple: If it’s not your "Core Product," don’t build it:
Here is the stack I'm using right now
• Auth: Clerk/Auth0
• Payments: Stripe
• Subscriptions: RevenueCat
• Database: Railway + Postgres
• Caching: Redis
• Search: Brave API
• LLM: Anthropic's API
• Embeddings: Voyage AI
• Analytics: Mixpanel
• Hosting: Railway
Ten roadmaps I don't have to manage. Ten things that will not break at 2am.
I still see teams and builders spending weeks writing their own auth flow. It is messy, it breaks, and users feel it. No user has ever chosen a product because the login screen was hand made.
P.S. If you're not "building" in 2026, you're already behind. Reply with what you're building.
P.P.S. This is NOT a sponsored post.