PMs from Google, Oracle & JP Morgan — their unedited check-ins, the essays they're shipping, and the products they're already building. Every word here is theirs.
The cohort doesn't stop at essays. These are live, working products members built and deployed — click through to try them.
Unedited reactions after the live build sessions — on the depth, the hands-on demos, and what finally clicked.
"I got an offer at eGov Foundation. I'll be working under the CTO of Aadhaar. The RAG session in the cohort was a game-changer. In my interviews, I was questioned on a product I hadn't directly owned, but the hands-on RAG work gave me the confidence to speak about it like I built it. I'm super invested — the learnings don't stop here, because I'll be owning AI products in my new role, and everything I learned in the cohort directly applies.
"I recently cleared the interview for a role at Rentickle. One of the questions was to design an AI furniture rental assistant for them — which was exactly what you covered in the Week 2 session. It really helped me in the interview. Thank you, Sir!
"I just landed a Senior Product Manager role in AI and data governance. I used so many learnings from the AI PM Cohort in my interview rounds — at one point I could literally see my interviewers' faces, shocked at how much I knew about AI. All thanks to Shailesh and Apoorva; the cohort learnings carried me through every round.
"The cohort pushed me to ship, not just plan. I built the Culture Welcome Portal for a paying client (4-figure monthly USD), a culture-first employee experience platform any company can brand and deploy in under an hour. It changed how I think about building products.
"My experience has been fantastic. I have not come across any other platform that explains AI fundamentals and also goes into the technical details with hands-on product case studies. Most courses do one or the other. This does both. That is what sets it apart. I would definitely recommend it to anyone looking to build real AI skills.
"I found the first two classes incredibly helpful. I had not touched statistics in a long while and I am not a hands-on technical person, so going back to the basics of where ML actually starts was exactly what I needed. I attended a different cohort before where they went deep into neural networks and transformers — I still do not remember most of it. But what you taught, it stuck. Your push to learn in public made a real difference too.
"I just completed a program from ISB before yours started. If I compare the two, the concepts here are much more hands-on — not just limited to theory. The recommendations, the references, the infrastructure — a lot of things are better here. It is genuinely worth the investment. When you explain them, it is much easier to relate to what we actually do day in and day out.
"Though I had been reading about RAG for a while, this session gave a lot of knowledge in a structural fashion. As a PM, it gave insights on what types of decisions should be taken for what use cases in order to improve the effectiveness of RAG implementation.
"Very insightful. The concepts were explained in a structured way, making it easier to understand. Looking forward to many more in-depth sessions like this.
"RAG session was really insightful. I think it was a lot to absorb in a two-hour class, so there is definitely an action item on us to go back and revisit. Truly appreciate the level of depth — I think that is important to break through into an AI PM role.
"The session was extensive and exhaustive. Appreciate the quality of content and the effort behind putting this together. Some more use cases and real-life examples during the demo would solidify our understanding. This definitely requires multiple runs through the lecture recording.
"RAG session was very insightful and in-depth. The AI Builder session with Claude Code and Vercel is super helpful.
"RAG Demo was super helpful and a very exhaustive build hour.
"RAG session was really helpful, had a lot of depth — would like to have similar depth for the coming sessions.
"The AI Builder session is super helpful. Gave confidence that we can build and deploy projects end to end.
"RAG is explained in a very easy and simple way. So much to revise and recall. Please share some supporting content — thanks a lot for such an insightful session.
"It was a good session — understood the topics. Need to revisit and practice a lot.
"The concepts were explained in a very easy to understand manner.
"This was packed full of information — thank you so much Apurva Mittal and Shailesh Sharma!
Verbatim from cohort feedback and Topmate. Roles and companies from the cohort roster.
Four weeks in, members are shipping and debugging real products — not just reading about them. These are their answers, verbatim.
"I'm definitely more confident. I built PMPrep AI — an AI PM interview coach that gives instant feedback — and it gave me clarity on how to structure AI products, especially around scoring and evaluation. Debugging is where the real shift happened: I test a lot, tweak prompts when they miss, and make sure the scoring rubric does what I expect."
"I'm far more confident now, both using and building AI products. I build end to end now. I've learnt RAG and agentic AI systems, with projects running in both. Debugging is no longer guesswork — with RAG I work through the pipeline in order: is the document parsed cleanly, do the chunks preserve meaning, is the retriever fetching the right chunks, is the re-ranker surfacing the genuinely best ones rather than the closest by embedding distance?"
"After the cohort, I feel extremely confident building AI products. Earlier I was only using Google AI Studio; now I'm comfortable with Claude Code. I built a RAG application from scratch, and at work I built a learning platform for newcomers on my team. I'm confident debugging too, because I now understand the components of an AI product: prompts, context management, guardrails, continuous evaluation."
"I've built a much stronger understanding of how to create AI-native products end to end. I can identify where AI creates value, frame the opportunity, and choose the right architecture. My biggest takeaway is a structured approach to debugging — instead of assuming the model is the problem, I isolate the issue across prompts, context, retrieval quality, embeddings, the RAG pipeline, model selection, latency, evaluation metrics, guardrails and business logic."
"During the cohort I migrated Recruiter OS, my agentic ATS for independent recruiters, from Lovable to Claude Code, with Cursor in the mix, so I could work at a more granular, technical level. In the last two weeks alone I've set up more development infrastructure than I had before — code review, codebase mapping, token management, TDD, new hooks, new skills — and co-written more playbooks than in the previous six months combined."
"I'm a lot more confident than I was. I'm on Codex right now, but I'm keen to switch to Claude going forward, especially after the session. I completely understand the RAG pipeline now. I can question our ML team and hold much better conversations with them about our internal AI product. It isn't a black box any more."
"I'm more confident in building AI products. I'm working with Lovable, Claude and Antigravity now, and when something breaks, I debug it by chatting with the model — sharing screenshots and feeding it more context."
"The program is increasing my confidence in building products with AI — I'm using Claude and Perplexity now. Through the course, the high-level architecture of AI products became clear to me, including how the layers are designed to improve the cost and latency of AI features."
"I'm confident with the RAG pipeline, and I've started exploring Claude. I've also started analysing the RAG architecture behind my organisation's pharmacy customer support tool."
"I feel much more confident than before. Whenever I use a product or app now, I automatically start thinking about where AI is being used, why it was added, and whether it's actually solving a user problem — the cohort has helped me think about AI products like a product manager rather than just a user. I've used these tools on a cloud kitchen management system for my college project."
"Yes. I am now getting much more comfortable with Claude Code, Kilo Code, Visual Studio and GitHub."
"After the cohort, I'm building with Claude Design and vector DBs — and I understand AI processes well enough now to spot the errors."
Verbatim from the Week 4 cohort check-in. Tools listed are the ones each member named as their current stack.
Two weeks in, every member answers the same question in writing. These are their answers, verbatim.
"Two weeks ago, AI and ML felt like a black box."
Now I can identify which ML algorithm fits a business problem — logistic regression for churn, clustering for segmentation — and explain why one fits better than another.
"Now I run the analysis myself and actually read what it's telling me."
Two weeks ago, a model was something I handed to a data scientist and waited to get a number back from. Now I know which method the situation actually calls for.
"That's the AI opportunity this cohort taught me to see."
I'm building Recruiter OS, an agentic ATS for independent recruiters. I applied the AI Opportunity Canvas to a real validation moment and published it as my first "learning in public" post.
Asked what they'll take back to their day jobs, members named live products and real workflows — in their own words.
"I applied structured prompting to design the workflows for Tailor Talk, an AI chat feature on my platform — automating query flows that earlier needed manual handling."
"Building a RAG-based chatbot for customer support queries."
"The guardrails for AI products — jailbreak, grounding prompts, reducing bias to cut hallucinations — is what I'm looking forward to applying at work."
"Have a better conversation with Eng on models and their performance — so I'm actively in decision-making rather than leaving it all to Eng."
"Apply logistic regression to analyse and classify call quality. Use decision trees to improve my AI voice bot."
"Tear down the agentic framework in my project to understand its prompting structure and the guardrails around it."
"Where AI should be used vs where it should not be — and applying the confusion matrix in real product development scenarios."
"Thinking in a strategic way about problem statements being thrown at me as a PM. Not every problem needs AI to solve it — ROI matters."
"Sharing strategic insights while solving product sense interview questions — and knowing when it's an LLM use case versus an ML one."
35+ posts & 20 live products published by cohort members so far — every card links straight to the member's own post.
Raw conversations. Real reactions. When a live story breaks, the cohort thinks it through together.
For OpenAI this is a long-term play: shift user behaviour so shopping happens natively on their interface, and they become the discovery layer. Brands pay a platform fee to get listed — but that could take years.
12Discovery is moving from search to LLMs — Nykaa gets to be where users spend time, plus AEO/GEO advantages. OpenAI gets a sticky vertical, deep beauty training data, and a leg up in the enterprise race.
9Amazon is the default search place for product discovery — they won't want to lose that, since a big chunk of their ad revenue depends on people starting their journey there.
Browsing, comparing, the whole cart experience — that stickiness is genuinely hard to replicate inside a chatbot.
Major ecommerce UIs are going to go conversational. Nykaa found a partner early — they'll live as a widget inside ChatGPT, and I'd bet others follow.
My 2 cents: Nykaa likely gets cheap tokens and white-glove support, OpenAI gets a model e-commerce player to showcase. Expect a "how Nykaa benefited" case study later.
Most retailers won't hand over the keys. They'll support discovery but keep the post-discovery journey. For Nykaa, this is really about replacing their own app with ChatGPT.
Beauty is highly conversational, so Nykaa just wants to show up first. For OpenAI, the category is a foot in the door to eventually own e-commerce discovery instead of Amazon or Google.
Ads stay relevant but shift from the attention economy to a trust & outcome economy. SEO becomes AEO — Agent Engine Optimization: catalogs, verified reviews and delivery reliability matter more than creatives. Brands pay for recommendation boosts or cost-per-purchase.
14A lot of ads get rewritten for agentic consumption — encoded for specific cohorts and ICPs. We won't see them, but agents will be filtered toward specific pages and purchase paths.
8If agents act on the website, they won't see ads. So what actually happens to the ad model?
The internet becomes layered — human-facing ads that also carry encoding for agentic ingestion. And there's still a journey: discovery, attention, activation, just redesigned with agents in mind.
The hardest thing to change is human behaviour. Will people really abandon comfort shopping, scrolling, list-building? I think not — but there'll be agentic layering: agent-to-agent, human-to-agent, and agent-to-system.
Ads move from the UX layer to the content layer — like YouTube embedding ads into video. Influence can happen at the model, data, marketplace, tooling and reasoning layers, with sponsored options shown as clearly labeled choices.
Ads won't disappear, they'll change: from an attention economy to an intent economy. Brands will compete to be the AI's recommendation through trust, reviews, accurate data and pricing.
Cursor has the users and the data, but they got blindsided when Anthropic called Claude Code an "educational experiment" and then made it commercial. Their challenge became model and compute capacity. SpaceX has massive compute — some of it leased to Anthropic already — so together they get a dedicated coding product with millions of users, which means instant distribution.
11A few reasons stack up: training data for their own coding model and harness, from developers who love the product and contribute code as they build. Composer is their first fine-tuned coding model — all of it to catch up to Claude Code and Codex. It's also a growth story for their IPO, and absorbing a full team of proven builders who ship fast is gold when talent is at a premium.
9Coding is one of the frontiers where a lot of money will come from — orgs want to remove engineering as a bottleneck to profitability, and coding agents are how they do it. Claude Code has already been a massive growth engine, and even OpenAI is chasing it with Codex. Acquiring Cursor buys a headstart: existing data, know-how and distribution to build on.
With Grok in the mix, a development platform like Cursor could pull in far more users if it defaults to Grok — and the platform itself becomes a way to keep improving the model. That's a real competitive edge against the likes of Claude.
Verbatim from the cohort's strategic-discussion channel. Each thread shows the two most-reacted answers; open a thread to read the rest. Speakers anonymized; reaction counts are illustrative.
Practitioners running the live build sessions — product leaders who ship AI, not just talk about it.
All 26 members of Cohort 01, by category. Open any profile to see where they came from, what they've shipped and what they're saying about the cohort — in their own words.
Educated at IIT Madras, IIT Delhi, IIM Bangalore, IIM Ranchi, XLRI, SP Jain, MDI Gurgaon, NIT, Yale, Duke, UT Austin, Virginia Tech, Symbiosis and more · spanning associate PMs to VPs of Product · united by one goal: building with AI.
Complete the 12 weeks and Topmate emails you a link to claim your certificate — issued in your name, signed by the faculty, and ready to put on your LinkedIn profile. Here's what it looks like:
has completed the AI PM Builder Cohort — 12 weeks of live sessions, build hours and products shipped in public.
The certificate is the paperwork. The 20 live products above are the actual portfolio.
If your team has a learning and development budget, this is exactly what it's for. Copy this email, fill in the four blanks, and send it to your manager.
Hey {manager},
There's a program called the AI PM Builder Cohort that I'd love to enroll in. It's a live, 12-week cohort run on Topmate, alongside product managers from companies like Google, Oracle and JP Morgan who are learning to build with AI.
A few highlights:
I anticipate being able to put my learnings directly into practice during the cohort. After each module, I can share a recap with the team.
The cohort costs {price}. You can review the full curriculum, the instructor's bio and what past members shipped here: {cohort link}
What do you think?
Thanks,
{Your Name}
P.S. Other teams have sent a few people together — it's an efficient way to build shared context on AI that otherwise takes months of back-and-forth. It might make sense for us to send a few folks.
Cohort link: topmate.io/technomanagers/f/ai-product-manager-builder-shailesh-sharma
"The best way to learn AI as a PM is to build with it — in public, alongside people doing the same."
Join the next cohort of PMs building real AI products, shipping essays, and getting hired — together.