Brought to you by
Mobbin - Discover real-world design inspiration
Get a full year of Mobbin Pro, along with 8 other AI tools, at bundle.aakashg.com.
Today’s episode
LinkedIn shut down its Associate Product Manager program and replaced it with an Associate Product Builder program.
If you’re a regular reader, you know I’ve been writing about this shift for a long time. What’s new is that the market just gave it a full-fledged title.
And a title in a job post comes with 3 things: a JD, interviews, and a comp range.
So I brought back the one guest I knew would bring the best-case data. Ankit Shukla, founder of HelloPM, ran 12,500 PM job descriptions through Claude and GPT to see what top companies hire for. His AI PM roadmap is still my most popular episode ever. His evals episode is #2.
This one goes further than both.
10 Key takeaways
1. The product builder role is real, and the data says so - 12,500 PM job postings analyzed with Claude and GPT. More than 30% now ask for hands-on AI building. The title has not caught up everywhere, but the requirements already have.
2. Judgment is the skill being paid for, not the tools - Across those postings, RAG, agentic AI and prompt engineering were not the number one skill. Identifying the right use case for AI was. The engineering part is what AI can already do for you.
3. The premium is real and it is bigger in India - More than 20% over traditional PM pay globally, with a $195,000 median in the US and senior roles reaching $250,000 to $560,000. Indian AI PM roles carry an even steeper premium over their non-AI counterparts.
4. Start with possibilities, not problems - Traditional product management starts with the problem. AI inverts that, because some problems you never framed as problems at all. You were treating them as harsh realities. Map what AI can do before you map what hurts.
5. Never start with the tool - The biggest takeaway Ankit gives is that POW comes before E. Possibilities, opportunities and workflows first. The moment you open n8n before you have done those three, you are building AI slop with extra steps.
6. The five levels, and why most people stall at level one - Prompting, then reusable Gems and custom GPTs, then skills and connectors, then vibe coding, then production builds with MCPs. Most people live at level zero and one. Level two is where the ROI actually starts.
7. Claude Code versus Codex is a settled debate - Ankit benchmarked both plus Cursor and a DeepSeek harness across e-commerce, social, and gaming builds. All of them shipped functional sites. For 90% of use cases the choice does not matter, so stop optimizing it.
8. The build was 20 minutes of human time and 20 hours of agent time - A production email platform on Node, TypeScript and AWS, built by someone who deliberately avoided the languages he already knew. He pulled the AWS MCP so he never touched the console, then slept while Claude shipped slices.
9. Audit what the agent spends - Claude proposed a $420 a month stack. Ankit knew AWS well enough to ask why it provisioned a machine that large and what Fargate was for. The revised stack came back at $110. The judgment is in the review, not the build.
10. The last 20% is where the job moved - Claude takes you 80% of the way overnight. Bug fixing, edge cases and errors are what remain, and in most companies the PM hands that part to an engineer. The PM's new job is producing the prototype that makes the conversation real.
Please subscribe on YouTube and follow on Apple and Spotify. It helps!
If you want to advertise, please email productgrowthppp[at]gmail[dot]com.