These are all the skills you need to become an AI PM like Aman Khan — who has been an AI PM at Arize (Evals Leader), Spotify, and Cruise (Autonomous Driving).
This was a fantastic post, Aakash. I love how you and Aman went beyond high-level theories and scoped out the real benchmarks to become an AI PM.
Some product folks claim to be AI-driven after using Lovable/Bolt for a few mockups and tools. But I think there's a real skill chasm between traditional PM and AI PMs of the future.
Where traditional rule-based SaaS products demanded clear PRDs and "test case" were conveniently delegated to QAs, it seems intrinsic knowledge of constructing and operationalizing evals will be a core skill that will separate the real AI PM crop from the mere chest-thumpers.
It's one thing to create a clickable prototype.
It's another thing to know what it takes to ship and maintain a production-level AI-powered product that doesn't break (functionally and possibly, ethically) when exposed to the millions of combinations of diverse, user-generated inputs at scale.
Also, IMHO, we'll soon see experts revisiting the good old debate around how technical PMs need to be.
Given Aman recommends Cursor as a home base for PMs to write and test evals beyond prototyping, and creating systems like this requires second and third-order thinking, I feel AI PMs will need a higher baseline of technical chops alongside their usual UX, business, and cross-functional responsibilities. People with CS backgrounds might be able to make the pivot faster.
I'm sensing product interviews, especially for AI-heavy roles, will also start evolving. Real-time prompt engineering will probably be on the menu.
Nailed it. AI PMs have to be more technical. And the interviews are becoming more technical, is what I hear.
As you said: "Where traditional rule-based SaaS products demanded clear PRDs and "test case" were conveniently delegated to QAs, it seems intrinsic knowledge of constructing and operationalizing evals will be a core skill that will separate the real AI PM crop from the mere chest-thumpers"
The result is a much higher technical bar.
People should check out your amazing newsletter for more such insights, thanks for commenting friend!
Really enjoyed this! I’ve used Bolt and Lovable before and was always impressed, but I was honestly a bit intimidated by Cursor until now. I used to think tools like these were just for devs, but the agent workflow demo totally changed my mind. Your breakdown of Bolt and the “why” behind choosing Cursor for the 0-to-1 demo made it all click. I’m already brimming with side project ideas—can’t wait to get my hands on Cursor!
This was a fantastic post, Aakash. I love how you and Aman went beyond high-level theories and scoped out the real benchmarks to become an AI PM.
Some product folks claim to be AI-driven after using Lovable/Bolt for a few mockups and tools. But I think there's a real skill chasm between traditional PM and AI PMs of the future.
Where traditional rule-based SaaS products demanded clear PRDs and "test case" were conveniently delegated to QAs, it seems intrinsic knowledge of constructing and operationalizing evals will be a core skill that will separate the real AI PM crop from the mere chest-thumpers.
It's one thing to create a clickable prototype.
It's another thing to know what it takes to ship and maintain a production-level AI-powered product that doesn't break (functionally and possibly, ethically) when exposed to the millions of combinations of diverse, user-generated inputs at scale.
Also, IMHO, we'll soon see experts revisiting the good old debate around how technical PMs need to be.
Given Aman recommends Cursor as a home base for PMs to write and test evals beyond prototyping, and creating systems like this requires second and third-order thinking, I feel AI PMs will need a higher baseline of technical chops alongside their usual UX, business, and cross-functional responsibilities. People with CS backgrounds might be able to make the pivot faster.
I'm sensing product interviews, especially for AI-heavy roles, will also start evolving. Real-time prompt engineering will probably be on the menu.
Nailed it. AI PMs have to be more technical. And the interviews are becoming more technical, is what I hear.
As you said: "Where traditional rule-based SaaS products demanded clear PRDs and "test case" were conveniently delegated to QAs, it seems intrinsic knowledge of constructing and operationalizing evals will be a core skill that will separate the real AI PM crop from the mere chest-thumpers"
The result is a much higher technical bar.
People should check out your amazing newsletter for more such insights, thanks for commenting friend!
Absolutely love the deep dive we did here - and went an hour over! Honestly awesome to see how much ground was covered. Thank you for having me on!
Favorite moment was definitely when the agent just wasn't working and we debugged it live - and got it up and running 😎
Yes this was all real and LIVE, thanks for doing it!
Really enjoyed this! I’ve used Bolt and Lovable before and was always impressed, but I was honestly a bit intimidated by Cursor until now. I used to think tools like these were just for devs, but the agent workflow demo totally changed my mind. Your breakdown of Bolt and the “why” behind choosing Cursor for the 0-to-1 demo made it all click. I’m already brimming with side project ideas—can’t wait to get my hands on Cursor!
Let's go! The agent makes it just as easy as lovable and bolt, you got this.