Back to index
OtherFour-week live sprint with 8 interactive sessions and 3 office hours a week, plus lifetime access to recordings·Not shown on the page between cohorts; check the enrolment screen when a cohort opens

Prototype to Production: The AI PM Playbook (Aman Khan)

4.3

The most hands-on of the popular AI PM cohorts. You build agents with coding tools and evaluate them yourself, which is the right way to learn this job. Just remember the instructor runs product at an evals company, and the course's view of the world reflects that.

What We Liked

  • You actually build things: agents with Cursor or Claude Code and LangGraph or CrewAI, not slide decks about agents
  • Evaluation is treated as a core PM skill, with metrics beyond accuracy and evals run on your own prototype
  • Aman Khan has real range, with product roles at Spotify, Cruise, Zipline and Apple before Arize
  • 4.8 out of 5 from 60 reviews is excellent for a course this practical
  • Engineers and AI PMs are on hand to help debug, which saves the course for non-technical students

What Could Be Better

  • Khan is head of product at Arize AI, which sells LLM observability and eval tooling, so expect the evals framing to lean toward that worldview
  • No price or cohort is shown between runs, which makes it hard to plan or budget
  • Eight sessions plus three office hours a week over four weeks is a heavy load next to a full-time job
  • Assumes you are comfortable installing developer tools; PMs who have never opened a terminal will struggle early

Detailed review

Most AI product management courses are about AI. This one makes you do the work. Aman Khan's sprint is four weeks long, and in that time you build and deploy agents for your own workflows using Cursor or Claude Code alongside frameworks like LangGraph or CrewAI, define evaluation metrics that go beyond simple accuracy or toxicity checks, run those evals on your prototype, learn to use coding agents well enough to collaborate with AI engineers, and finish with an H1 and H2 AI roadmap and a strategy document you could put in front of leadership. The last section on going from prototype to production, with real user data and implementation plans, is where many courses go quiet, so I was glad to see it included.

I think this is the right shape for an AI PM course in 2026. The PMs who are doing well in AI teams are the ones who can open Claude Code, get a rough agent working, and then look at the traces and say why it fails. Talking about AI strategy without that hands-on feel produces roadmaps that engineers quietly ignore. Khan's background helps here.

Before becoming head of product at Arize AI he held product roles at Spotify, Cruise, Zipline and Apple, and he taught a deeplearning.ai short course on evaluating AI agents. The 4.8 rating from 60 reviews backs up the impression that students come away with real skills. There are two things to weigh. The first is the conflict of interest.

Arize sells observability and evaluation tooling for LLM applications, and its head of product teaching that evals are central to the PM job is not a neutral position. I happen to agree that evals matter a great deal, and nothing on the course page suggests it pushes Arize specifically, but you should notice the alignment. The second is practical. Eight live sessions and three office hours a week is a lot of calendar, and the page did not show a price or an open cohort when I checked, so you cannot easily plan around it.

Courses like this tend to run a few times a year, and Maven usually shows the price once enrolment opens. There is also a real setup hurdle if you have never used developer tools. The helpers are there to debug with you, but expect the first week to be frustrating if a terminal is new to you. For a PM who wants to genuinely understand how AI products are built rather than just talk about them, this is the course I would recommend first.

[ final ]

The verdict.

If you want to be the kind of AI PM who can build a working prototype and argue about eval results with your engineers, this is the course I would pick over the bigger bootcamps. Clear four weeks of evenings first, and keep in mind whose tooling the instructor sells.