How to Run an AI PM Study Group That Actually Accelerates Learning
TL;DR
Aspiring AI PMs who learn alone often quit by month three. Aspiring AI PMs who learn in a focused weekly study group of five to seven peers tend to ship two to four real artifacts in the same period. The compounding factors are accountability (showing up matters when others expect you), specialization (different members go deep on different topics and teach each other), and feedback (you cannot self assess your own writing or prototypes). This guide covers how to recruit the right peers, the 90 minute weekly meeting structure that survives, the four exercises that produce the most learning per hour, and the failure modes (vague topics, no homework, no shipping) that kill most groups by week six.
How to Recruit Five to Seven Peers Who Will Actually Show Up
The single biggest predictor of study group success is who you recruit. Most groups fail not because the curriculum was wrong but because the peer mix was wrong. The four traits below matter far more than seniority or current job title.
Trait 1: Track record of finishing things
Recruit peers who have shipped something in the last 12 months: a product, a side project, a written piece, a course completion, anything that took 20 plus hours of disciplined work. People who consistently start and abandon projects will do the same in your study group and demoralize the rest. The simplest filter is asking what is the last thing you finished and how long did it take. The answer is diagnostic.
Tradeoff: Filtering on this trait shrinks the pool. You may pass on enthusiastic peers who lack track records. The math still favors strict filtering: one persistent finisher contributes more than three enthusiastic starters across a 12 week study group. Be selective even if it means starting with three peers and growing to five.
Trait 2: Different but adjacent backgrounds
Five members who all come from the same background (all consumer PMs, all enterprise sales engineers, all data scientists) produce a thin study group because every member sees the same blind spots. A mix of backgrounds (one designer, two PMs from different industries, one data scientist, one engineer in transition) produces better discussions because each member brings different examples and instincts. The aim is adjacent (everyone speaks the language of product) not identical.
Tradeoff: Recruiting across backgrounds is harder than recruiting from your immediate network. It requires reaching into adjacent communities (product Slack groups, design forums, engineering meetups). The investment of 5 to 10 hours upfront in recruiting compounds across the entire life of the group.
Trait 3: Comparable time availability
If half the group can commit 5 hours per week and half can commit 1 hour, the group fractures. The members who do the homework feel held back by the ones who do not. The members who do not feel embarrassed and drop out. Be explicit upfront about expected weekly time commitment (90 minutes for the meeting plus 3 to 5 hours of homework) and recruit only people who can credibly meet it.
Tradeoff: Some excellent peers may not have the time available right now. Be honest with them and yourself rather than recruiting and watching the misalignment surface in week four. Start the group with the people who can show up and invite the busier peers to the next cohort.
Trait 4: Willingness to give and receive direct feedback
A study group where everyone is polite and nobody says this case study has problems produces little growth. The whole point of the group is honest mutual feedback. Recruit peers who can both give and receive critical feedback without flinching. Ask in the recruiting conversation: tell me about a time someone told you your work needed major changes; what happened. The answer reveals whether they will engage substantively or default to platitudes.
Tradeoff: Recruiting for direct feedback culture means screening out conflict averse candidates, even friendly and competent ones. The discomfort pays back: study groups with feedback culture produce members who are interview ready by week 12. Polite groups produce members who feel encouraged but have not improved.
The 90 Minute Weekly Meeting Structure That Holds for 12 Weeks
The most common failure pattern is a meeting that starts as 60 minutes, drifts to 90, then 120, then nobody can attend reliably. The structure below holds at 90 minutes, runs at the same time every week, and produces measurable progress.
Minutes 0 to 10: Status round
Every member shares in 60 to 90 seconds: what I committed to last week, what I shipped, what I struggled with. Strict time limit per person enforced by a timer. The status round creates accountability because nobody wants to repeatedly say I did not get to it. It also surfaces blockers early so the group can help.
Tradeoff: Strict timing feels rigid for the first two weeks and then becomes essential. Without it, the status round eats 40 minutes and there is no time for substantive work. Appoint a rotating timekeeper who is empowered to interrupt, not the same person every week.
Minutes 10 to 40: Topic deep dive
One member presents a 20 minute prepared deep dive on the week's topic (a paper, a system pattern, a teardown of an AI product, a framework). The group asks questions for 10 minutes. Rotate the presenter weekly so every member presents twice in a 12 week cohort. Presenting forces the kind of deep understanding that listening alone never produces.
Tradeoff: Preparing a 20 minute deep dive takes the presenter 4 to 6 hours. This is heavy but it is also the single highest leverage learning activity in the group. Members will resist the load initially; the ones who prepare seriously will improve fastest and become natural leaders within the cohort.
Minutes 40 to 70: Workshop a member's artifact
One member brings a piece of in progress work (a draft case study, a prototype, a PRD, an evaluation harness) and the group spends 30 minutes giving structured feedback. Use a feedback frame: what is working, what is unclear, what would I change, what is one specific next step. Rotate so every member gets workshopped twice per cohort.
Tradeoff: Workshopping is the most uncomfortable part of the meeting and the highest impact. Members will sometimes feel raw afterward. The compensating discipline is to always close the workshop with three concrete next steps, so the member leaves with action rather than just critique.
Minutes 70 to 90: Commitments and topic for next week
Each member commits in writing (in the shared doc) to one specific deliverable for the next week. The group picks the next week's topic and presenter. End on time, every time. Members who stay to chat can do so, but the meeting itself ends at the 90 minute mark to protect everyone's schedules.
Tradeoff: Ending on time is a discipline that pays for itself. Members who know the meeting always ends at 90 minutes will reliably attend. Members who experience meeting drift will start skipping or arriving late, which corrodes the group within four to six weeks.
Four Exercises That Produce More Learning Than Reading
Reading and discussion alone produces shallow understanding. The four exercises below force the kind of active practice that builds real intuition. Pick one or two for each 12 week cohort and do them deeply rather than spreading thin across all four.
Exercise 1: The shared evaluation set
Pick one task (summarization of customer interviews, classification of support tickets, generating PRD outlines). Build a 50 input evaluation set as a group. Each member writes their own prompt and runs it against the same set. Compare results in week four. This single exercise teaches more about prompt engineering and evaluation than 20 hours of reading.
Exercise 2: The rotating teardown
Each week one member tears down a different shipped AI product and presents in 20 minutes. Over a 12 week cohort the group covers 12 products. The cumulative pattern recognition (here is the 8th time we have seen RAG with reranking) builds intuition that no single teardown can produce.
Exercise 3: The pair prototype
Pair members for 4 weeks. Each pair builds a small AI prototype together (one weekend session per week). At the end of week 4, each pair presents what they built and what they learned. Pairing forces both members to articulate their thinking aloud, which surfaces gaps that solo work hides.
Exercise 4: The mock interview ladder
Starting in week 6, the last 30 minutes of each meeting becomes a mock interview. One member is the interviewer (using questions from real AI PM interviews), one is the candidate, and the rest observe and give structured feedback. Rotate weekly so every member does both roles. This exercise alone is worth a six week interview prep course.
Pick a graduation artifact at week one
Decide at the first meeting what every member will have produced by week 12 (a portfolio piece, a passing mock interview score, a working prototype with usage data, a published case study). Without a defined outcome, the group drifts and ends with nobody able to point to what they got out of it. With a defined outcome, every weekly commitment connects to a destination, and the energy compounds rather than dissipates.
Learn AI PM in a Cohort That Holds You Accountable
Cohort based learning, peer feedback, and the structured practice AI PMs need are core to the AI PM Masterclass. Taught live by a Salesforce Sr. Director PM.
The Failure Modes That Kill Most Study Groups by Week Six
The list below is built from observing dozens of AI PM study groups. Most groups die for predictable reasons. Knowing the failure modes upfront lets you design against them.
Failure 1: No homework, just discussion
Groups that meet and only talk about what they read produce minimal skill growth. Without prepared deliverables, members read passively and the discussion stays surface level. Fix: every meeting requires every member to bring something they made (a prompt, a paragraph of analysis, a diagram, a piece of code). Discussion happens around artifacts, not in their absence.
Failure 2: One person carries the group
A common pattern: the most enthusiastic member ends up doing all the prep, all the recruiting, all the timekeeping. They burn out by week six and the group dissolves. Fix: rotate every role (presenter, timekeeper, workshopper, note taker) every week. No member does the same role twice in a row. The load distributes and members develop different skills.
Failure 3: Topics chosen week to week without an arc
Picking topics ad hoc each week means the group covers a random walk of subjects. Members learn fragments and never go deep enough to build real fluency. Fix: pick a 12 week curriculum at week one. The arc can be along skills (week 1 to 4 evaluation, 5 to 8 architecture, 9 to 12 stakeholder management) or along an artifact (build one prototype across 12 weeks).
Failure 4: No shipping deadline
Groups without an external deadline tend to drift. The deadline can be soft (we will publish a group case study by week 12) or hard (we will all submit applications to specific roles by week 10). The deadline creates the urgency that produces real work. Fix: pick the deadline at week one and update the timeline every two weeks against it.
Failure 5: Tolerating chronic absences
When one member misses two meetings in a row, they almost never return. Worse, the rest of the group sees their absence as permission to skip themselves. Fix: have a frank conversation after one absence. After two, the member is out of the cohort with no hard feelings; they can join the next one. Holding this line preserves the group for the members who are committed.