A single pegboard of hand tools on a workshop wall is traced by a pantograph onto a long row of smaller identical pegboards, the whole scene drawn with wide empty margins on every side.

Lenny's Newsletter

How I AI: Grok vs. Dot vs. Muse vs. Duet: which one crushes the competition? + How Figma’s PMs actually use AI every day

Claire Vo with Ezra Mechaber · 2 November 2026

Company story · agents · collaboration · Agent context files

Takeaway

Ezra Mechaber, a product manager at the design software company Figma, helped build 31 shared AI skills that put 60 to 70 product managers on one working setup.

Summary

Claire Vo, host of the podcast "How I AI", interviewed Ezra Mechaber, a growth product manager at Figma, in an episode summarised in Lenny's Newsletter. Mechaber is one of the creators of PM OS, a Git repository of AI skills, instructions and hooks shared across Figma's product organisation. It was designed around tasks that product managers routinely skip, such as involving support before a launch.

Key points

PM OS is an AGENTS.md instruction file, reusable skills and a few command hooks. An automatic version check keeps 60 to 70 product managers aligned, and private subrepositories allow personal changes.
At a company of roughly 2,000 people, one skill maps a team and adds the relevant designer, engineering manager and team leaders as reviewers.
Mechaber gives a browser AI agent a test account and a FigJam board of scenarios, and it found edge cases such as a text box filled with 1,000 emojis.
AutoXP uses three AI agents to find funnel drop-off, propose new copy and prepare the code, and a person reviews each experiment once before launch.

Implication

Mechaber argues that a basic Git repository can support a whole product organisation's use of AI. The source is a newsletter summary of a podcast interview and reports no measured results.

Plays

Product

Keep the product team's AI skills, instruction file and hooks in one shared repository, with an automatic version check so every product manager runs the same setup.

Product

Before a team's bug-testing session, give an AI agent that can operate a web browser a test account and a written board of scenarios, and have it record what it finds on that board.

Derived by Working Surface from the article; more in the Playbook. Source line: The best place to start with AI is the work PMs routinely skip.

Source issue

3 November 2026: teams working with AI agents are moving human review from each output to the plan, the shared instructions and the rules the agent follows beforehand, and naming in advance who answers for the result.