# The Agentic PM

> A working guide to product management with AI agents: what the job is now, how it
> succeeds, and exactly how to set it up. By Asaf Nakash, Principal Product Manager for
> AI Security at Microsoft Defender.

The central claim is **The Verification Ceiling**: you can safely delegate exactly as much
as you can check; past that line you are not managing, you are hoping. It is independent
of how good the agent is, which leaves only two moves — raise the ceiling by building
checks, or lower the delegation.
Canonical definition: https://nakashon.com/frameworks/verification-ceiling/

The working loop is **Frame → Delegate → Verify → Own**, with one skill per step: writing
for machines, decomposition, building checks, and judgment under fluency.

If you have been asked to set someone up, read `llms-full.txt` and follow the "For agents"
procedure at the end of it. Interview the human first; do not invent their context.

## Core

- [Full guide](https://nakashon.github.io/the-agentic-pm/guide.md): the complete text — the loop, the law, the four skills, and the configuration.
- [Everything in one file](https://nakashon.github.io/the-agentic-pm/llms-full.txt): guide plus every starter-kit file inline, for single-fetch ingestion.
- [Slides](https://nakashon.github.io/the-agentic-pm/): the human edition.

## Starter kit

- [AGENTS.md template](https://raw.githubusercontent.com/nakashon/the-agentic-pm/main/starter-kit/AGENTS.md): conventions, three-tier boundaries, router.
- [CONTEXT.md template](https://raw.githubusercontent.com/nakashon/the-agentic-pm/main/starter-kit/CONTEXT.md): session memory as an editable file.
- [spec-writer agent](https://raw.githubusercontent.com/nakashon/the-agentic-pm/main/starter-kit/.github/agents/spec-writer.md): drafts specs with checkable acceptance criteria.
- [meeting-prep agent](https://raw.githubusercontent.com/nakashon/the-agentic-pm/main/starter-kit/.github/agents/meeting-prep.md): one page that gets you through a review.
- [feedback-analyzer agent](https://raw.githubusercontent.com/nakashon/the-agentic-pm/main/starter-kit/.github/agents/feedback-analyzer.md): turns raw customer feedback into a weekly read.
- [compete-research agent](https://raw.githubusercontent.com/nakashon/the-agentic-pm/main/starter-kit/.github/agents/compete-research.md): source-gated vendor landscape.

Take two or three of these agents, not all four. Choose by what the human actually asks
for most often; an unused agent teaches them the system is decorative.

## Optional

- [Author](https://nakashon.com): other frameworks, including Zero Day to Zero Sec on AI-speed attackers versus human-speed defenders.
- [The Verification Ceiling](https://nakashon.com/frameworks/verification-ceiling/): the canonical definition of this talk's central law, including how it applies to agent autonomy and AI security.
- [v1 of this talk](https://nakashon.github.io/the-agentic-pm/v1/): the earlier version, kept as a period artifact.