Agentic PM
Use this section when: You're a product manager who wants to integrate AI agents into your daily workflow — not as a novelty, but as a core tool for speed, consistency, and leverage.
What this is
This section contains practical, copy-pasteable workflows for PMs who use AI agents (Claude, ChatGPT, Gemini, or any LLM) as part of how they work every day.
These aren't theoretical. Each workflow describes:
- When to use it — the specific situation that triggers this workflow
- What to prepare — the inputs the agent needs from you
- The prompt — copy-pasteable, with placeholders you fill in
- What to review — what to check in the agent's output before you trust it
- Common mistakes — what goes wrong and how to avoid it
Workflows
These are the PM jobs-to-be-done where agents provide the most leverage:
| Workflow | Use this when... | Guide |
|---|---|---|
| Standup prep | It's 9:45am and standup starts in 15 minutes | workflows/standup-prep.md |
| Backlog refinement | The backlog needs grooming — stories are stale, scope is unclear, priorities have shifted | workflows/backlog-refinement.md |
| Stakeholder prep | You have a stakeholder meeting tomorrow and need to be ready | workflows/stakeholder-prep.md |
| Research synthesis | You have raw research data (interviews, surveys, support tickets) and need actionable insights | workflows/research-synthesis.md |
| Sprint reporting | The iteration ended and you need to communicate what happened | workflows/sprint-reporting.md |
| PRD drafting | You need to write or refine a product requirements document | workflows/prd-drafting.md |
Patterns
These are reusable mental models for how to work with agents effectively. They apply across all the workflows above:
| Pattern | What it is | Guide |
|---|---|---|
| Agent as Drafter | The agent writes, you review and refine | patterns/agent-as-drafter.md |
| Agent as Analyst | You provide data, the agent finds patterns | patterns/agent-as-analyst.md |
| Agent as Facilitator | The agent runs a ceremony script while you focus on the conversation | patterns/agent-as-facilitator.md |
| Review loops | How to avoid accepting bad first drafts — the single most important skill | patterns/review-loops.md |
Continuous alignment
When agentic systems move to production, alignment becomes an ongoing concern — not a one-time review. The Continuous Alignment Techniques (CAT) framework provides structured methods for instrumentation, evaluation datasets, and offline/online/inline evaluation loops. If you're building agentic systems for clients, start there.
The key principle
Agents multiply your output, not your judgment.
An agent can draft 10 stories in the time it takes you to write 1. But if you don't review those stories carefully, you've just created 10 mediocre stories instead of 1 good one.
The PM's value shifts from writing to directing and reviewing. You become the editor-in-chief, not the staff writer. This is more valuable — but only if you maintain high review standards.
See patterns/review-loops.md for how to build this skill.
Getting started
If you're new to working with AI agents as a PM:
- Start with standup prep — low stakes, high learning. Do it for 3 days and you'll understand the rhythm.
- Then try story writing — use the story writing guide to draft stories for your next iteration.
- Then try research synthesis — upload a set of interview notes and see what the agent surfaces.
- Read review loops — before you go further, internalize the review discipline.
Don't try to use agents for everything at once. Build the skill in one workflow, then expand.
Tool recommendations for PMs
| Tool | Best for | Why |
|---|---|---|
| Claude Projects | Story writing, PRDs, ongoing backlog work | Persistent context across sessions; strong reasoning for acceptance criteria |
| ChatGPT Projects | Same as above; also strong for data analysis | Persistent context; ADA for spreadsheet/data analysis |
| Claude Chat | One-off analysis, quick drafts, stakeholder prep | Fast, no setup needed |
| Gemini + Google Workspace | Analyzing Sheets data, summarizing Docs | Deep Workspace integration |
| NotebookLM | Research synthesis from recordings and documents | Audio-native; good for processing interview recordings |
| Claude Code | Advanced PM workflows (automated backlog analysis, CI integration) | File system access; can read/write project artifacts directly |
See tools/by-tool/ for detailed guides on each tool.