Read every file in /research/interviews/. For each transcript, extract: - Interviewee role and company context - Top 3 pain points with direct quotes - Current workaround - Feature requests mentioned Then write /research/interview-synthesis.md with: - The 5 most common pain points ranked by frequency - Contradictions or outlier perspectives worth flagging - 3 hypotheses worth testing - What is still unknown
AI for Product Management: the first 5 workflows to start with
The 2026 competence gap is not "do you use ChatGPT." It is whether messy context becomes a saved research synthesis, a ranked Now list, a PRD, or a stakeholder update without rebuilding the prompt every time. Five workflows with inputs, outputs, and copy-paste prompts. Start with research synthesis, then priorities, PRD, handoff, and updates.
Open AI PM OS
Skip building these five workflows from scratch. AI PM OS ships them ready to run.
What is AI for product management?
AI for product management in 2026 means using agents and reusable workflows to turn product context into discovery synthesis, PRDs, specs, and decisions. Competence is a workflow problem, not a chatbot problem. Start with five workflows. An AI PM OS is the packed version of that operating layer.
This page: AI for PM craft
How you research, write, prioritize, hand off, and update with an agent on your files.
Not this page: PM of AI products
Evals, model risk, AI feature scoping, pricing AI. Important, different job — covered by dedicated AI-product guides, deliberately not here.
By George NurijanianLast updated: October 01, 2026
Workflow vs chatbot: why chat does not compound
A chat thread dies when the tab closes. A workflow compounds when it has:
- Named inputs (files, folders, tickets)
- A saved output on disk
- Standing context (CLAUDE.md / AGENTS.md, skills, MCP)
If you re-paste the same company background every session, you do not have an AI PM setup yet — you have a faster notepad. Score that gap in two minutes with the AI PM Setup Scorecard.
The first 5 workflows
Research synthesis is marked Start here because its saved output grounds the PRD workflow. Then priority critique, PRD drafting, build handoff, and stakeholder updates. Open any row for a copy-paste starter prompt.
Read @context/product.md, @notes/, and the open tickets in @tickets.md. Produce /work/priorities-[today].md with: - Now (this sprint): max 3 items, each with the evidence that puts it here - Next (this quarter): up to 5 items - Later / park: everything else, with one line on why it waits - Assumptions that would change the ranking if proven wrong - One question to ask the team before locking the Now list
Read @brief.md and @research/interview-synthesis.md. Draft /work/prd-[feature].md covering: - Problem statement grounded in the evidence - Proposed solution with a clear scope boundary - Success metrics (leading + lagging) - Open questions for engineering and design - What is explicitly out of scope Flag any claim that is not backed by the research file.
Read @work/prd-[feature].md and @context/product.md. Produce: 1. /work/stories-[feature].md — user stories with acceptance criteria, ordered by dependency 2. /prototypes/[feature]-v1.html — a single-page clickable prototype of the primary flow with realistic mock data Ask before inventing UI that the PRD does not describe.
Read @sprint/this-week.md and @decisions/. Write /comms/update-[today].md with: - 5-sentence exec summary (outcome first) - What shipped / what slipped, with owners - Decisions needed this week (max 3) - FAQ for the broader team - Draft release-note language for anything user-facing Match the voice of examples in /comms/examples/ if that folder exists.
Open AI PM OS
Once the map is clear, install AI PM OS so the same jobs run against your product context.
Chat vs agent vs Linear MCP vs AI PM OS
Buyers mix these up. ChatGPT is a conversation. Claude Code is an agent on files. Linear MCP is a tracker connector. An AI PM OS is the craft layer.
| Layer | What it does | What it is not |
|---|---|---|
| Chat (ChatGPT, Claude.ai) | One-off drafts | Standing context, files, MCP |
| Agent (Claude Code, Cursor) | Multi-step work on files | Encoded PM methodology |
| Linear MCP | Where work lives in Linear | An AI PM OS |
| AI PM OS | Skills, workflows, subagents — how PM thinking runs | An issue tracker |
How do I know my AI PM setup is ready?
A strong PM with an ad-hoc setup still stalls: blank chats, no context files, no review loop. The free AI PM Setup Scorecard grades setup completeness across eight dimensions — tooling, reusable skills, context, data access, workflows, quality control, improvement loop, portability — not your fluency as a PM, and not whether your company ships AI products.
You get a 0–100 score, a tier, and up to three weak spots as next steps. Share the result with a link. No signup.
If Tooling or Context is weak, start with How to set up Claude Code as a PM (~30 minutes).
Tool map by workflow
This starter map pairs meeting capture with Granola, document synthesis with NotebookLM, PRD drafting with Claude Code, prototyping with Cursor or v0, and ticket work with Linear. Full testing notes and rankings live on AI tools for product managers — start with a 4-tool stack, then expand.
| Workflow | Primary | Also |
|---|---|---|
| Capture context / meetings | Granola | Otter, Fireflies |
| Research synthesis (your docs) | NotebookLM | Claude Projects |
| PRD / strategy drafting | Claude Code | Claude Projects, ChatPRD |
| Prototype / UI handoff | Cursor, v0 | Lovable, Framer AI |
| Execution / tickets | Linear (+ MCP) | Jira AI |
Claude Code + Cursor for PMs
Claude Code is Anthropic's terminal-based agent for autonomous, multi-step work on files. Cursor is an AI code editor for interactive edits and visual prototypes. Both can use standing context such as CLAUDE.md or project rules, reusable skills, and a context library. For the PM-angled hub around Claude Code, see Claude Code for product managers.
DIY stack vs a packaged AI PM OS
DIY wins when
You need the install, one context file, one MCP connection, and one workflow that leaves an artifact on disk. Use the setup guide plus the five prompts on this page.
Packaged OS wins when
You are stuck in the blank-system phase: a folder and a CLAUDE.md exist, but inventing skills, guided workflows, and review loops from scratch is the real cost. That is what AI PM OS is for (231 PM skills, 11 guided workflows, review subagents; $99 one-time).
Either path: finish one saved artifact this week. Do not collect prompt lists.
What PMs are saying in 2026
Dated signals behind this starter kit, with links to the sources.
“Starting every AI chat from scratch is the fastest way to burn out… The 30% of professionals getting 15+ hours back… are using Skills and Systems inside Claude.”
“A production-ready Claude Code configuration for product managers. Drop these files into your project and Claude Code immediately understands PM work. Includes a CLAUDE.md context file, 6 PM skills, and 4 templates.”
Also circulating: Lenny × Anthropic Head of Product Dianne Penn on “evals are the new PRDs” (2026-07-27) — useful advanced framing once the five starter workflows are habit.
Frequently asked questions
What is AI for product management?
AI for product management means using agents and repeatable workflows to turn product context into discovery synthesis, PRDs, prototypes, stakeholder updates, and decisions. What lasts is a workflow with clear inputs, a saved output, and standing context. A one-off chat thread dies when the tab closes.
Where should a product manager start with AI?
Start with research synthesis. Drop interview transcripts or feedback into Claude Code, NotebookLM, or Claude Projects and save themes, quotes, and uncertainty flags as a file. That file grounds the next PRD. Then move to priorities, build handoff, and stakeholder updates.
Chat vs agent vs MCP vs AI PM OS — what is each for?
Chat (ChatGPT, Claude.ai) is for one-off drafts. An agent (Claude Code, Cursor) does multi-step work on files. MCP gives the agent live access to tools like Linear. An AI PM OS is the craft layer — skills, workflows, and review — so the agent is not a blank prompt. The layers table on this page maps all four.
How do I know my AI PM setup is ready?
Use the free AI PM Setup Scorecard. It grades setup completeness across eight dimensions — tooling, reusable skills, context, data access, workflows, quality control, improvement loop, and portability — and returns a 0–100 score, a tier, and up to three weak spots as next steps. It scores the workspace, not the PM.
Do product managers need Claude Code, or is ChatGPT enough?
Chat is fine for quick questions. File-based agents win when the job must read a folder, write a file, pull tickets via MCP, and chain steps without re-pasting context. Keeping both works: an agent for workflows and a chat tool for one-offs. See the Claude Code for product managers guide.
DIY stack vs AI PM OS — when does each win?
DIY wins for the install, the first context file, the first MCP connection, and the first workflow that leaves an artifact on disk. A packaged OS wins when skills, guided workflows, and review loops are the gap. AI PM OS ships 231 PM skills and 11 guided workflows for $99 one-time. Either path: finish one saved artifact this week.
Will AI replace product managers?
No. AI compresses synthesis, first drafts, and status writing. It does not replace product sense, prioritization judgment, or cross-functional leadership. The leverage comes from workflows that keep their context and outputs between sessions; a blank chat starts from zero every time.
What is the difference between a prompt list and a PM workflow?
A prompt list is a pile of one-shot instructions. A workflow is a repeatable path with named inputs, a saved output, and standing context (CLAUDE.md, skills, MCP). Blank-chat prompting does not compound. Skill packs and operating systems do, which is why standing context (CLAUDE.md, skills, MCP) beats re-pasting every session.
Is Linear MCP an AI PM OS?
No. Linear MCP connects Claude Code or Codex to Linear so the agent can find, create, and update issues. That is where work lives. An AI PM OS encodes how PM thinking runs: skills, workflows, and review subagents. They complement each other. The full comparison is AI PM OS vs Linear MCP.
Should prioritization be automated with AI?
Prioritization is still under-automated relative to research and PRDs. A useful first step is a priority critique workflow: feed notes and tickets, get a Now/Next/Later draft with assumptions flagged, then decide as a human. Full automation of roadmap calls is not the goal, a better first draft is.