Start your learning journeyAI Playbook
Log inStart learning
Role playbook

Product management

Synthesize research with an evidence trail, write specs a coding agent can build from, prototype before you commit engineers, design experiments that can fail, and get feedback triage and competitor watch done before Monday.

Checked 13 Oct 202621 min read16 prompts1 skill1 hook2 routines

For Product managers, product owners and group PMs at software companies

Turn research and feedback into specs a coding agent can build

  • Interview synthesis in about an hour per round, with every theme linked to verbatim quotes
  • Specs with testable acceptance criteria that a coding agent builds without a round of questions
  • A clickable prototype in front of five users in the same week as the idea
  • Experiment briefs with one primary metric, guardrails and a sample size before launch
  • Feedback triage and a competitor digest waiting for you every Monday morning
Works withClaudeChatGPTGeminiMicrosoft CopilotClaude CodeLinearJiraNotionAmplitudeFigma Make
On this page
  1. Where AI helps in this job
  2. Start in 15 minutes
  3. The tools people in this role use
  4. Members only: Workflow: interview synthesis with an evidence trail
  5. Members only: Workflow: a spec a coding agent can execute
  6. Members only: Workflow: prototype before you spec
  7. Members only: Workflow: experiment brief and readout
  8. Members only: A skill for product managers
  9. Members only: Guardrails and hooks
  10. Members only: Routines that run without you
  11. Members only: Prompt ladder for product managers
  12. Members only: Measure it
  13. Members only: Risks and what not to automate

Most PM weeks repeat one loop: read what customers said, decide what to build, write it down so engineers and coding agents can build it, and check whether it worked. AI is fast at the reading, the counting and the first drafts. You own the decisions, the evidence and the promises.

Where AI helps in this job#

TaskWhat AI does wellWhat you still own
Interview and survey synthesisCodes 10 to 30 transcripts into themes with counts and verbatim quotes in minutesWhich themes matter, and checking quotes against the source
Feedback triageTags hundreds of tickets, reviews and NPS comments against your taxonomyThe taxonomy, and the call on what moves the roadmap
Specs and acceptance criteriaDrafts user stories, states, edge cases and Given/When/Then criteria from your notesScope, trade-offs and what you leave out
PrototypesBuilds a clickable app from a prompt and a few screenshotsWhich question the prototype answers, and talking to users
Competitor watchReads changelogs, pricing pages and release notes every week and lists what changedWhether a change matters to your strategy
Experiment designDrafts hypotheses, metric definitions, guardrails and sample-size mathThe decision rule, and stopping a broken test
Metrics reviewExplains a dashboard export, spots anomalies, drafts the weekly narrativeWhether the number is right, and what you do about it
Roadmap narrativesTurns a list of bets into a one-page story for execs, sales or customersThe bets themselves and the dates you promise

AI is weak at the parts of the job that make a PM worth hiring: knowing which customer to believe, saying no, and noticing what nobody wrote down. It counts mentions, not revenue, so a loud segment looks like a trend. It also invents plausible quotes and numbers when the source is thin. Treat every quote and figure in a synthesis as unverified until you have opened the transcript or the dashboard it came from.

Start in 15 minutes#

These work in Claude, ChatGPT, Gemini or Microsoft Copilot (formerly Microsoft 365 Copilot). Paste text or attach files.

Prompt

Themes from a round of interviews

Level 2Any assistant
Here are notes or transcripts from [NUMBER] customer interviews about [TOPIC]. Each one starts with a participant ID such as P01.

[PASTE OR ATTACH]

Find the five strongest themes. For each theme give:
- A one-line name and a two-sentence summary.
- How many participants raised it, with their IDs.
- Two verbatim quotes, each with its participant ID. Copy quotes exactly. Never paraphrase inside quotation marks.
- Any participant who said the opposite.

End with the questions these interviews did not answer.

Good output names participant IDs for every count. Pick three quotes and search the transcripts for them. If one is not there word for word, treat the whole synthesis as a draft.

Prompt

Acceptance criteria for a feature idea

Level 1Any assistant
Turn this feature idea into user stories and acceptance criteria: [DESCRIBE THE FEATURE, WHO IT IS FOR AND WHY]. Write each criterion as Given, When, Then. Cover the empty, loading, error and no-permission states, and list what is out of scope.

Test each criterion by asking whether a tester could fail it without asking you a question. Rewrite any that use words like fast, simple or intuitive.

Prompt

Roadmap narrative for next quarter

Level 2Any assistant
Here are our bets for next quarter. For each: the problem, the evidence, the metric we expect to move and the rough size.

[PASTE YOUR LIST]

Write a one-page narrative for [AUDIENCE: EXECS, SALES OR CUSTOMERS]: why these bets now, what we are deliberately not doing, how we will know each bet worked, and the biggest risk. Use only the bets and evidence above. Do not add dates, features or numbers that are not in my list.

Check that every sentence maps to a bet in your list. Delete anything new, especially dates.

Warning

Interview transcripts and support tickets contain names, emails and company details. Use the workspace your company approved for customer data, or replace names with participant IDs before you paste. Never paste unreleased financials or board material into a personal chat account.

The tools people in this role use#

  • A general assistant. Claude, ChatGPT, Gemini or Microsoft Copilot for synthesis, writing and analysis. Claude Research (Pro plan and above) and Gemini Deep Research can search the web together with connected Google Workspace content such as Gmail and Drive.
  • Claude Cowork with Anthropic's product-management plugin. Open-source skills including write-spec, synthesize-research, competitive-brief, metrics-review, roadmap-update and stakeholder-update, pre-wired for Linear, Jira and Confluence, Notion, Figma, Amplitude, Pendo, Intercom and Slack.
  • Prototyping tools. Claude artifacts on every Claude plan, Figma Make (fullest on paid Full seats), v0, Lovable and Bolt with free tiers and two-way GitHub sync, and Google AI Studio's Build mode. The prototype workflow below compares them.
  • Coding agents that take a spec. Claude Code (terminal, desktop, cloud sessions, and @claude on GitHub through its GitHub Action), the GitHub Copilot cloud agent (renamed from coding agent in April 2026, on all paid Copilot plans), OpenAI Codex cloud tasks, Cursor cloud agents, Google Jules and Devin. Most of them read an AGENTS.md file in the repository.
  • Spec toolkits. GitHub Spec Kit runs a specify, plan, tasks and implement sequence (/speckit.specify, /speckit.plan, /speckit.tasks, /speckit.implement in its docs; the exact spelling you type depends on your agent) and writes spec.md, plan.md and tasks.md under specs/. Amazon Kiro keeps requirements.md (in EARS notation: WHEN a condition THE SYSTEM SHALL a behavior), design.md and tasks.md per feature.
  • MCP connectors. Official remote servers exist for Linear, Atlassian (Jira and Confluence), Notion, Amplitude, Mixpanel, PostHog, Statsig, Dovetail, Productboard (beta) and Intercom. Linear's and Notion's read and write. Intercom's searches conversations and can add internal notes and edit help articles.

Workflow: interview synthesis with an evidence trail

Run it after each research round (5 to 12 interviews) or once a month for continuous discovery. Inputs: transcripts or notes (text, VTT or DOCX), your research question, and a participant list with IDs and segments but no names. 1. Replace names and emails with ...

Members only below this line

Get the full Product management playbook free

What you get on this page:

  • 4 step-by-step workflows
  • 13 prompts
  • 1 skill
  • 1 hook
  • 2 routines

Free. You also get Stratiflux Weekly: one email a week, unsubscribe anytime.

Also in the members part

  1. Workflow: a spec a coding agent can execute
  2. Workflow: prototype before you spec
  3. Workflow: experiment brief and readout
  4. A skill for product managers
  5. Guardrails and hooks
  6. Routines that run without you
  7. Prompt ladder for product managers
  8. Measure it
  9. Risks and what not to automate

Sources

Checked against these on 13 Oct 2026.

  1. Anthropic knowledge-work-plugins: product-management plugin (opens in a new tab)github.com
  2. Claude Help Center: Publishing, remixing and sharing artifacts (opens in a new tab)support.claude.com
  3. Claude Help Center: Use Research on Claude (opens in a new tab)support.claude.com
  4. GitHub Docs: About Copilot cloud agent (opens in a new tab)docs.github.com
  5. GitHub changelog: Research, plan and code with Copilot cloud agent (2026-04-01) (opens in a new tab)github.blog
  6. GitHub Docs: Adding repository custom instructions (AGENTS.md, CLAUDE.md) (opens in a new tab)docs.github.com
  7. Claude Code docs: GitHub Actions (opens in a new tab)code.claude.com
  8. Claude Code docs: Claude Code on the web (opens in a new tab)code.claude.com
  9. Claude Code docs: Memory (CLAUDE.md and AGENTS.md) (opens in a new tab)code.claude.com
  10. OpenAI Codex docs: AGENTS.md (opens in a new tab)learn.chatgpt.com
  11. AGENTS.md (opens in a new tab)agents.md
  12. GitHub Spec Kit (opens in a new tab)github.com
  13. Kiro docs: Specs (opens in a new tab)kiro.dev
  14. Linear docs: MCP server (opens in a new tab)linear.app
  15. Atlassian Rovo MCP Server (opens in a new tab)github.com
  16. Notion docs: Notion MCP (opens in a new tab)developers.notion.com
  17. Intercom developer docs: MCP server (opens in a new tab)developers.intercom.com
  18. Figma Help: Push from Figma Make to GitHub (opens in a new tab)help.figma.com
  19. v0 docs: GitHub (opens in a new tab)v0.app
  20. Lovable docs: GitHub integration (opens in a new tab)docs.lovable.dev
  21. Google AI for Developers: Build mode in Google AI Studio (opens in a new tab)ai.google.dev
  22. Statsig docs: Power analysis (opens in a new tab)docs.statsig.com
  23. Statsig docs: Sample ratio mismatch (opens in a new tab)docs.statsig.com
  24. GrowthBook docs: CUPED (opens in a new tab)docs.growthbook.io
  25. Eppo docs: Guardrail metrics (opens in a new tab)docs.geteppo.com
  26. Microsoft Learn: Microsoft Copilot requirements (rename note) (opens in a new tab)learn.microsoft.com
  27. Claude Code docs: Hooks reference (opens in a new tab)code.claude.com
  28. Claude Code docs: Routines (opens in a new tab)code.claude.com
  29. Claude Help Center: Schedule recurring tasks in Claude Cowork (opens in a new tab)support.claude.com
  30. Claude Help Center: Use skills in Claude (opens in a new tab)support.claude.com
  31. Claude Help Center: Use connectors (tool permissions) (opens in a new tab)support.claude.com

Related

GuidePrompting, from easy to hardHow the 16 prompts on this page workGuideSkills: teach an agent a procedure once, reuse it everywhereHow the skill on this page worksGuideHooks: guardrails that run every timeHow the hook on this page worksGuideRoutines: scheduled agents that open pull requestsHow the 2 routines on this page work