The robots aren't taking your job.
But a PM who knows how to use them will.
The gap between PMs who get this and PMs who don't is widening fast.
The toolkit, prompts, and workflows here are built around one idea: that the scarce resource in product right now is clear thinking, not engineering capacity. Built to work across whatever tools come next, not locked to any one platform. Start with the toolkit. See what changes.
Everything you need to work with agents like a pro
Most PM resources teach you what AI can do. This one teaches you how to work with it: specifically, how to shape problems clearly enough that agents can act on them, how to feed them the right context, and how to evaluate what comes back.
The ai-pm-toolkit is a free, open source collection of prompts, skills, context templates, workflows, and IDE configs built for the AI-native PM workflow. It's not a course. It's not a framework. It's not a platform that owns your workflow. It's the practical infrastructure that sits inside your editor, works across every major AI tool, and makes every agent session more useful than the last.
What's inside
The most underrated skill in AI-native PM work isn't prompting, it's context. These templates help you capture who you're building for, what's been tried, what good looks like, and what constraints actually matter. So your agent starts from signal, not scratch.
Reusable prompts organised by workflow stage: shaping ambiguous problems, running first-pass prototypes, synthesising user research, evaluating output, making ship-or-kill calls. Every prompt is designed for a specific moment, not general use.
Instruction sets that change how your agent thinks across an entire session. Load the problem-shaping skill and your agent stops accepting vague briefs. Load the evaluation skill and it starts asking whether output actually solves the problem, not just whether it runs.
End-to-end playbooks for the situations PMs hit repeatedly: going from idea to working prototype in an afternoon, converting research into a spec an agent can act on, making a prioritisation call with limited data. Each one references the right prompts and skills at each step.
The prompts and skills above, wired up as slash commands for Claude Code and Cursor. Type /clarify-ambiguity and your agent opens a guided problem-shaping conversation. Type /build-session and it assembles your skill, context, and task into a ready-to-use session brief. No copy-pasting required.
A Python MCP server that exposes the entire toolkit as callable tools inside Claude Code. The build_session tool is the headline: one call assembles your skill, context docs, and prompt into a complete session opener. The infrastructure version of what the slash commands do manually.
Don't want to clone the repo? Run npx ai-pm-toolkit list prompts to browse what's available, npx ai-pm-toolkit search "clarify" to find what you need, and npx ai-pm-toolkit copy prompts/problem-shaping/clarify-ambiguity to put it straight on your clipboard. The full toolkit, accessible in under 60 seconds.
The toolkit works natively with Cursor, Claude Code, and Windsurf today. It will work with whatever comes next. Every prompt is plain text. Every workflow is a markdown file. Every context doc is yours to own, version, and take with you. Multi-tool support is the whole point.
Open source. MIT licensed. Built for PMs who want to rule the robots, not be ruled by them.
The AI tools landscape is moving faster than any market in recent memory. Platforms that exist today will not all exist in two years. Some will be acquired. Some will pivot. Some will be made obsolete by the next generation of models.
The PMs who will be most capable in three years are not going to be the ones who used the best platform today. They will be the ones who spent that time building something portable: a toolkit they own, stored somewhere permanent, designed to work across whatever tools and platforms come next.
That is what this is. Plain text files in a GitHub repository. Prompts that work in any model. Workflows that belong to you, not to a subscription.
Build something you own. Make it better over time. Take it everywhere.
Where are you starting from?
Level 1
You use ChatGPT or Claude but never in a structured way
Start with the no-code prompt cards. Copy one into your AI tool of choice and get something useful in under two minutes. No setup required.
Get the prompt cardsLevel 2
You use AI regularly but every session starts from scratch
Set up a Claude Project once and every future session starts with your product context already loaded. Twenty minutes of setup, permanently better output.
Set up your workspaceLevel 3
You want to use an AI editor but don't know where to start
The first IDE session guide walks you through Cursor from installation to a complete working session. No coding experience needed.
Start your first sessionAlready using Cursor, Claude Code, or Windsurf? The full toolkit is on GitHub.
Thinking About What's Changing
Guides, workflows, and points of view for product managers navigating the AI-native era.
The AI landscape for PMs - what you actually need to know
And why this isn't just a faster way to do the same job.
Read articleChoosing your AI tools as a PM - a practical guide
Why the tools matter less than you think, and one decision that matters more than most PMs realise.
Read article