Slide 1 · Title
Workshop · Claude Code · Product Leadership

Product Mindset, Shipped.

Product thinking and execution have always been unwritten; we want to codify them so you can scale, accelerate, and focus on what matters.

Claude Code for PMs · April 2026
Amir Fish — AI Builder · Ex-Google & Meta GPM (17 yrs)
Slide 2 · The Problem
Product discipline

What good looks like — and what's in the way.

Left: what any product team wants. Right: what's blocking teams today.

What good looks like
Move fast
No queueing behind other teams' cycles.
Focus on the work, not the admin
Spend time on the problem, not the PowerPoint.
Build high-quality products
With research, framing, and judgment baked in.
Build the right things
Not just what's easiest to build.
What's in the way
Manual workflows eat the week
"Nothing is automated."
Context is scattered everywhere
"Keeping on top of everything — meetings, Slack, Teams, emails."
Redoing work that should already exist
"Recreating Figma takes a lot of time. Requirements aren't clear."
Product thinking doesn't scale past the PM
"I don't have the vocabulary to use an AI agent effectively."
The answer: systems, not sessions. Files, not chat. Tools anyone can invoke — even when the PM isn't in the room.
Slide 3 · The Foundation
Using markdown files — the easiest format for machines to digest

Encoding your org's knowledge.

Foundation · Organizational Knowledge Base
Strategy & decisions
Vision, bets, and a dated ledger of what you chose and why — including review-meeting decisions.
Goals and KPIs
3/6/12-month targets with the numbers that tell you if it's working.
Products
One file per product — capabilities, key users, current state, roadmap.
Personas
Who the users are, what they care about, how they behave.
Research
Raw inputs: interviews, competitive scans, customer signals.
Org structure
Who works with whom, key stakeholders, how decisions actually happen.
Style guides
PRD format, meeting notes, VP review — the quality bar in writing.
my-product/ ├── strategy/ │ ├── vision.md │ ├── bets.md │ └── decisions/ │ ├── 2026-04-q2-review.md │ └── 2026-04-vp-design.md ├── goals-kpis.md ├── products/ │ ├── iPhone-17-Pro-Max.md │ ├── iPhone-17.md │ ├── AirPods-Pro-3.md │ └── Apple-Watch-10.md ├── personas/ │ ├── upgrade-buyer.md │ └── first-time-user.md ├── research/interviews/ ├── style-guides/ │ ├── prd-format.md │ ├── meeting-notes.md │ └── vp-review-format.md └── org-structure.md
Slide 4 · The Ideal
The vision

An AI-based Product Playbook.

An execution layer (skills, agents, workflows) on top of your organizational knowledge base. Invokable by anyone — even teams without a PM.

Skills
On-demand · one job · the PRD lifecycle
/prd-writer  ·  /prd-critique
Drafts a PRD, then stress-tests it like a skeptical PM
/brainstorm
Superpowers brainstorming — stay on course
/collect-feedback
Structures a request for team input
/synthesize-feedback
Clusters responses into themes & conflicts
Agents
Persistent · hold a PM lens
@skeptical-pm
"What would a VP push back on?"
@customer-voice
What would real users actually say?
@data-realist
Prove every claim with real numbers
@design-critic
User-first lens on any mockup
Workflows
Chained · run on cadence
/daily-brief
Outlook + Teams + Jira → one MD
/product-a-to-z
FOR TEAMS
Step-by-step launch path — for any team
/launch-prep
Pre-launch checklist, all systems
/innovations-watch
DEMO
Scans mailing list · flags what's relevant
Foundation · Organizational Knowledge Base
strategy/ · goals-kpis/ · products/ · personas/ · decisions/ · research/ · org-structure/ · style-guides/
The markdown files the execution layer reads from and writes to.
Slide 5 · Anatomy of a Skill
Backup — the demo you're about to run, broken down

What's actually inside a skill?

Four 2-minute interactive demos that lay the groundwork.

A skill, reduced to its parts
PART 1
System Prompt
the skill itself
+
PART 2
User Input
what anyone types
=
RESULT
Output
the shipped artifact
01
Prompt engineering
Zero-shot · few-shot · chain-of-thought — same question, very different answers.
02
Task decomposition
Hard task → ordered sub-tasks → execute each. How agents stay reliable.
03
ReAct (Reason + Act)
Think → act → observe → repeat. The loop Claude Code runs under the hood.
04 · THE PAYOFF
🧩 Anatomy of a skill
Edit a real system prompt live. Pick a preset. Watch output change.
Slide 6 · Use Cases in Action
Files + Layers = Shipped

Now you can actually ship things.

Four patterns, one architecture. Each example = files + a layer + a result you ship.

01
A team without a PM ships a proposal
FILES
launch-guide.md, templates/prd.md
LAYER
/product-a-to-z walks idea → approval → review → launch
SHIP
An eng team clears leadership review on a proposal — without booking a PM's time.
02
Data work
FILES
research/interviews/, clustering-rules.md
LAYER
/cluster-feedback skill
SHIP
200 messy tickets → themes + quotes, in 90 seconds. No pandas, no code.
03
Prototyping
FILES
personas/, design-principles.md
LAYER
/screenshot-to-component skill
SHIP
Figma screenshot → working HTML you can open in a browser. Designers: this is your lane.
04
Workflow automation
FILES
triage-rules.md, team-routing.md
LAYER
/triage-weekly workflow (Jira + Teams)
SHIP
Monday: 30 bugs classified, deduped, routed. The "manual grind" pain — gone.
Slide 7 · Example Output
What the Playbook generates — a concrete example

Three systems → one file.

Example of a workflow output: the scattered-context problem, collapsed into a morning briefing.

SOURCE 1
Outlook
SOURCE 2
Teams
SOURCE 3
Jira
→
ONE FILE
daily-brief.md
Priorities · Blockers · Follow-ups
This is what a /daily-brief workflow produces — built from a few files, runs every morning.
Slide 8 · Bonus Demo
Elaborating on tip #8

Google Stitch — prompt to polished UI.

A bonus demo for the design-to-code folks. Complementary to Claude Code, not a replacement.

Prompt → full UI mockup
Describe a screen in plain English. Stitch generates a real mobile or web UI with realistic data.
Variants on demand
Don't like v1? Generate 5 variations of the same screen. Mix styles, layouts, themes.
Design system capture
Apply a consistent style tokens across every screen. Your brand, not generic Material.
Export to code
HTML + CSS out the other end — drop into Claude Code to wire up the logic.
Google Stitch example — prompt-generated UI mockup
Why this is here: Multiple of you named Figma ↔ code as your bottleneck. Stitch + Claude Code is the fastest path from "idea" to "something to look at."