one-step-better-ai-pm

name

one-step-better-ai-pm

description

Get one actionable improvement for your AI product based on the latest GenAI PM briefs. Fetch the last 5 days of curated AI PM insights from genaipm.com, analyze the current repo/project, find synergy between trending topics and the user's work, then research the source material and apply a concrete improvement. Use when the user wants to improve their AI product, get coaching on AI PM best practices, apply the latest industry insights to their codebase, or run "/one-step-better-ai-pm". Requires a GenAI PM subscriber email (set GENAIPM_EMAIL env var or provide when prompted).

One Step Better AI PM

Get 1% better at AI product management every day. Pull the latest curated insights from GenAI PM, find what applies to the current project, and apply one concrete improvement.

Prerequisites

• GenAI PM subscription (free at https://genaipm.com)
• Subscriber email via GENAIPM_EMAIL env var, pm-config.json, or provided when prompted

Workflow

Phase 1: Fetch the Latest Briefs

1. Get subscriber email: check pm-config.json → genaipm_email field, then GENAIPM_EMAIL env var, then ask the user
2. Fetch briefs:

   WebFetch https://genaipm.com/api/feed/latest?email=<email>
   

3. Parse the JSON response — data array contains up to 5 entries, each with date, title, and content (full HTML)
4. Extract key insights across all briefs:
   • New AI capabilities, model releases, API changes
   • Developer tools, frameworks, libraries
   • Real-world implementation patterns and case studies
   • Claude Code, Cursor, and AI coding assistant tips
   • Product management frameworks, methodologies, processes
   • Infrastructure, deployment, and DevOps patterns

If the API returns a 401, tell the user to subscribe at https://genaipm.com and set their email.

Phase 2: Build a Repo Profile

Create a structured summary of the project across 4 dimensions. This profile drives relevancy matching in Phase 3.

Step 1: Read universal discovery files (check each, skip if missing):

• README.md, CLAUDE.md, .cursorrules, .cursor/rules — project description and conventions
• package.json, pyproject.toml, requirements.txt, Cargo.toml, go.mod — dependencies and stack
• docs/ directory listing — look for product briefs, architecture docs, or design docs and read them
• .claude/settings.json, .claude/hooks.json, .claude/skills/ — AI assistant setup

Step 2: Scan the codebase structure:

• List top-level directories to understand project layout
• Grep for AI/LLM SDK imports (openai, anthropic, langchain, langgraph, google.generativeai, xai, cohere, replicate, huggingface, etc.)
• Grep for API keys/env vars referencing AI services
• Identify the main entry points and core business logic files

Step 3: Summarize into 4 dimensions:

1. Product/Business — What does this product do? Who is it for? What problem does it solve?
2. AI/ML Usage — Which AI models, APIs, and providers are used? What does the AI do?
3. Technology Stack — Languages, frameworks, databases, hosting, key libraries.
4. Dev Tooling — CI/CD, testing, linting, AI coding tools, hooks, skills, MCP servers.

Write this summary internally before proceeding — it's the lens for matching briefs.

Step 4: Read .one-step-better/history.json if it exists — skip previously applied improvements.

Phase 3: Match, Rank, and Present (Approval Gate)

Do NOT proceed to Phase 4 without explicit user approval.

Step 1: Score each brief item against the repo profile.

For every distinct insight in the briefs, score it on these criteria (highest priority first):

1. Core product relevance — Does this directly relate to what the product does?
2. AI/ML pipeline relevance — Does this improve or optimize the AI capabilities?
3. Technology stack relevance — Does this relate to the specific frameworks in use?
4. Dev tooling relevance — Does this improve the development workflow?

Items matching criteria 1-2 should always rank above items matching only 3-4.

Step 2: Present the top matches.

1. "Your repo profile:" — Show the 4-dimension summary
2. "From the latest GenAI PM briefs:" — List 2-3 highest-scoring items with relevance explanation
3. "Recommended improvement:" — What to do, which files affected, expected benefit, estimated time
4. Ask: "Want me to research this and apply it?"

Wait for user approval.

Phase 4: Deep Research & Apply

Once approved:

1. Research the source — Extract URLs from the brief, use WebFetch to read original articles/docs
2. Apply the improvement — Make the concrete change
3. Explain what changed — Summarize files modified, why, and how it helps

Phase 5: Track Progress

1. Create .one-step-better/history.json if it doesn't exist
2. Append an entry with date, brief info, improvement description, and files changed
3. Report total improvements applied

Daily Brief Integration

When called from the daily brief engine (via --brief-mode), run a lightweight version:

• Skip Phase 2 (use cached repo profile from .one-step-better/profile.json)
• Skip Phase 4 (don't apply, just recommend)
• Return a single JSON object: {"insight": "...", "action": "...", "source_url": "..."}
• The brief engine includes this as the "💡 AI PM Insight" section

Guidelines

• Always wait for approval before making changes
• Skip improvements already in history
• Prioritize core product improvements over dev tooling
• If no briefs are relevant, say so honestly