Build an LLM prompt library for PM workflows

Shared Prompts Beat Individual Experiments

Teams using AI for PM workflows typically produce a mess of individual prompts, each re-invented by each PM, each iterating on the same starting point. Anthropic's writing on prompt engineering and Reforge's AI adoption research both document the compounding value of shared prompt libraries: when one PM improves a prompt, the whole team benefits, and quality becomes a team-level metric rather than an individual-level one.

How the Build an LLM prompt library for PM workflows Prompt Works

The prompt structures a prompt library with versioning, quality examples, and governance rituals. The "workflow we'd NOT put in the library" output is the discipline: some PM work is too judgment-heavy for a shared prompt, and forcing libraries into those workflows produces worse results than individual judgment.

When to Use It

• Multiple PMs use AI inconsistently and outputs diverge.
• A new PM team is scaling and needs shared infrastructure.
• Leadership is asking how AI is actually used across PM team.
• A workflow is being re-invented by every new hire.
• A prompt improvement was discovered and needs distribution.

Common Pitfalls

• No versioning. Prompts change, and without version tracking, you can't debug regressions.
• No quality examples. Users can't tell good output from bad without examples. Include both.
• Over-libraryfying. High-judgment work (strategic prioritization) does not benefit from shared prompts.

Sources

• Anthropic Research — Anthropic
• AI Adoption in Product Orgs — Reforge
• The Product Engineer Role — PostHog
• PostHog Blog — PostHog