AI Exponential Pm Shifts

Instructions

Use this workflow when model capability is improving quickly and you want your product process to keep up.

1. Plan in short sprints (run side quests)

   • Encourage short, self-directed experiments outside the main roadmap.
   • Treat an afternoon prototype or capability test as a first-class output.

2. Prefer demos and evals over docs

   • Build a rough prototype early to make the conversation concrete.
   • Create lightweight evals to measure whether the feature works and where it fails.

3. Revisit features with new models

   • Be a daily active user and deliberately test tasks you assumed were too hard.
   • If users are stitching together manual workflows, treat that as scaffolding to productize.

4. Do the simple thing that works

   • Avoid clever workarounds that only exist to compensate for current model limits.
   • Favor simple implementations that become easier to upgrade as models improve.

5. Optimize for capability first

   • Don’t cut token usage too early; first confirm the feature is possible and valuable.

Examples

Example: Turn a written spec into a prototype

Use the template below as a starting point:

• templates/spec-to-prototype.md

Example: Create a small eval set for a risky feature

Start with:

• examples/lightweight-eval-starter.md

Example: Daily active testing prompt

“Try doing the thing we assumed was too hard. If it works, suggest how we should productize the manual steps users are doing today.”

Source

• Blog post: https://claude.com/blog/product-management-on-the-ai-exponential