Summary

AI-native product development often behaves less like classic roadmap execution and more like R&D under uncertainty.

Key points

• Effort and impact estimates drift faster in AI-heavy work.
• Stable core engineering and exploratory AI work benefit from different operating rhythms.
• Build-vs-buy assumptions can flip quickly as model capability moves.
• Taste and qualitative QA become strategically important where deterministic testing is incomplete.

Operational relevance

This concept is useful for deciding which initiatives deserve exploratory lanes, short loops, and different review logic instead of normal backlog treatment.

Open questions

• Which current workstreams should be treated as exploration rather than normal roadmap items?
• Where should qualitative QA stay human-led even as the harness gets stronger?

Sources

• wiki/sources/beyond-the-prompt-richard-white.md