Most code is a frozen decision whose why has already evaporated.

I work the other way around. I hold on to the intent, then treat the implementation, trade-offs, and claims around it as things to be re-derived and verified, not assumed.

When only the intent is left standing, the real work is decomposition: pulling buried decisions back into the open, naming the alternatives, separating evidence from guesses, and leaving open questions open on purpose.

The same move applies to people, not just code.

A craftsman's estimate. A domain expert's gut call. A judgment that works, but was never written down.

That tacit knowledge usually lives in one head and dies there. I go to where it lives, sit inside the problem, and turn the implicit into a digital asset that something else can query, version, and verify.

This is close to what a Forward Deployed Engineer does: less "here is a tool," more "let me learn your judgment and encode it."

• Intent is the source of truth — implementation is derived, not sacred.
• Decisions get decomposed — alternatives, non-goals, and assumptions are named.
• Evidence beats claims — nothing is done until it has been exercised.
• Tacit knowledge becomes an asset — judgment gets encoded into something queryable and verifiable.
• Local-first substrates — tools that let AI agents act on intent and prove what they did.

Building local-first substrates where agents can see, act, and leave evidence — plus a knowledge layer that compounds across sessions instead of resetting to zero.