re:compoundThinking
Essays on what AI has made cheap, what has become scarce, and how serious professional work should now be produced, reviewed, and continued.
Featured essay
Why the old review interface breaks when finish stops carrying reliable signal about the thinking underneath — and why polished work can quietly place itself beyond the reach of checking.
A fiduciary case for reinvesting the gains from task automation into the human capabilities that frame, steer, evaluate, synthesize, and preserve understanding.
On the operating conditions that let a team — or a human and a model — think together under pressure without polish substituting for truth.
Why the fiduciary case against stripping judgment-layer capability is stronger than the compassion case — and what boards should ask about where the freed capital actually goes.
A fourteen-month matter closes, and the handoff note defeats the partner three Saturdays running — because what carried the matter was a maintained live state, and the standard memo can only flatten it into a record.
Kalman filters, correlated errors, and the agent-count fallacy — why a governed pair, one human and one AI, differently wrong, beats a swarm of similar agents on problems that are unstable while you solve them.
Why excellent summaries can still destroy valuable understanding — and why serious organizations need checkpoints designed to resume thought, not merely artifacts designed to record it.
From theory to practice
re:compound turns these ideas into repeatable human behavior and model-side operating assets inside one real working relationship.
A practical first step
If the thesis is familiar, the useful next question is where it can be tested against consequential work and real review.
Discuss where to begin →