The newsletter for builders at the intersection of human judgment and AI speed.
The framework we keep coming back to: where humans are still the source of truth, where the seam is, and where AI actually earns its keep. Every issue maps the same three vertical bands.
Intent, taste, context, accountability — the call that has to be made on incomplete information. We document the calls that didn't fit a prompt.
Where the handoff happens — review loops, eval harnesses, escalation rules. The unglamorous infrastructure that decides whether the whole thing works.
Model picks, latency budgets, retrieval, cost math. What's actually changed in the past seven days that should change what you ship.