Done Is a Claim, Not a State: Building Evidence Gates for Coding Agents
A field guide to separating completion claims from attempts, readbacks, durable artifacts, and authorized transitions.
Read the field guide currently building the SDLC for an AI-first world
Building toward an SDLC where AI is a first-class participant.
Most claims about AI and engineering run on vibes. I’d rather show the work: six years of turning “this should be faster” into measured outcomes, now spent making agent output safe enough to ship. More about how I got here →
Selected work
view all →The claim, and the receipts.
vibes → evidence
Replacing AI-productivity anecdotes with measurement
50K+ repos
Separation-of-duties compliance, audit prep down ~75%
$1M saved
Annual cloud spend, ~2,000 VPCs consolidated
hrs → s
Account + VPC teardown, automated end to end