An evaluator with two outcomes will pick one
A pass-or-fail evaluator forced to rule on insufficient evidence produces a confident guess. A third verdict blocks like a failure while asking for something else.
Agents
Driving AI agents to software that is verified on a real device, not just compiled.
39 posts
Stop counting findings and start counting empty rounds
Discovery of unknown size cannot target a count. Convergence on consecutive empty rounds finds the tail that a fixed number of passes always misses.
Say what you truncated or the report is a lie
Top-N, sampling and no-retry all produce reports shaped like full coverage. A bound that is not named in the output is indistinguishable from completeness.
Say sorry once and it happens again
When an agent is told it behaved wrongly, an apology changes nothing. The response that holds is an edit to the runtime plus a gate that catches a recurrence.
Write the failing fixture first or you have written a description
A gate that has never failed might detect nothing. The red fixture is the only evidence it works — and one that trips on two classes localises neither.
Greening the check by weakening it
An auto-fix can satisfy a failing check by repairing the code or by weakening the check. Both go green, and only one of them is a repair.
An agent drove a KMP app to the Play Store. The codegen was the easy half.
Autonomous codegen produces an app that compiles; what makes it one that ships is a gate stack refusing any claim that carries no evidence.
Where your AI coding bill actually goes: I measured one long session
22 articles and 110 generated assets in one session. What the retries cost, and the one change that cut attempts per asset from roughly 1.8 to 1.1.
Running an AI agent in CI without it spending your budget
An agent in CI cannot be asked a question. Budget caps that actually throw, local reproduction before pushing, and the gate that degraded instead of failing.
Agent memory that is just files
Most questions an agent asks about past work are exact, not semantic. What a directory tree answers well, and the one query shape it cannot answer at all.
What should an AI agent never be allowed to do?
A permissions model derived from incidents rather than imagination: three tiers, human-only bypasses, and capabilities removed from reach entirely.
The agent that writes the check that blocks it
When the same agent writes the fix and the gate, the gate agrees by construction. RED before GREEN is what makes a machine-written check mean anything.