The audit was green, specific, and about the wrong artifact
An audit passed on 113 pages while the source directory held 145, because it inspected a stale build. A correct check on the wrong artifact reports confidently.
AI-Driven Development
Shipping production software with coding agents — context, evals, guardrails, and what breaks at scale.
36 posts
Never attempted, and the summary said complete
A missing trailing newline dropped the last entry from a loop. Nothing failed, the summary matched what ran, and one item was never attempted.
A warning is not a refusal, and the driver could not tell
A driver keyed on the word ok halted a batch of 49 at item 19, because the check reported a warning rather than a failure. Gate on the blocking count.
Name the fault, and rule out the wrong fix
An error that describes the request sends you to change the request. A good message names the real condition and closes off the plausible wrong repair.
An agent-driven refactor, with the receipts
Before and after on a real module migration: what proved the change was inert, what the agent got wrong, and the receipt that made it safe to run at all.
The most useful thing my agent does is refuse
Refusals are only useful if they name the remedy. Four that fired during one session, what each redirected me to, and the one that fired on nothing at all.
The context window is not your constraint — what you put in it is
Scoping a shared instruction bundle saved 8.2% per worker, against 40% guessed. Why residency beats capacity, and the surface that was never loaded at all.
I swept the parameter instead of arguing about it
A threshold argument settled by measuring four values over the real corpus. The knee, the trade that stops paying, and why the numbers go in the comment.
The same rule in two files is how agent-written code drifts
A protected-path list was typed by hand and had drifted nine entries behind reality. Derive the list instead of copying it, and gate the copies you keep.
The prompt is not the artifact — the runtime is
A behaviour defect belongs in the command's runtime file, not in better instructions next time. What to edit, how to lock it, and where that stops working.
429 is not an error, it is a schedule
Transient and terminal failures need opposite responses and look identical. How to tell them apart, and why partial progress must be persisted before it is believed.
Six guardrails that stop my AI agent mid-command
Six PreToolUse hooks that refuse an agent's command before it runs. What each one caught, and why a guard the agent can talk its way past is not a guard.