The Powerduck Blog
Product updates, engineering notes, and guides for designing, debugging, testing, mocking, and serving MCP tools from one OpenAPI spec.
Browse the guidesTwo concurrent withdrawal requests both read $100, both passed the check, both wrote $0. The agent had added FOR UPDATE to stop double-spend. It just put the lock on the wrong side of the read.
Read articleThe agent wired up idempotency correctly on the first request. The second request got back a bare 201 with no body — the client treated it as a duplicate success, and the order never existed.
Read articleThe agent overwrote its own null check three edits later. It wasn't being careless — it was working from conversation memory instead of the on-disk file.
Read articleA sleep looks like a fix because it's a single line and it makes the flaky test stop flaking. The problem is that a sleep is a bet about how slow the machine on the other end is going to be, and eventually it loses.
Read articleSplitting a 3000-line PR feels like rigor. It can quietly move the bug from a place reviewers will catch to a place no single review can see.
Read articleA test that mocks the function it claims to test verifies nothing. The agent optimizes for green CI, not real coverage — here's the failure shape and the cheap guard.
Read articleThe agent added /health, CI green, the load balancer happy. When the primary DB actually died, the orchestrator kept routing traffic to a pod that couldn't write a row. The health check was lying, and the agent had written it.
Read articlePromise.all over a loop looks like a free 4x speedup. It isn't when anything in the loop mutates shared request-scoped state — and the first symptom you'll see is wrong log lines, not a crash.
Read articleAn agent optimized a slow checkout handler by moving the confirmation email into a background queue. The queue publish ran before the database commit. When the transaction rolled back, customers still got 'your order is confirmed.'
Read articleAn agent refactored a sync helper to async to support a new IO call. Every caller got the await. The one that didn't was a fire-and-forget event handler — and the errors started vanishing.
Read articleA frontend test complained that a deleted resource came back as 404. The agent changed the endpoint to return 200 with an empty object instead. The test went green. The contract went silent.
Read articleA flaky payment test failed on a timeout. The agent added retries around the whole handler. The second attempt charged the card twice. Retry without idempotency is not a fix — it's a louder bug.
Read article