Adrià Cidre

Coverage tells you a test ran.

Confidence is what the model thinks of itself.

Append-only history and GDPR-style erasure look incompatible.

My agent pipeline runs for hours, survives Ctrl+C, and resumes from where it stopped.

Every system I built had a tenant_id column and a discipline problem.

Two agents with identical latency can feel completely different.

A maintenance agent filed 559 bugs and fixed 412 on its own.

Feature branches and merge queues were how we coped with slow humans.

I stopped feeding hospital PDFs to a vision model.

Most builders assume end-of-turn detection is a silence threshold.

An 8th specialist that touches the browser — navigating, clicking, and typing to find bugs that scripted journeys miss.

The leap from AI-assisted coding to autonomous batch processing.

Seven specialists, each with their own checklist — QA, UX, UI, Security, Performance, Data Leakage, Language.

Exponential backoff, dual timeouts, SSE heartbeats, idempotency caches — the unglamorous patterns that keep LLM-powered systems running at 3am.

Real browser, predefined journeys, specialist agents checking every step.

Users ask similar questions in different words.

The AI picks up a planned issue, implements it, tests it, commits, and marks it done.

Context windows fill up fast in long AI conversations.

Vague instructions produce vague implementations.

The bottleneck isn't the AI — it's me.