Notes from the workshop

What we learned building things, written up while it was still fresh. No thought leadership, no listicles.

A postmortem format that people do not dread

Most incident reviews either assign blame quietly or avoid the interesting question entirely. We rewrote our template around one rule.

ProcessQualityEngineering

Process · 5 min

18 Jun 2026

The tests we write for a model-backed feature before any UI

Testing an AI feature through its interface is slow and unreliable. We test the reasoning layer directly, before a single screen exists.

TestingAIQuality

Quality · 6 min

17 Jun 2026

Pricing a fixed-scope project when part of it is unknown

Clients ask for a fixed price on work that includes a genuinely unknown integration. Here is how we split the quote so both sides are honest about the risk.

ProcessScopingClients

Process · 6 min

16 Jun 2026

What changes when a model can run the tests itself

Giving an agent the ability to execute the test suite, not just read it, changed the kind of mistakes it made and the kind we had to watch for.

AITestingEngineering

AI · 6 min

15 Jun 2026

The week we spent only on error messages

A client asked why support tickets kept arriving for problems the system already knew about. The answer was in the error copy nobody had reviewed in two years.

Case studyProductQuality

Case notes · 6 min

12 Jun 2026

Acceptance criteria a tester can disagree with

Criteria written to be satisfied are less useful than criteria written to be argued about. The distinction shows up the first time someone tries to break them.

TestingQualityProcess

Quality · 5 min