04 / Writing
Notes tagged “reliability”
Engineering notes, experiments, and the occasional release. Written as we go, so expect working detail over polish.
01Designing reliable AI workflowsMost agent failures are not model failures. They are workflow failures. Here is the shape we keep arriving at when the output has to be right.02What we log when an agent failsThe exact fields we record for every step of an agent run, and why each one has earned its place during an incident.