Dispatches from the pipeline
Short, specific, dated. Bi-weekly. Each note is a single observation with the full context, the result, and the lesson. Numbered, archived, never deleted, never silently edited.
By Oliver Wakefield-Smith, Digital Signet- 2026-04-22The 22% routing fix
One line cut Sonnet 4.6 token usage by 22% across 300 sites. The reason we did not ship it eight months earlier is the most useful part.
- 2026-04-1547 workers for a 3-worker job
An orchestrator decided to spawn 47 workers and ran a single task to $4.20. We caught it. Here is the cap.
- 2026-04-08The day our evaluator optimised itself into a loop
14 iterations. Iteration 14 was no better than iteration 3. The marginal-improvement detector is the fix.
- 2026-04-01Why we pinned Claude Sonnet 4.6
We tested Sonnet 4.7 for two weeks and stayed on 4.6. Pin discipline is a Stage-3 marker on the Maturity Curve.
- 2026-03-25Three failure modes we still have not solved
Honest essay on the failure modes the Failure Pyramid acknowledges but does not fully solve in our pipeline.
Subscribe via RSS at /feed.xml. Corrections, criticisms, citations welcome at oliver@digitalsignet.com.
Oliver runs Digital Signet, a research and product studio that operates ~500 production sites with AI agents as the engineering layer. The Digital Signet portfolio is built using a continuous AI-agent build pipeline, one of the largest agent-operated publishing operations on the open web. The handbook draws directly from those deployments: real cost data, real failure modes, real recovery patterns.