Blog
Writing, worked out in the open.
The living counterpart to the Philosophy page. Essays argue ideas; use cases examine starter topologies; notes record product decisions. Newest first. RSS.
- ESSAY Agents need roads — and real guardrails Legible structure helps agents stay on course. Capability roots, Test-mode stubbing, and Production approval also enforce boundaries they cannot cross.
- ESSAY Not all judgment is delegable Accountability cannot be transferred to a model. A tool for serious work has to be built around that fact.
- USE CASE The analyst's weekly memo, as a flow A linear pipeline that turns a week of source material into a reviewed, sent memo — with a human sign-off in the middle.
- USE CASE Intake triage with routing and escalation A branching pipeline that classifies incoming requests, routes by urgency, and escalates the hard cases to a person.
- NOTE Retries should fail closed around side effects Pure work can retry freely. External writes need an explicit replay protocol, because a crash can leave the truth unknowable.
- USE CASE Synthesizing feedback from many sources A fan-in pipeline that merges several feedback streams, extracts entities, compares across sources, and validates before it writes.
- ESSAY Verifiability is necessary but not sufficient for trust Being able to check an output after the fact is not the same as being able to trust the process that produced it.