Writing
The complete chronological archive.
All writing
Page 7 of 11
Turn Painful Agent Runs Into Evaluated Skills
Jul 2026·Build AI agent skills from hard-won runs by extracting invariants, bounding authority, and testing new cases instead of saving an unverified prompt.
Define AI Agent Work as Verifiable Execution Slices
Jul 2026·Use AI agent task decomposition built around one outcome, changed surface, proof path, rollback, and stop condition instead of arbitrary small tasks.
Make Agent Integration Failures Explicit Before Asking Humans
Jul 2026·Classify AI agent integration errors before asking humans for context. Preserve missing input, unavailable capability, denied access, and stale sources.
Before an AI Agent Asks a Question, Make It Search
Jul 2026·Use AI agent context gathering to reduce avoidable team interruptions. Separate cheap lookups from decisions that genuinely need a human.
Restart AI Coding Sessions When Context Stops Being Verifiable
Jul 2026·Restart an AI coding session when context stops being verifiable. Use a checkpoint that preserves facts, uncertainty, and the next safe probe.
When Vault Migration Is Really a Cryptography and Trust-Boundary Problem
Jul 2026·Vault transit migration can fail after the database copy; validate key parity, token scope, and app decrypt paths before cutover.
When the cluster was ahead of the code
Jun 2026·Terraform drift can make a clean apply silently roll back live changes. Learn how to spot reverts, stop, reconcile code, and deploy safely.
Measuring Performance Work Before Declaring Victory
Jun 2026·Use performance attribution to compare old and new paths, calculate database deltas, control confounders, and decide what to retain.
When Project Risk Is an Accountability System, Not a Staffing Problem
Jun 2026·Project risk accountability fails when ownership, readiness, and escalation are invisible. Use ledgers and gates before adding staff.
When Agents Can Write Their Own Skills, Retrieval Becomes the Product Boundary
Jun 2026·Wrong skill retrieval makes agent skill persistence risky. Treat skills as gated, versioned product surface with evals and tool limits.