Build log

Every entry, newest first. Four words cover it: shipped, patched, killed, wrote. Each line has a permalink; the whole thing has an RSS feed.

2026.10.13KILLEDKilled the reflex to reach for AI first. Sometimes the better answer is existing software or deterministic automation; reserve agents for judgment, and keep a human checkpoint.#

2026.10.13WROTEThe economics of AI may hinge on workflow data. Every business has its own strange, valuable operating patterns; agents make that uniqueness useful instead of incidental.#

2026.10.13PATCHEDBeen building a lot lately: turning repetitive operations into boring, auditable systems, with human checkpoints where judgment still matters.#

2026.10.06SHIPPEDLinked the new Substack essay to my site using the existing feed sync. The build matched the deployed files; I checked the title and link on the live writing page.#

2026.09.26WROTEPassed the Claude Architect Foundations exam. Officially certified.#

2026.09.26PATCHEDPatched a few operations bots. The details stay private; the reliability work keeps moving.#

2026.09.20WROTEMock exam scores finally hit 95%. One last focused week for the Claude Architect Foundations certification. Let’s go!#

2026.09.20WROTEA few good examples can unlock surprising agent behavior. Capability is cheap; deciding where autonomy stops is the architecture.#

2026.09.20WROTESilent failures are the real danger. Anything with meaningful, ongoing consequences needs deterministic validation or a human checkpoint before it acts.#

2026.09.10WROTEStill taking my time learning Substack. Thoughts are being jotted down now; the posts can come when they're ready.#

2026.09.10KILLEDPut a lot of things on pause to study for the Claude Architect Foundations exam. Choosing focus over fake momentum.#

2026.08.28WROTEPrivate work is moving toward public checkpoints. The distinction that matters: deterministic automation can be audited; agentic judgment needs a human checkpoint before it acts.#