I'm Mike McMahon — infrastructure/SRE engineer, AI practitioner, homelab builder. This site is where I document the work: what I'm building, how it broke, and what I learned along the way.
Use the sidebar to navigate. Start with How I Got Here if you're new, or jump straight into the Session Notes for technical content.
Subscribe to new posts: Email digest (daily) or RSS feed
Mid-task messages now reach the agent while it is still working, and it changes the shape of the session more than any capability upgrade has. Notes from a day where three wrong turns each cost one sentence to correct instead of a full re-brief, plus the honest limit: steering got better, correctness did not.
A day that started with a name-constrained root CA and turned into a study of status reports that were not facts: a fix that sat merged and undeployed for 50 days, three copies of a certificate where a stale one would have worked forever, a curl call that reported pushing metrics it had not pushed, and a pull request GitHub still says it merged into a commit that does not exist.
A failure taxonomy from instrumenting a multi-agent system against live infrastructure: what each mode looks like, the metric signal that detects it, and the control that contains it. Includes the cost data showing a 57x price difference for identical diagnostic quality, and an honest account of the failure class none of this instrumentation catches.
Two days on OpenBrain's ingest gate and retirement airlock. Both were built, reviewed, merged, and covered by tests. Neither had ever executed once. What surfaced when they finally did: a foreign key that made deletion impossible, a smoke check that could not fail, migration files that lied about their own status, and an anon role that could have emptied the vault.