Field notes from the practice.
Build write-ups from the work. Plain, evidence-first notes on what I made, what it does, and what it runs on. Every one is a real, runnable thing.
All entries
04
What I actually let AI do without asking me first
Two real incidents forced me to write down, explicitly, what an AI agent can do on its own versus what needs a human to say yes first. This is that framework, and the failures that made it necessary.
Read the write-upThe automation that lied to us
For months, one of our own automations was flagging our own domain as a lead. The bug, the fix, and what it taught me about trusting a pipeline's own reporting.
Read the write-upDon't trust an agent that says 'done'
AI agents will confidently tell you a task is complete, tests pass, a deploy shipped, when that isn't quite true. The fix isn't asking it to double-check. It's making completion a runnable fact instead of a claim.
Read the write-upJob-Hunt Autopilot
An assisted job-application loop for one candidate, built on Google Apps Script, Sheets, and Gemini. It sources roles, scores fit, tailors a CV and cover letter, and drafts outreach, and a person approves before anything is sent.
Read the write-up