Three things you can check yourself.

A personal AI operations layer that runs three businesses day to day. Not a demo: it schedules jobs, sends real notifications, and blocks its own mistakes with rules written into code rather than left as good intentions.

What it proves: judgement about where automation can be trusted and where a person has to stay in the loop. The counts above are pulled straight from the live system.

A tool that reads AI-assisted writing and flags the lines that have drifted into machine cadence, naming the rule each one breaks rather than rewriting it for you.

What it proves: the same judgement, applied to someone else's problem, against a fixed brief and a fixed deadline. Built for Clief Notes competition 9, Honourable Mention. Check it at github.com/patkelman/author-voice-checker.

A script that runs before the day starts, checking eighteen separate points of failure across email, calendar, backups and infrastructure, fixing what it can safely fix and naming the rest.

What it proves: unattended automation has to know the difference between "fix it" and "tell someone". Get that wrong and it becomes the thing that quietly breaks everything. Ask for a walkthrough and watch it run.