Specify
I decide what gets built and why, set the constraints, and write the acceptance criteria the work has to meet.
AI in Practice
How I work with AI across the things I build.
I use AI daily, on real projects with real constraints. Not to replace the thinking, but to move faster through the parts that would otherwise stall me: research, first drafts, code I can scope and read but leave the tool to write.
The senior version of this skill isn't prompting. It's judgment. I set the requirement, I direct the work, and I know when the output is wrong. It's the same instinct I've brought to fifteen years of commercial work: hold the brief, review hard, sign off on nothing I can't stand behind.
Direct
I brief the tool, review what comes back, and send back what doesn't hold up. Most of the work is in the rejection.
Verify
I check claims against reality before I accept them. A confident answer that's wrong is worse than an honest gap, so nothing ships on trust alone.
Operate
I do the hands-on work the tool can't reach. I run the builds and the deploys myself.
AI principles
The rules I hold myself to when the work matters.
I hold the requirement. The goal, the constraints and the threat model are mine, and the work serves them rather than the other way round.
Verified means verified. I don't accept a claim because the tool sounds sure. I check it against the source, and a confident answer that turns out wrong costs more than an honest "I don't know."
A human decides. On anything that matters, the tool proposes and I dispose. The judgment call stays with me.
I keep a record. Long builds run across dozens of sessions, so I keep written state of what's decided and what's open, and I don't carry a contradiction forward.
The consumer HealthTech venture and the civic-transparency project in Experience. Home automation, the private browser, the data-centre atlas and the vehicle tracker in Interests. Each one used AI to build; each one is mine to explain.