I lead AI engineering at OFX, building production agents that take on real operational work with humans still in the loop. Outside that, I build small agents for everyday problems and put them online. Here are a few of them.
An agent fleet that hunts loyalty points while you sleep. Governed by design: it refuses, asks, and logs before it acts.
An AI agent security platform: intelligent monitoring for agents that take real actions.
A multi-agent system that works out exactly what is going on with your energy bill.
An agent-driven advisor for navigating compliance questions.
An agent that takes real actions has to earn being left running. Mine refuse, ask, and log.
LLMs are good at finding and reasoning. Anything a person acts on, deterministic code does, so the numbers are right.
The wins and the failure modes both go public. That is most of the credibility.