AI use

I build AI systems, and I use them in my own work. Models and agents help me research, test ideas, write and review code, automate tasks, and turn rough notes into clearer explanations.

That includes drafting. Some wording and implementation on this site starts with AI assistance. I do not want to imply that every sentence or line of code was produced without it.

What matters is whether the work holds up. A convincing model response is not evidence. Research claims need sources; engineering claims need working code, measurements, and tests. I remain responsible for what I publish here, including mistakes.

Where AI plays a material part in a project or result, I want that contribution to be clear. The aim is not to hide the tools. It is to make the work useful, understandable, and open to scrutiny.

My agentic fleet

I work with multiple agents across research, implementation and review. I divide work into clear tasks and use independent reviews to challenge results. More agents do not remove the need for judgement; coordination and verification are part of the engineering.

Orca

I use Orca as a workspace for agent-assisted engineering. It helps me bring multiple workstreams together rather than treating each conversation as an isolated task.

Ledger

Ledger is an agentic memory system I built for my working life. It connects notes, commitments and sourced history so I can return to the context behind a decision, not just its summary. Its earlier failures taught me that a useful memory system needs evidence, corrections and clear human decisions, not merely more extracted tasks.