I read everything.
Email is the only channel I check reliably. I try to reply to anything substantive within a week; if you have not heard back in two, send it again — it did not offend me, it got buried.
Collaboration
Especially if you have run agents against real systems — not a demo — and think some part of what I have written is wrong. Disagreement from practitioners is the most useful mail I get.
Peer review and programme committees
I review in agent engineering and evaluation, constrained and neuro-symbolic modelling, AI governance and privacy, and graph learning for financial systems. Send the venue, the deadline and the volume.
Speaking
Talks on designing the agent harness, evaluating systems with no ground truth, and what a control tester actually asks of an AI-mediated process. Copy-ready bios are on the about page.
Press and citation
Happy to be quoted on any of the above. If you are citing an essay, the citation line is at the bottom of every post.
- Requests to review or endorse something I have not read.
- Anything asking me to comment on my employer's systems, roadmap or numbers. I will not, in any form.
- Generic outreach that does not reference anything on this site.