Now
Now
Last updated November 2026
A snapshot of what has my attention. It is deliberately short, and it is allowed to be wrong by the time you read it.
Building
The next release of this site, a small applied-AI evaluation harness, and a set of notes on where oversight stops being a slide and starts changing a decision.
Learning
How agent workflows fail when individually safe tools compose; how to make assurance evidence legible to the people who have to sign it; how much structure a small team actually needs.
Questioning
What evidence should a person need before delegating a decision to an agent? When does human-in-the-loop become a ritual? How do you design for degraded conditions without pretending the system can handle everything?
Didn't work
A few ideas are still in audit rather than shipped as demos. The risk-based MFA/SOC thread is waiting for a repository audit. I also spent longer than I should have on visual directions that were quieter than the argument.
If something here overlaps with a problem you are working on,start a conversation.