Postponed · new date to follow · SeedLegals, London
Save tokens. Ship with confidence.
Your agent works in the demo. Then your customer's risk team asks what it did last Tuesday before they grant write-access, and the only answer you have is the agent's own log.
That log is a diary, not an audit.
When the agent gets confused it doesn't stop. It loops in retries, and token spend on agent runs is fat-tailed: a handful of looping runs burn many times the budget of tasks that should have taken three calls.
You write the spec. The instrument reads every change the agent makes against it and colours it: green inside the lane your spec drew, amber at the edge, red somewhere you never asked for.
You'll leave with
- • A spec-first way of working with coding agents that shows you a looping run early, so you stop it before it becomes the fat tail of your token bill.
- • What your customer needs to see before they let your agent act, and how to show it.
- • The kind of record the EU AI Act asks high-risk systems to hold, kept as you work, not written up afterwards by the agent that did the work.
The 9 November evening is postponed.
A new date follows. Anyone who requested a seat hears it first, and the seat carries over.
About the speakerA short talk, a live demo on a real agent, Q&A, food and drinks. House rule: anyone who calls their agent "deterministic" buys the next round. Runs locally in VS Code; the install and the measurement are free and open source (MIT), and only the warranty and attestation layer is licensed. Speaker: Elias Moosman, ThetaDriven, author of Tesseract Physics. Hosted at SeedLegals, London.