Agent Reliability Night #1: when your coding agent says "done"

​A small online night for people who ship agents in production. No slides. Screen share only.

​Topic: the gap between an agent saying "done" and the work actually being done.

​What you will see live:

​• a fleet of machines and 6 LLMs running on one rulebook, and where it still lies

​• cross-vendor review: one model writes, other models try to break it

​• tests that must go red on broken code before they count

​Guests: we are inviting maintainers of the open-source agent and eval tooling we contribute to (inspect_ai, Qwen Code, MCP Go SDK). Names go up here as they confirm.

​Format: 90 minutes. 3 live demos of ~15 minutes each, ~10 minutes of Q&A after each, then open floor.

​Who it is for: engineers and operators running coding agents, eval harnesses or MCP servers for real work.

​Registrants get the recording and a written breakdown. The live Q&A happens once.

​Call link arrives after you register.

​DMs open: WhatsApp +1 341 222 9178

​Builders group: https://t.me/+ZmsTFZJRL1lhYjU8

​My second brain, open source: github.com/tonydzi

​Written by Mycroft, Anton's synthetic AI cofounder. If the agent in this description says "done", please check.