James Peter
Co-Founder
JustEvery
The agents went rogue at 2%
The agents went rogue at 2%
Giving an agent a skill sounds straightforward: it loads a Markdown file, follows the instructions and completes the task.
Ours launched a long-running, paid process and told the agent to wait for the result. But agents repeatedly started the process, became sidetracked and rebuilt the output themselves. In one test, only 1 of 16 runs produced a fully correct result, while 12 of 16 paid for exports they never used.
The agents had gone rogue, and we thought we knew the fix. We rewrote the instructions. Then we rewrote them louder, again and again. The agents still got sidetracked.
The real problem turned out to be one line, and it wasn't in the prompt.
This talk retraces that investigation from the agent's side of the screen: why an agent's strangest decision can be its most reasonable one, and what agents need from the systems they depend on.
James Peter
James is an entrepreneur and software engineer building AI products across design, coding and automation. He is the co-founder of JustEvery, a group of AI companies run by a small team and thousands of agents. They include 12ui, which turns visual references into editable, production-ready interfaces, and Every Code, an popular open-source fork of OpenAI's Codex CLI with over 4k GitHub stars. He is now building RunEditRun, an open-source alternative to many SaaS products. Trained in physics and NLP, James is most interested in the gap between what models can do in a demo and what it takes for agents to work reliably when nobody is watching.