Break Your Own Harness
Sunday, July 12 in San Francisco. 12:30 to 7pm.
Most people test their agent harness by hoping it works. We test it by trying to break it.
This is a half-day build-and-stress-test. You ship one agent harness in the first half, then spend the second half trying to snap it. Edge cases, adversarial inputs, the failure modes you usually pretend won't happen. The builders whose harness survives the longest walk away with the real prize: a harness they actually trust.
Hosted by a fast-moving agent-tooling team, with compute and infrastructure backing from a leading GPU cloud and an open-source model lab.
Who it's for: agent builders, ML engineers, and applied-AI founders who would rather find the cracks themselves than ship them to production. Room for 80 to 150.
Bring a laptop and something you've been afraid to test under load.