{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/727528de450a48d29a2ac20b279e26fc\" frameborder=\"0\" width=\"1846\" height=\"1384\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1384,"width":1846,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1384,"thumbnail_width":1846,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/727528de450a48d29a2ac20b279e26fc-ef2ed0f9424fa7e0.gif","duration":233.259,"title":"SPEC-27 for Spec-Driven AI Agent Validation and Monitoring 🚀","description":"Hi, I am Steve, and I walked you through SPEC-27, a new product for specification driven validation and ongoing monitoring of AI agents. We define the system in specs, then check behavior against them across multiple iterations. In one example with a Chain based filing Q and A agent, we saw about 85 percent clean accuracy on tough criteria, with robustness lower under variations like lexical substitution. We also run red team attacks to test out of domain behavior. We are in early access, and I asked you to come kick the tires and reach out for a demo."}