SPEAKER
Dru Knox
Head of Product & Design · Tessl
Dru Knox is Head of Product & Design at Tessl, where he leads the development of AI-native tools purpose-built for developers. A seasoned product manager, Dru has spent his career working on deeply technical and developer-facing products at companies like Google and Airtable. Over the past several years, he’s focused on machine learning and generative AI at scale, driving product innovation at Grammarly, his own startup, and the AI-native social network Cantina.Dru is passionate about making LLM-powered products that are fast, useful, and intuitive, especially in the codegen space. He brings a thoughtful, pragmatic approach to building tools that serve real developer needs, grounded in experience across both big tech and startups.Outside of work, Dru’s interests include improv comedy, Dungeons & Dragons, and philosophy. His favorite podcasts are Cortex, The Adventure Zone, and Sharp Tech. Originally from Virginia, he’s currently based in London and still counts winter (and snow) as his favorite season.
Upcoming sessions
See complete program4 November
12:25 PM · 30 min
Harness Engineering in Practice: The Loops That Make Agents Trustworthy
A coding agent is not just a model. It is the model plus the system around it: context, permissions, checks, reviews, evaluations, and maintenance loops that determine whether its output can be trusted.
This session moves beyond defining an agent harness and shows how one works in practice. We will follow a real reliability loop through its full lifecycle: the failure it was designed to catch, how the team encoded the right boundary, what happened when the loop fired, and how the result changed the harness itself. Along the way, we will examine the trade-offs between strict controls and useful autonomy, why one-off guardrails decay, and how teams can turn production failures into reusable improvements.
Attendees will leave with a practical model for designing harness loops that catch problems before merge, improve with use, and reduce the amount of manual supervision needed as agent adoption grows.
