AI coding agents move the bottleneck in software teams. Once a model can produce a decent patch in minutes, the scarce work is steering: giving it the right context, catching drift early, preserving architecture, and making failure easy to catch and cheap to fix.

This talk makes Harness Engineering concrete: the platform work around agentic coding. A useful harness turns the repository into an environment agents can operate inside safely. That means project commands, task-specific skills, architecture maps, coding rules, tests the agent can run and understand, feedback loops, and guardrails that keep changes observable and contained.

You will leave with a practical checklist for building your own harness: making your codebase more agent-ready, reducing drift and cognitive debt, while keeping accountability where it belongs: with humans.

​Building a solid mental model of Edge AI and hybrid patterns is the first step toward transforming UI state and structured data into reliable AI inputs. This session moves past the hype to show how on-device AI can power actual user-facing features while maintaining low latency, privacy, and predictable behavior by keeping sensitive data off the cloud.

​We will break down the engineering trade-offs, context design, and safety mechanisms required to close the agentic loop. You’ll leave with practical proof-of-concepts for building AI features that are useful, stable, and ready for production.