Spec-Driven Agentic Development: Velocity Without the Slop

Agents should be trusted with big-picture feature work; that's how you increase velocity. But throwing a ticket at an agent and saying "implement this" isn't a recipe for working code, much less high-quality code. You need a spec-driven process that aligns you and your agent before the build, so you reach a high-quality solution the first time.

In this session, we'll break down what goes into a good product spec and tech spec, then codify those principles into reusable Skills that generate high-signal, easy-to-review documents every time. We'll also share our findings from evaluating models across the full workflow, so you can deploy the right agent (and harness) for every job.

Plus: our model and harness evaluation findings across spec generation, self-improvement, and code implementation.

Who it's for: Engineers and engineering leaders delegating feature work to coding agents who want a repeatable process for quality output.