The talk argues that managing many coding agentsAn AI coding agent is a tool-using AI system that can inspect, modify, and validate software within a repository. can exhaust the person who must steer, inspect and approve each one. It borrows the overview and quick controls of a strategy game: a map and activity panel show what agents are working onAI agent observability makes an agent's state, actions, tool use, failures, resource use, and outcomes visible enough to understand and operate it., while attention shortcuts help a human jump to a pending question or approval.
The speaker then describes agents proposing tasks from a codebase, an orchestrator breaking larger requests into isolated workAgent orchestration coordinates AI agents, tools, people, tasks, state, and control flow so a larger workflow reaches a verified outcome., and recurring loopsAn agent loop repeatedly observes a task's state, chooses an action, checks the result and decides whether to continue or stop. that start background jobs. A review kit gathers changed files and visual evidenceAgent completion verification checks evidence that an agent achieved the requested outcome rather than merely producing intermediate work. so a person can assess several implementations and choose among them. These are demonstrations of a particular tool, not evidence that unattended agent work is generally safe or effective.
The final sections explore shared workspaces where teammates and agents can coordinate, and a simpler game-like interface intended for people who do not want file-level controls. The latter is presented as an early concept. The reusable idea is to match the amount of agent detail and control to the person doing the review.
Watch on YouTube




