Demonstrates verifying coding agent output by inspecting the rendered page, plus structured visual feedback as agent input, instead of trusting the agent's self-report.
Key takeaways · AI-distilled
Amazon AGI Lab's perception agents workshop builds on two primitives: visual annotation, where a person selects an element, draws on the page or drops a pin and exports structured JSON feedback, and visual verification, which checks the rendered site.
The annotation step is deterministic, while verification lets a coding AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → inspect the rendered result of its change against design expectations and user flows, instead of accepting the agent's claim that the change is done.
Participants connect a browser extension to a coding agent through a local bridge and add an Apply Changes button, so exported feedback flows into code edits, a verification run and a report. Several audience extensions were still at the concept stage.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
videoHow Secrets Leak Through AI Agents (and How to Stop It) — Venice AI
videoFrom Zero to Leaderboard: Agent Evaluation — Wolfram Ravenwolf, Weights & Biases
videoBuild a Platform and Watch It Burn — Michael Forrester, Accenture & Whitney Lee, Datadog
videoContinuously Improving Agents with Langfuse — Annabell Schäfer & Lotte Verheyden, ClickHouse
