AI Engineer

Build with Perception Agents — Emile Baizel & Shruti Arora, Amazon AGI Lab

Oct 11, 2026AI Engineer
SummaryAI, from the source post

This talk details the development of a system that integrates coding agents with visual perception primitives for frontend development. It introduces two core components: deterministic visual annotation, which allows users to select elements, draw, or pin on a webpage to export structured feedback (JSON) for coding agents, and visual verification, which checks the rendered site against design expectations and user flows. The system connects a browser extension (Amazon Nova Act) to a local bridge and a coding agent, creating a loop where feedback is applied, code is edited, verification is run, and a report is generated. The talk showcases this loop with a podcast landing page example and discusses various audience-driven integrations and concepts, emphasizing visual human-agent collaboration and verifying rendered results over agent claims.

More from the AI-Assisted Code Verification and Deterministic Testing trail