OpenGov engineer Gabe De Mesa describes OG Assist, an agent embedded across government software products. He explains how product teams expose tools and how the assistant can use page context and highlight actions.
Gabe De Mesa describes replacing a LangGraph implementation with an Effect-based TypeScript agent loop, and using an agent-to-agent protocol as a shared interface contract. He discusses customer feedback and automated evaluations of real completionsEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria..
The talk covers human approval for consequential tool callsHuman-in-the-loop AI keeps a person involved in reviewing, correcting, approving, or guiding an AI system's work., isolated sandboxesAn AI agent sandbox is an isolated execution environment that limits which files, processes, networks, credentials, and external systems an agent can access., rolling summaries for long conversations and tracing for observabilityAI agent observability makes an agent's state, actions, tool use, failures, resource use, and outcomes visible enough to understand and operate it.. Gabe De Mesa also describes generating interfaces at runtime and using agents in internal development.
Watch on YouTube
