Agents in Production: How OpenGov Built and Scaled OG Assist - Gabe De Mesa, OpenGov

2026-06-26 · Source: AI Engineer · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Advanced, long

Summary

OpenGov has successfully built and scaled OG Assist, an AI agent system integrated across its ERP software products like budgeting, procurement, asset management, and permitting. OG Assist allows users to interact with an agent that makes tool calls against product data, and can even interpret and act on screen content. The system leverages an "Effect Native Agent Loop," moving from LangGraph, and relies heavily on the open-source TypeScript library Effect for structured concurrency, fine-grained control, tracing, and logging. Key architectural components include an Agent-to-Agent (A2A) protocol for defining agent routes, human-in-the-loop for tool call approvals, and sandboxing for safe code execution. OpenGov manages long context through rolling summarization and implements generative UI. Feedback collection via thumbs up/down and automated CI evals drive continuous improvement, while internal agents like Claude and Cursor enhance developer velocity.

Key takeaway

For AI Engineers building production-grade agent systems, prioritize full control over the agent loop and robust observability. You should adopt a structured framework like Effect for fine-grained control, tracing, and logging, ensuring debuggability and maintainability. Implement human-in-the-loop approvals and sandboxing for safety, especially with mutating operations or code execution. Integrate continuous feedback and automated evaluations to rapidly iterate and improve agent performance, accelerating your development and deployment cycles.

Key insights

OpenGov's OG Assist demonstrates a robust production AI agent architecture, emphasizing control, safety, and continuous improvement.

Principles

• Bet on open-source libraries for core agent loops.
• Rigorous protocols like A2A drive development alignment.
• Continuous feedback and automated evals are crucial.

Method

OpenGov developed an "Effect Native Agent Loop" for full control, integrating structured concurrency, tracing, and logging. It uses rolling summarization for long context and sandboxing for agent code execution.

In practice

• Implement thumbs up/down for user feedback.
• Use internal agents (Claude, Cursor) for dev workflows.

Topics

• AI Agents
• Production Systems
• Effect TypeScript
• Agent-to-Agent Protocol
• Observability & Evals
• Human-in-the-Loop

Best for: AI Engineer, Machine Learning Engineer, MLOps Engineer

Related on AIssential

• AI Observability Starter Kit for Microsoft Foundry agents — Production-grade AI observability requires instrumented traces, automated quality and safety evaluations, and proactive alerting.
• Architecting Modern AI Systems — Local hosting of open-source models offers cost control, data sovereignty, and predictable performance over closed-source APIs.
• OpenClaw Architecture - Part 6: Reliability, Observability, and Evaluation — Production-ready agents require robust control planes, durable evidence, and continuous evaluation beyond simple demos.
• New in Opik: Native OpenClaw Observability, Custom Dashboards, Optimization UI Upgrades, & More! — Opik's updates enhance AI development visibility through agent observability, prompt optimization, and flexible dashboards.
• Tracer-Cloud / opensre — OpenSRE provides an open-source framework for AI SRE agents to automate production incident investigation and response.
• Agent Tracing and Observability: Log & Debug Complex AI Systems — Multi-agent systems require specialized observability beyond traditional logging to debug coordination and self-modification failures.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI Engineer.