Summary
Cole Medin introduces Omni Agent, an open-source meta-harness released by Databricks — driven by their CTO — that sits above individual AI coding assistants and orchestrates them in coordinated, multi-step workflows. The core idea is that top engineers are no longer relying on a single model or terminal session: Claude Code, Codex, and Pi each have different strengths, and a meta-harness like Omni Agent makes it straightforward to delegate implementation to one and code review to another without manually managing handoff documents across terminals.
The video demonstrates Omni Agent’s built-in orchestrators Poly and Debbie, showing how Poly takes a GitHub issue or freeform prompt, kicks off a Claude Code subprocess for implementation, monitors its progress, and then routes the output to Codex for review — all using existing CLI credentials without re-authentication. Medin reports getting this workflow running in under 10 minutes. He also walks through Omni Agent’s primitive-based config system for building custom orchestrators with configurable skills, agent pools, and sandbox execution.
Medin frames this as the next major evolution in AI-assisted development, arguing that harness quality now matters as much as — or more than — the underlying model, particularly given recent disruptions to LLM availability. Omni Agent’s additional features include shared sessions for team collaboration and guardrail support for production environments.
📺 Source: Cole Medin · Published June 15, 2026
🏷️ Format: Tutorial Demo
