OpenForge RL brings end-to-end training to harness-based AI agents
burkov · x · 2026-07-26
Modern AI agents increasingly depend on elaborate inference harnesses such as Claude Code or Codex to handle multi-turn reasoning, tool use, and access to external systems.
The paper argues that these stateful, multi-process harnesses are hard to train end-to-end with today’s open SFT/RL stacks, because they cannot naturally express that kind of inference setup. To address that gap, researchers from Columbia and Microsoft present OpenForge RL, an open-source framework for training harness-based agents end-to-end across diverse environments.
Related event: Microsoft Open-Sources OpenForgeRL for End-to-End Agent Training(4 posts)→
More from coding & agent
- Kolab Benchmark Uses Pro Screen Recordings to Show AI Agents Overestimate Collaboration Quality — minnesotanlp · 2026-10-27
- Dev argues Claude burns far more tokens than GPT models for the same coding work — haider1 · 2026-10-27
- Stop building agents that can't change models: stable contracts, external state, real evals — rseroter · 2026-10-27
- Cheap models can cost more in agent harnesses than frontier ones, researcher warns — vasuman · 2026-10-27
- Kanzo: A 5k-Line Rust Kanban Binary With MCP for Claude Code and Codex — allmudi · 2026-10-27
- 'A human approves everything' is a trap: approval fatigue and how to actually design safe agents — Efistoffeles · 2026-10-27