https://derekwang2002.github.io/ai-intelligence/en/projects/agent-lightning/ Material changes in AI models, agents, engineering and infrastructure, with primary sources and recommendations. en Mon, 02 Nov 2026 16:56:00 GMT https://derekwang2002.github.io/ai-intelligence/en/events/ev-20261007-07/ ev-20261007-07-published Sat, 10 Oct 2026 15:56:58 GMT New TRIAL <p>Microsoft Research open-sourced a rebuilt Agent Lightning v1.0. An OpenAI-compatible proxy observes model calls from an existing agent harness, avoiding a rewrite of its execution loop inside the training framework. The roughly 3,500-line control plane runs local or Kubernetes rollouts; the authors report that about 6,000 samples raised Qwen3.5-9B from 41.8% to 56.4% Pass@1 on SWE-bench Verified.</p><p><strong>Why read: </strong>Real harnesses can enter RL directly, reducing divergence between training and deployment.</p><p>Recommendation: TRIAL</p><p>Sources: <a href="https://www.microsoft.com/en-us/research/blog/agent-lightning-v1-0-a-3500-line-lightweight-agentic-rl-framework-for-training-agents-with-real-harnesses/">Agent Lightning v1.0</a> · <a href="https://github.com/microsoft/agent-lightning">microsoft/agent-lightning</a></p>