---
type: "post"
slug: "agent-lightning-v1-0"
title: "Agent Lightning v1.0"
description: "Agent Lightning v1.0 names the pattern where the deploy-time harness owns the environment loop during reinforcement learning, and catalogues the training problems that creates."
retrieval_nugget: "Agent Lightning v1.0 names the pattern where the deploy-time harness owns the environment loop during reinforcement learning, and catalogues the training problems that creates."
published_at: "2026-09-01"
updated_at: "2026-09-12"
record_date: "2026-08-25"
date_kind: "discovered_at"
topics: ["agent-security","agents","ai"]
entities: ["arXiv"]
source_url: "https://arxiv.org/abs/2608.17528"
source_title: "Agent Lightning v1.0: Towards Harnessed Agentic RL"
source_domain: "arxiv.org"
source_terms: ["Lightning","v1.0","Agentic","Harnessed","training"]
summary_word_count: 189
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:agent-lightning-v1-0"
status: "published"
source_urls: ["https://arxiv.org/abs/2608.17528"]
visuals: []
editorial_provenance: {"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:77555b82a9a61971ad984ec7ea04b0364476cb19f263f7043bd7d1a2b255ee89","reviewed_at":"2026-09-12T10:00:00Z","source_evidence_count":1,"verified_claim_count":2,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":2,"implication_count":1,"watch_condition_count":1,"related_record_count":3}
analysis: {"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"A Microsoft-affiliated group published Agent Lightning v1.0 on 18 August, arguing that the harness managing tools, context and control flow is a critical part of an agent system rather than scaffolding around the model.","observed_facts":[{"text":"The paper defines harnessed agentic RL as a paradigm in which the deploy-time harness directly participates in model post-training.","source_urls":["https://arxiv.org/abs/2608.17528"]},{"text":"It lists retokenization, sample merging, advantage calculation and loss normalization as the resulting technical challenges.","source_urls":["https://arxiv.org/abs/2608.17528"]}],"mechanism":"the harness, not the training engine, owns the environment interaction loop, while the trainer observes only sequences of LLM request-response pairs","why_now":"The original Agent Lightning introduced a disaggregated architecture connecting arbitrary agents to reinforcement learning through an LLM endpoint proxy, an approach the paper says was later adopted by verl Uni-Agent, AReaL 2.0, slime and Polar.","implications":["That inversion is what creates the listed difficulties in retokenization, sample merging, advantage calculation and loss normalization — the trainer has to reconstruct structure it no longer controls."],"evidence_boundary":"This is a preprint abstract: the framing and adoption list are the authors' own, and no independent reproduction is cited.","watch_conditions":["The condition to watch is whether harnessed training runs publish the harness version alongside the model, because a result produced under an unnamed harness is not reproducible by anyone else."],"related_records":[{"url":"https://newruntime.com/posts/arc-agi-3-result-shows-the-harness-is-part-of-the-score","relation":"Prior coverage showing the harness inside the measured result."},{"url":"https://newruntime.com/posts/agent-work-is-moving-into-harnesses-factories-and-accepted-artifacts","relation":"Prior coverage of harnesses as the unit of agent work."},{"url":"https://newruntime.com/topics/agent-harness","relation":"Topic hub for harness design."}]}
routes: {"html":"https://newruntime.com/posts/agent-lightning-v1-0/","markdown":"https://newruntime.com/posts/agent-lightning-v1-0.md","json":"https://newruntime.com/posts/agent-lightning-v1-0.json"}
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
---

# Agent Lightning v1.0

## Retrieval answer

Agent Lightning v1.0 names the pattern where the deploy-time harness owns the environment loop during reinforcement learning, and catalogues the training problems that creates.

A Microsoft-affiliated group published Agent Lightning v1.0 on 18 August, arguing that the harness managing tools, context and control flow is a critical part of an agent system rather than scaffolding around the model. The original Agent Lightning introduced a disaggregated architecture connecting arbitrary agents to reinforcement learning through an LLM endpoint proxy, an approach the paper says was later adopted by verl Uni-Agent, AReaL 2.0, slime and Polar.

The paper names that arrangement harnessed agentic RL, where the deploy-time harness directly participates in model post-training, and its claim is that this differs fundamentally from traditional agentic RL: the harness, not the training engine, owns the environment interaction loop, while the trainer observes only sequences of LLM request-response pairs. That inversion is what creates the listed difficulties in retokenization, sample merging, advantage calculation and loss normalization — the trainer has to reconstruct structure it no longer controls.

This is the same conclusion arrived at empirically in [an ARC-AGI-3 result showing the harness is part of the score](https://newruntime.com/posts/arc-agi-3-result-shows-the-harness-is-part-of-the-score/), and it is the training-side counterpart to [agent work moving into harnesses, factories and accepted artifacts](https://newruntime.com/posts/agent-work-is-moving-into-harnesses-factories-and-accepted-artifacts/). The [agent harness hub](https://newruntime.com/topics/agent-harness/) tracks the pattern.

This is a preprint abstract: the framing and adoption list are the authors' own, and no independent reproduction is cited. The condition to watch is whether harnessed training runs publish the harness version alongside the model, because a result produced under an unnamed harness is not reproducible by anyone else.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Agent Security](https://newruntime.com/topics/agent-security.md): Explore the agent-security topic hub.
- topic hub: [Agents](https://newruntime.com/topics/agents.md): Explore the agents topic hub.
- topic hub: [Ai](https://newruntime.com/topics/ai.md): Explore the ai topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)