{"type":"post","id":"what-the-harness-runs-is-a-surprisingly-simple-e35cab","slug":"what-the-harness-runs-is-a-surprisingly-simple-e35cab","title":"What the harness runs is a surprisingly simple loop.","description":"## ReAct makes the agent loop legible as a research object","retrieval_nugget":"## ReAct makes the agent loop legible as a research object","published_at":"2026-10-10","updated_at":"2026-10-10","record_date":"2026-10-10","date_kind":"published_at","topics":[],"source_urls":["https://arxiv.org/abs/2210.03629"],"schema_version":"newruntime-agent-readable-v0.2","stable_id":"post:what-the-harness-runs-is-a-surprisingly-simple-e35cab","status":"published","visuals":[],"editorial_provenance":{"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:39623834f7825da8b87e8029c76e985b5f1ba229a5a68a3124790c009b5ce5ff","reviewed_at":"2026-10-10T09:04:10.120Z","source_evidence_count":1,"verified_claim_count":3,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":3,"implication_count":2,"watch_condition_count":1,"related_record_count":0},"analysis":{"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"ReAct’s contribution is to make an agent’s reasoning and environment-facing actions an interleaved loop rather than treating reasoning and action generation as separate stages.","observed_facts":[{"text":"The paper evaluates ReAct on question answering, fact verification, ALFWorld, and WebShop.","source_urls":["https://arxiv.org/abs/2210.03629"]},{"text":"It reports absolute success-rate gains of 34 percentage points on ALFWorld and 10 points on WebShop against the imitation and reinforcement-learning methods described in the paper.","source_urls":["https://arxiv.org/abs/2210.03629"]},{"text":"On HotpotQA and FEVER it reports using a simple Wikipedia API to address hallucination and error propagation seen with reasoning-only approaches.","source_urls":["https://arxiv.org/abs/2210.03629"]}],"mechanism":"The authors describe reasoning traces as helping a model induce, track, and update action plans and handle exceptions; actions let it query external sources or interact with environments to gather information.","why_now":"Later harness discussions often treat loops, tools, and execution traces as implementation choices; this paper provides an early research account of why coupling reasoning to observable actions can matter.","implications":["Builders can use the loop as a design vocabulary for separating a plan update from an environment action and the observation that follows.","The benchmark findings motivate that architecture, but do not establish present-day reliability, safe permissions, or production performance."],"evidence_boundary":"This note is limited to the claims and benchmark comparisons summarized in the arXiv paper; it does not independently reproduce the experiments or generalize them to current models.","watch_conditions":["A useful next test is whether modern agent evaluations preserve the paper’s separation between reasoning, action, and resulting observation while measuring task success and error recovery."],"related_records":[],"new_branch_reason":"This starts a distinct analytical record because no prior New Runtime record is supplied for this research paper."},"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"]},"routes":{"html":"https://newruntime.com/posts/what-the-harness-runs-is-a-surprisingly-simple-e35cab/","markdown":"https://newruntime.com/posts/what-the-harness-runs-is-a-surprisingly-simple-e35cab.md","json":"https://newruntime.com/posts/what-the-harness-runs-is-a-surprisingly-simple-e35cab.json"},"next_reads":[{"type":"collection","path":"/posts/","url":"https://newruntime.com/posts/","title":"Field Notes","reason":"Open the latest editorial analysis.","media_type":"text/html","json_url":"https://newruntime.com/posts.json"},{"type":"collection","path":"/source-ledger.json","url":"https://newruntime.com/source-ledger.json","title":"Source Ledger","reason":"Inspect the public source evidence graph.","media_type":"application/json"}]}