Harvey presents post-trained RLM agents for end-to-end M&A diligence, potentially extending professional agents from isolated legal tasks to an integrated diligence workflow.
state: seedheat: lowuncertainty: highconvergesscott: mediumresearch-agents enterprise-agents long-horizon-orchestrationHarvey
What is this?
Harvey published “Post-Training RLM Agents for End-to-End M&A Diligence,” dated September 8, 2026, with authors including Niko Grupen, Julio Pereyra, and Gabe Pereyra. In the supplied X snippet, Harvey says it partnered with Baseten to post-train recursive language model (RLM) agents and claims that model-harness co-optimization meaningfully improves performance in long-horizon environments. The claim is progress toward end-to-end M&A diligence, not demonstrated completion: the supplied article excerpt contains no experimental results, performance figures, or training details. A separate Harvey snippet describes extending Legal Agent Bench to M&A diligence to evaluate end-to-end legal work and support open-model training and agent research.
Why it matters to Scott
Harvey’s claimed model–harness co-optimization for M&A diligence converges with Scott’s Model-Plus-Harness Benchmark Unit position, opening a concrete publishing opportunity around a professional-agent builder treating capability as a joint systems property rather than weights alone. The supplied radar hits do not track this same development, but absent results or architectural details, this is convergence in stated approach—not validation of Scott’s long-running architecture or a demonstrated challenge to his preference for bounded AI judgments over end-to-end autonomy.
ip:concept.model-plus-harness-benchmark-unitradar:concept.recursive-agentsradar:concept.long-running-orchestrationradar:concept.agent-evaluationradar:concept.enterprise-agents
queries asked of Scott's wikis
- model-harness co-optimization and agent post-training
- recursive language models and long-horizon agent orchestration
- agent evaluation in realistic end-to-end task environments
- professional agents moving from isolated tasks to integrated workflows
- document-intensive research agents and multi-document diligence
Measured heat
now 0 pts/hpeak 0 pts/hcomments 0/hpeers p14momentum: steady2 platformsage 771h
points/hour across evidence · reading as of 2026-10-12 02:59:37.977291+11:00 · deterministic, not a model opinion
How the heat travelled
pace: p11 vs 519 stories at the 720h mark (now 771h old) — behind addom-local-coding-harness (0.5x)
Evidence (2) — ⭐ canonical anchor
Interpretation history
2026-09-09T13:34:09Z
No substantive new evidence changes this from a relevant research-direction claim into a demonstrated integrated diligence workflow. The reconstructed Harvey account and its HN link are one evidence line, not independent corroboration; the claimed model–harness co-optimization remains interesting but unquantified.
2026-09-09T13:30:00Z
grounded: converges/medium — Harvey’s claimed model–harness co-optimization for M&A diligence converges with Scott’s Model-Plus-Harness Benchmark Unit position, opening a concrete publishin
2026-09-09T13:27:36Z
case created — A first-party account of a bounded professional-agent workflow merits a seed without assuming unreported performance.
Decision trace
- 09-22 11:27review_dormantscheduled targets exhausted or 28 quiet days
- 09-22 11:27drop_targetsquiet through full ladder or over cap 8
- 09-09 23:34repriceNo substantive new evidence changes this from a relevant research-direction claim into a demonstrated integrated diligence workflow. The reconstructed Harvey account and its HN link are one evidence l
- 09-09 23:34alert_silentThere is no new engineering, performance, or access delta warranting interruption. The attributed research announcement can wait for the next briefing; substantive implementation details or results wo
- 09-09 23:34alert_routeThere is no new engineering, performance, or access delta warranting interruption. The attributed research announcement can wait for the next briefing; substantive implementation details or results wo
- 09-09 23:32alert_silentHarvey’s first-party post identifies post-training RLM agents for end-to-end M&A diligence as a concrete research direction relevant to Scott’s professional-agent coverage. However, the supplied e
- 09-09 23:32surface_candidateHarvey’s first-party post identifies post-training RLM agents for end-to-end M&A diligence as a concrete research direction relevant to Scott’s professional-agent coverage. However, the supplied e
- 09-09 23:32alert_routeHarvey’s first-party post identifies post-training RLM agents for end-to-end M&A diligence as a concrete research direction relevant to Scott’s professional-agent coverage. However, the supplied e
- 09-09 23:30groundHarvey’s claimed model–harness co-optimization for M&A diligence converges with Scott’s Model-Plus-Harness Benchmark Unit position, opening a concrete publishing opportunity around a professional-
- 09-09 23:27createA first-party account of a bounded professional-agent workflow merits a seed without assuming unreported performance.