---
schema_version: "newruntime-agent-readable-v0.1"
type: "raw_signal"
id: "tg-2614"
slug: "harvey-trains-legal-agent-with-applied-compute"
title: "Harvey trained a legal agent with an applied compute loop"
description: "Harvey describes domain experts, task environments, evaluation, and iterative training as one system for legal agent performance."
observed_at: "2026-07-03"
why_it_matters: "High-stakes vertical agents need expert-designed tasks and feedback loops around the model rather than a domain prompt added at deployment time."
novelty: "notable"
verification_level: "source-inspected"
signal_type: "field-report"
evidence_kind: "mixed"
status: "published"
telegram_message_id: 2614
telegram_url: "https://t.me/qwgai/2614"
topics: ["vertical-agents","evals","legal-ai"]
entities: ["Harvey"]
related_patterns: []
source_urls: ["https://harvey.ai/blog/training-a-legal-agent-with-applied-compute"]
import_batch: "telegram-2026-07-17-v1"
routes: {"html":"https://newruntime.com/signals/harvey-trains-legal-agent-with-applied-compute/","markdown":"https://newruntime.com/signals/harvey-trains-legal-agent-with-applied-compute.md","json":"https://newruntime.com/signals/harvey-trains-legal-agent-with-applied-compute.json"}
source_format: "telegram-export-normalized-json"
---

# Harvey trained a legal agent with an applied compute loop

## Observation

Harvey describes domain experts, task environments, evaluation, and iterative training as one system for legal agent performance.

## Why it matters

High-stakes vertical agents need expert-designed tasks and feedback loops around the model rather than a domain prompt added at deployment time.

## Provenance

Normalized from QWG AI Telegram message 2614. The original Russian-language record remains available at https://t.me/qwgai/2614.