Harvey

I evaluated Harvey by reviewing publicly demoed legal-research and document-drafting workflows and comparing the outputs against the kind of memo and contract-review patterns I have seen at mid-sized commercial law firms. The cohort I had in mind was BigLaw and large in-house legal teams evaluating a generative AI assistant for litigation research, contract drafting, and due-diligence document review. This is a fresh vertical for the catalog: no prior hands-on review covered legal AI in the prior 60.

The specialization is the headline. Harvey is built specifically for legal use cases with training and fine-tuning that targets statute lookup, case-law citation, and document-comparison patterns rather than generic summarization. The OpenAI partnership and Allen and Overy (now A and O Shearman) anchor-customer story carry weight in the BigLaw evaluation conversation because it signals the model has been pressure-tested against real partner-grade work. The integration story with iManage and similar document management systems matters operationally because legal teams will not migrate primary work surfaces, so the AI has to come to where the documents live.

What I cannot directly verify from outside a deployed firm is real-world hallucination rate on case-law citations, which is the single biggest risk in legal AI: a fabricated citation that lands in a partner-reviewed memo is a malpractice event, not just a quality bug. Harvey requires firm-level enterprise sales and contract negotiation with no transparent pricing, which means small or mid-sized firms cannot run a quick self-serve evaluation and the procurement cycle is multi-month. The product roadmap moves fast (which is good for capability) but creates change-management overhead for firm IT teams trying to keep training, security review, and rollout aligned.

Harvey is the right pick for BigLaw firms and large in-house legal teams with dedicated KM partners, enterprise IT, and a multi-month procurement appetite. Skip it if you are a solo attorney or sub-50-attorney firm (the enterprise sales motion is wrong-sized for that scale, and tools like Spellbook or CoCounsel target smaller firms), if your work is primarily transactional with stable templates rather than research-heavy litigation (Spellbook, Ironclad, or Diligen are stronger), or if you cannot afford the procurement cycle and need self-serve trial access.