
Uri Barak (opens in a new tab)
Herzog Fox & Neeman
Lawyer or legal adviser · Team size: More than 50 · Used for 1 to 2 years ·
- Problems solved
- Due-diligence - what used to be a tedious multi-day process now takes only hours with the Vault feature; Document benchmarking - reviewing and redlining documents against benchmarks. This isn't a unique feature or something that Claude or ChatGPT cannot do but Harvey is more cost-efficient for the heavy law firm user due to seat-based pricing, and they have a great interface; Quick work through multiple tasks, in parallel - both are edges. I run multiple workflows, projects and tasks concurrently. While human in the loop is still necessary for the final QA, it saves time and increases efficiency. Harvey also tends to be quite quick, even while working on complex tasks. Seat-based pricing - While using Claude with Opus 5.5 on a Team/Enterprise plan might get similar or better results (putting Vaults aside), you'd quickly run out of tokens while Harvey keeps going. It might change with a move to full or partial usage-based pricing. Playbooks - allows preserving knowledge and preferences and replicating them in repeatable workflows across the same document type or use case. It isn't a Harvey unique feature, but is highly useful for my needs as a daily user. Overall, it's a great product and it feels like having another, highly capable member on the team.
- What could improve
- Better agent/playbook/workflow builder - currently a bit rigid and not as intuitive as Claude Access to stronger models - for obvious cost saving reasons, the topmost models one can currently access on Harvey are Claude Opus 4.8 and GPT Sol 5.6. Since Harvey is in a way subsidizing tokens (and I've heard they'll move to usage-based pricing next year), this is understandable but still would be great to be able to access at least the previous generation of high-end models (Opus 5.0, Sol 6.0). Output too long and granular - out of the box, Harvey's outputs - whether when drafting or revising contract clauses or when replying to queries - are way too long and "legalese". When representing startups v. large enterprises, where every redline matters and less is more, and time is the most precious resource, a lot of time is spent on trimming down Harvey's proposed language or dropping suggested language which isn't critical. I know that the product itself already has features to deal with it, like Memory and writing styles, and this can also be addressed via workflows and playbooks, and not all clients are the same so better to be more conversative, but in future versions Harvey might consider going for shorter, more to the point style from the get-go.
