Every mid-sized law firm lives with the same bind. Clients want sophisticated work, fast, with AI used responsibly. The team is excellent but finite. Hiring ahead of demand is a financial risk; waiting for the work to arrive stretches partners and associates thin.
Harvey, an AI platform built for legal work, published a short guide on how mid-sized firms are getting out of that bind. The thesis fits in one line: a firm’s capacity no longer has to scale in lockstep with its headcount. Here is our read, and what applies to any firm, whatever tool it runs.
Where this comes from
Start with one workflow, not an “AI rollout”
The guide’s most useful advice comes before any tool: pick one workflow the firm already runs every week, apply AI to a live matter and compare against the old way. The first question is which pain shows up most:
- Thousands of pages to digest before a deadline? Start with large-document review.
- Hours assembling a first draft before senior or client review? Start with drafting.
- A data room that eats time and margin? Start with diligence and structured extraction.
- A lean team taking a new matter from intake to the first filing? Start with building the matter in connected stages.
What happened at the firms
The examples come from firms of 55 to 160 lawyers in the US and Italy. A few numbers stand out:
- Bankruptcy, under 24 hours. A filing landed at 3 a.m. ahead of a 1 p.m. hearing, over a holiday weekend with most of the team out. With every matter document in one place, the partner identified the relief affecting her client, drafted the objection and prepared the cross of the debtor’s CFO. The objection was on file an hour before the hearing.
- Two days down to two hours. A corporate partner loaded 50-plus closing documents to pull every restrictive covenant in a sale with an unusual structure.
- A week and a half for three associates, done in a day. A general counsel wary of AI agreed to one diligence test. Accuracy came in above 90%, and the client now uses AI in-house.
- Six weeks of team work, done by one partner. On a contingency-fee matter, a partner used AI from early document review through the complaint, including the timeline and venue analysis.
The immediate gain is time. The gain that matters is capacity: work that used to be hard to staff or price starts to fit.
The attorney still owns the work
In none of the four workflows does AI sign anything. It prepares, organizes, compares and flags discrepancies; the attorney picks the precedent, sets the instructions, reviews the substance and owns the result. Everything comes back linked to source, so every conclusion can be checked.
One partner built trust in a simple way: he asked the tool to critique a motion he had already argued and knew cold. Testing on work you know well is the best way to calibrate how far to trust it.
The survey numbers
The guide closes with June 2026 research by RSGI on firms using Harvey:
Reported time savings run from three to five hours a week for light users to more than ten for the heaviest. The guide itself says these figures are context, not a guarantee. We agree: the real business case comes from the firm’s own work.
Proven? Make it repeatable
The last step is what separates a test from a change in how the firm works. At one firm, the partner in charge of innovation built agents alongside each practice group leader and rolled them out firmwide. One partner’s way of handling a document set becomes a workflow the whole group runs, and the time recovered compounds instead of staying with one matter.
- Pick the pain. A workflow that repeats every week.
- Write down the baseline. How long it takes today, how many review rounds, how many handoffs.
- Run it on a live matter. With a question the team already needs answered.
- Compare and decide. If the gain shows up, turn the workflow into an agent for the practice.
What this means for your firm
The guide is Harvey’s material, but the design doesn’t depend on any one tool. The same workflows run on Claude, which many firms already use, or on agents hosted inside the firm’s own environment when confidentiality calls for it. And the four workflows are examples: docket and deadline tracking, pre-bills, client reporting and plenty of other repeat tasks belong on the list too.
- Our AI Analyst does the first read: maps the firm’s workflows, sets the baseline and points to the first pilot.
- Our Forward Deployed Engineer works inside your operation and builds the agents with each practice leader, with an attorney approving everything that goes out.
- Side-by-side training gets partners and associates ready to use and look after these workflows day to day.
One workflow, one live matter, one baseline. That’s how a mid-sized firm starts taking on more without hiring ahead of demand.
Enjoyed this article?
Get the next one before it hits the blog.
A short email per issue with what actually matters about AI agents at work. No spam.