The Risk of AI in Law Firms Isn't the AI. It's Supervision.
Why the firms most worried about AI are worried about the wrong thing.
The biggest risk of AI in law firms is not the technology. It is using AI output that no licensed attorney reviewed before it reached a client or a file. The AI itself is not the danger. The danger is the shop that lets a machine draft a demand letter, qualify a claim, or answer a client and then ships that work without a lawyer ever looking at it. Get supervision right and the tool is an asset. Get it wrong and no vendor logo will save you.
Most of the worry about AI in law firms is aimed at the wrong target. Attorneys ask whether the model is smart enough, whether it hallucinates, whether it can be trusted. Those are fair questions. But they miss the one that actually decides your exposure, which is who reviews the output before it counts. A supervised draft from an average tool beats an unsupervised draft from a brilliant one every time.
Why is everyone asking the wrong question about AI in law firms?
The popular framing treats AI as a replacement that either works or does not. Either the software is good enough to trust on its own, or it is too risky to touch. That is a false choice, and it is the reason so many small personal injury firms freeze on the sidelines while their intake still goes to voicemail at night.
The useful framing is narrower. AI in a law firm should never be the thing that decides. It should be the thing that drafts, so the attorney spends judgment where judgment belongs and stops spending hours on the first ninety percent of a document. The question is not whether you trust the machine. The question is whether a human stays in the loop on everything that leaves the building.
Framed that way, the fear dissolves into a design problem. You do not need software you can blindly trust. You need software that hands finished work to a person who checks it, and a workflow that makes skipping that check impossible.
What actually goes wrong when AI runs unsupervised?
The failure mode is rarely a dramatic robot mistake. It is quieter than that. An unsupervised system sends a client a confident answer that happens to be wrong for their matter. It drafts a demand and the numbers never get a second read. It tells a caller something a lawyer would have caught in five seconds. The technology did its job. The process had no brakes.
This is why the supervision chain matters more than the model. A licensed attorney carries the duty of competence and the duty to supervise nonlawyer assistance, and neither duty disappears because the assistance is software. If anything the obligation gets sharper. The output is faster and there is more of it, so the review step is the only thing standing between speed and a problem.
The firms that get burned are not the ones using AI. They are the ones using it as an oracle instead of an assistant.
What does supervised AI look like in a personal injury firm?
It looks like a set of narrow roles that each produce a draft, and an attorney who adopts, edits, or rejects that draft before it becomes real. The work moves faster because the blank page is gone. The judgment stays human because nothing skips the review.
- •The intake coordinator answers the 9pm call from a car accident lead, gathers the facts, and books the consult. It does not decide whether you take the case. It hands you a qualified lead and a clean record so a person makes that call.
- •The demand letter drafter turns the file into a complete draft. It does not send anything. The attorney reads it, fixes what the facts require, and signs off. Nothing goes out unreviewed.
- •The contract analyst reviews an agreement and flags the terms that matter, so the lawyer reads a shortlist instead of forty pages cold. The flags are a starting point for review, not a verdict.
- •Case preparation assembles what the team needs to move the matter forward, organized and ready for a human to work from.
- •Collections follow-up keeps post resolution outreach running on schedule so nothing quietly slips, with the firm in control of what gets said.
Notice what none of these do. They do not give legal advice, decide whether a claim has merit, resolve conflicts, or calculate a deadline. Those are practice of law, and they stay with the attorney. The agents handle operations. The lawyer practices law and reviews everything. That division is the entire point.
Doesn't supervision cancel out the time savings?
This is the objection worth taking seriously, and the answer is no, because reviewing is not the same work as creating. A demand letter that sits half drafted for two weeks is not slow because signing it is hard. It is slow because starting it is hard, and because the person who has to start it is buried. Hand that person a complete draft and the two week bottleneck becomes a same day read.
The same holds across the firm. Reading a shortlist of flagged contract terms is faster than reading the whole contract. Reviewing a qualified intake summary is faster than fielding every call yourself. Supervision is where the leverage lives, not where it dies. You are trading hours of production for minutes of judgment, and the judgment is the part clients actually pay a lawyer for.
How should a small firm judge an AI vendor on supervision?
Ask three questions before you sign anything. The answers separate a real operations layer from a black box.
- •Does every output land as a draft a human reviews, or does the system act on its own? If it acts on its own, walk.
- •Can you see exactly what it did and why, so an attorney can supervise it the way the rules require? Opaque is a liability, not a feature.
- •Is the pricing published and flat, or is it a quote and a percentage of your recovery? A vendor who wants a cut of the fee has an interest in the outcome. A vendor charging a flat, published rate per completed unit of work does not.
That last one matters more than it looks. Nimbus charges flat, published rates per unit of work, the same regardless of how a matter resolves. Qualified intake runs about $18. Contract review is $65. A demand letter draft is $170. The CRM itself is free and never invoiced. No percentage of any recovery, no opaque quote, and the company never holds or transmits client funds. Compare that to a part time intake specialist at $3,000 to $4,000 a month who still works business hours.
The safest AI in a law firm is the AI that never gets the last word.
The firms that will win with this are not the ones who trust the software most. They are the ones who wire a licensed attorney into every step and let the tools carry the volume underneath. If that is the firm you want to run, you can sign up now and start on the free CRM, or see it on a live call and watch the review chain work end to end.
Frequently asked questions
Does Nimbus give legal advice or decide whether a case has merit?
No. The agents handle operations like answering calls, drafting, and follow up. Every output is a draft. A licensed attorney reviews it, and the attorney is the one who advises clients, evaluates merit, and makes every legal decision.
Does Nimbus take a percentage of settlements?
Never. Every fee is a flat, published rate per completed unit of work, identical regardless of how a matter resolves. The company never holds, transmits, or takes custody of client funds, so no one has a hand in your recovery.
Can one attorney realistically supervise all of this?
Yes, because supervision means reviewing finished drafts, not producing them from scratch. Reading and adopting a complete draft takes minutes where creating it took hours, which is exactly why the model frees an attorney's time instead of consuming it.
What does it cost to start?
The CRM is free and never invoiced. Beyond that you pay flat published rates only for the work you use, such as about $18 for qualified intake or $170 for a demand letter draft, with no seats and no minimum on pay as you go.
Put your firm's operations on autopilot. Reviewed by you.
Nimbus Marlowe is the AI operations layer for personal-injury firms. The CRM is free, pricing is flat and per-unit, and every output is a draft your attorneys review and adopt.