Attorney Supervision of AI: How the Review Chain Works
Why every draft an agent produces waits for a licensed lawyer, and what to check before you adopt it.
Attorney supervision of AI works when the review step is built into the workflow itself, so no draft leaves the firm until a licensed lawyer has read it and adopted it. An agent produces the work product, whether that is a structured intake record or a demand letter draft, and then it waits. A lawyer decides what happens next, every time.
That answer sounds obvious until you look at how much legal software skips the step. Tools that auto-send follow-ups, or push a generated letter straight to an adjuster's inbox, put your name on work you never saw. For a solo or small personal-injury firm, one bad paragraph in one unreviewed letter costs more than the tool ever saved.
Nimbus Marlowe is the AI operations layer for personal-injury law firms, starting with intake. It sits in front of the practice-management system you already run, and every agent inside it, from the intake coordinator to collections follow-up, delivers output to a review queue rather than to the outside world. This post walks through how that chain works across a small PI firm's week, and what to look for before you adopt a draft as your own.
What does attorney supervision of AI look like in practice?
Start with a Tuesday night. The intake coordinator answers the 9pm call from a car-accident lead, asks the qualifying questions your firm wrote, and books a consult on your calendar. What lands on your desk Wednesday morning is a structured intake record: who called, what happened, when it happened, where treatment stands, which carrier is involved. The agent did not decide whether to take the case, because it cannot. You read the record and run your own conflicts check before deciding anything.
The same shape repeats across the layer. When the contract analyst reads an agreement, its flags show up as annotations you can accept or dismiss, and a dismissed flag changes nothing in the document. Case preparation hands you an organized package rather than a decision. Demand letter drafts sit in your queue until you have worked through them line by line, and the collections reminder waits for your sign-off before any client sees it.
Every agent output is a draft. It becomes the firm's work only when a licensed attorney reviews it and adopts it.
Picture a two-attorney shop in Fort Worth, or a three-lawyer firm north of Atlanta, running thirty active matters. The review queue becomes the morning ritual: coffee, then fifteen minutes moving through overnight intake records and pending drafts. Nothing that stacked up overnight went anywhere on its own. The agents did the assembly work in the dark, and the morning belongs to the judgment calls only a lawyer can make.
Why does the review step belong in the software?
Because a step that lives in the terms of service gets skipped on a busy Friday. Plenty of vendors write review language into their disclaimers, then ship an auto-send button that makes skipping the default. Nimbus removed the temptation at the design level. When the demand letter drafter finishes, there is no send option on the agent's side of the screen. The draft opens with its source documents beside it, and the only path forward runs through your edits and your adoption.
That structure also protects you against tools that hide the step. If a product markets itself on how little you have to touch the output, ask where the adoption moment lives. A lawyer who cannot point to the moment they took responsibility for a document has lent out their bar card, whatever the marketing said. Responsible AI for law firms starts with a review moment the software refuses to skip.
What should a lawyer check when reviewing an AI draft?
Lawyer review of AI work product is a skill, and it sharpens fast once you know where drafts tend to go wrong. When a demand letter draft or an intake record hits your queue, read it against the file and work through a short list.
- •Names, dates, policy numbers, and provider details, checked against the source documents. Small transcription slips are the most common edit you will make.
- •Every factual assertion. A statement the file cannot support gets cut, or the missing document gets requested.
- •Anything that reads like a legal conclusion or a valuation. Those judgments belong to you, and a draft that wanders into them needs a heavier hand.
- •Tone for the specific recipient, since a carrier you have negotiated with for a decade may call for a different register than the default.
- •What is missing. The agent drafts from the documents in the matter, so a thin file produces a thin draft, and the gaps tell you what to request next.
The agents stay out of the legal questions entirely. Whether a claim has merit is your judgment, and the same goes for conflicts checks and for every date on the calendar. The software handles the drafting and the organizing, and the decisions stay with the person holding the license.
What do the ethics duties around nonlawyer assistance require?
AI legal ethics for small firms rests on a duty most PI lawyers already carry: when nonlawyers assist on client matters, the lawyer makes reasonable efforts to ensure that assistance stays compatible with the lawyer's own professional obligations. Bars have applied that principle to paralegals and outside answering services for decades, and software that drafts a letter sits in the same seat. The signature at the bottom carries the responsibility, no matter who or what prepared the first version. None of this is legal advice about your jurisdiction, and your own bar's guidance controls.
Those duties get easier to meet when the tool keeps a record. Inside Nimbus, every draft carries its source documents with it, and the version that leaves the firm is the version you adopted. If anyone ever asks how a letter was prepared, the honest answer holds up: an assistant produced a draft, and the attorney of record revised and adopted it.
Where does the cost of supervised drafting land?
Flat per-unit pricing keeps the review economics simple. A qualified intake runs about $18, a contract review costs $65, case preparation is $95, a demand letter draft is $170, and collections follow-up carries its own flat per-unit rate. You know the cost before the work starts, and the price never moves with the size of a recovery. Compare that with part-time intake help or a legal assistant at $3,000 to $4,000 per month, before anyone has produced a single reviewable draft.
The CRM underneath is free at $0 forever, with contacts, matters, documents, and a client portal included, so starting costs nothing. Firms that want discounted per-unit rates can commit at a $500 per month minimum, settled as a simple true-up. You never prepay, and there is no balance sitting somewhere waiting to expire.
No Nimbus fee is ever a percentage of a recovery, and the company never holds or transmits client funds.
Frequently asked questions
Does the AI ever send anything without attorney review?
No. Every agent output, from an intake record to a demand letter draft, lands in a review queue where a licensed attorney edits and adopts it before anything moves. The software has no path around that step.
Does Nimbus take a percentage of settlements?
Never. Every rate is flat and published, and it stays identical regardless of the outcome of the case. Nimbus also never holds or transmits client funds.
Does the AI give legal advice or evaluate my cases?
No. The agents handle operational work, like answering an intake call at midnight or preparing a demand letter draft for your queue. Every legal judgment, including whether a claim has merit, stays with the attorney.
Does Nimbus replace my practice-management system?
No. Nimbus is an operations layer that sits in front of the system you already use. Your matters stay where they are, and the agents feed drafts and records into a review queue that fits your existing workflow.
The free CRM costs $0 forever, so the fastest way to judge the review chain is to run one matter through it yourself. Sign up now and start free, or book a call to watch an intake record travel from agent to attorney on a live screen.
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.