Most small business owners assume that hiring a lawyer in 2026 looks the same as hiring one in 2016: a rate sheet, a retainer, and a promise that a human will read every page. The reality has flipped. Behind the engagement letter you're about to sign, a mix of software, private language models, and outside vendors is probably touching your matter, and the letter itself rarely says so.

None of that means the firm is a bad choice. It does mean you should ask sharper questions before you sign. The old engagement letter and the new one are doing very different jobs, and knowing where they diverge is what separates a client who gets the upside of AI-driven legal work from one who inherits its risks.

Document Review and Discovery Are Where the Money Moves

Discovery used to be the single largest line item in most litigation budgets, and it was mostly bodies. Junior lawyers coded documents by the thousand while the meter ran. AI-assisted review has compressed that work meaningfully. A task that took a team a month can now take a small group a week, with the machine surfacing the responsive material and humans confirming the calls.

The savings are real. Whether they land on your invoice is a separate question. Some firms pass the efficiency through as a lower bill. Others hold rates steady and bank the margin.

Two questions belong in writing before you sign:

• Which discovery and review tasks the firm now handles with AI, and how that changes the hours billed compared with a fully manual workflow.
• Whether the efficiency shows up as a lower fee, a flat rate, or a cap on the discovery budget, and, if not, why the client rate is unchanged.

Drafting With a Machine Changes Who Is Actually Writing

Drafting is the other place AI has moved fastest. First drafts of NDAs, employment agreements, demand letters, and routine motions are increasingly produced by a model trained on the firm's own precedent, then edited by a lawyer. Done well, you get a tighter draft faster, and the lawyer's time goes to the parts that need judgment.

Done poorly, you get a document that reads fluently and is wrong in a way nobody caught. Courts have already sanctioned lawyers for filing AI-generated work without checking it. The safeguard is a verification standard the firm will commit to on paper: which categories of documents get a full human read, who signs off, and what happens when the model cites authority that turns out not to exist.

Confidentiality Is the Question the Old Letter Never Had to Answer

Nothing about AI matters if the confidentiality story doesn't hold. When your matter runs through a language model, someone is deciding whether your data trains that model, where it's stored, and who else can see it. California's ethics guidance is direct: lawyers should consult IT experts on AI security and should never feed confidential client information into tools that use inputs for training, per a Bar Association of San Francisco summary of the state's COPRAC advice.

Privilege is the harder edge. In a recent federal decision analyzed by Ogletree Deakins, a court found that sharing material with a third-party AI platform waived work-product protection, and the firm recommends that engagement letters and litigation-hold notices name AI platforms explicitly. That is exactly the kind of clause a small business client should ask for.

Ask the firm to name every AI tool that will touch your matter, describe where your data lives, and confirm in writing that your inputs are not used for model training. If the answer is vague, the risk is yours.

The Economics Only Work If You Negotiate for Them

There's a reason firms are investing heavily in this shift. A recent overview from Law.co on the shift toward legal AI on the shift toward legal AI lays out how much of the routine work, including review, summarization, extraction, and first-pass drafting, can now be handled by software with a lawyer in an approval role. That's a genuine productivity gain. It is also a pricing conversation the client has to start.

Small business clients who never raise the topic tend to keep paying the old rate for the new workflow. Clients who ask specific questions, such as flat fees for defined tasks, capped budgets for discovery, a written verification protocol, and a named list of AI tools in use, tend to get better terms and a cleaner record if something goes wrong later.

The engagement letter is the one moment when the client holds the pen. Use it before you sign, not after the first invoice arrives.