It's late, a solo attorney is staring at a huge discovery batch, and there's a hearing in three days. A few years ago that meant Red Bull and a paralegal working the weekend. Now the same lawyer runs the batch through a document intelligence tool, gets a first-pass privilege log by midnight, and spends the next morning reading the twenty documents that matter.
The client never sees the tool. The client sees a lawyer who is calmer, sharper, and ready earlier.
That late-night batch is the thread running through this piece. The question small firms keep getting wrong isn't whether to use AI. It's where to let AI into the work and where to keep it firmly out. Cross the line in the wrong direction and you either burn out doing work a machine could do, or you file a brief citing cases that don't exist. Both are avoidable.
The Late-Night Batch Is Where AI Earns Its Keep
Document review at volume is the clearest win for a small firm. The work is repetitive and pattern-based, and the lawyer's job is to catch the exceptions: a privileged email that got miscoded, the smoking-gun attachment, a date that contradicts the timeline. A well-configured tool does the sorting; the attorney does the judging. Fair trade.
The same logic holds for intake triage, calendar coordination, deposition transcript summaries, and first-pass contract redlines against a known playbook. None of those tasks require the lawyer's judgment on the first pass. All of them punish a small firm that insists on doing them by hand.
Adoption reflects this. Recent surveys show that nearly seven in ten legal professionals now use generative AI tools for work, a figure that more than doubled in a single year, and yet most firms still lack formal AI policies or training programs. That second half is the problem.
Individual lawyers are running ahead of their firms, using tools the firm hasn't put any guardrails around.
The Same Batch, One Step Further, Is Where It Gets Dangerous
Rewind that late night. Suppose the lawyer, riding the momentum, asks the same tool to draft the reply brief and cite the controlling authority. That's the move that keeps getting attorneys sanctioned.
The most-cited example is still the Avianca matter, where a New York attorney filed a brief built on six ChatGPT-generated cases that did not exist. That was 2023. The docket has only grown since.
The pattern is consistent enough now that reporters track it as a running beat rather than a novelty. Courts are revoking pro hac vice admissions and making bar referrals over fabricated citations. The tool didn't sign the brief. The lawyer did.
So the human line, on the same batch of documents, sits between sorting and asserting. A model can propose. Only the attorney can vouch.
Draw the Line Before You Draft the Prompt
The firms that use AI well decide, before a matter starts, which tasks the tool is allowed to touch and which it isn't. Written down, shared with staff, revisited quarterly. A few practical divisions worth stealing:
The Client on the Other End Still Wants a Person
Efficiency inside the firm is invisible to the client. What they notice is the phone call. And on the phone, they still want a human being. As AI improves, that preference is hardening rather than softening.
Small firms that lean into the human channels, especially at intake and during hard conversations, hold onto clients who otherwise shop around on price. That doesn't mean the answering service has to be a person at 2am. It means the person is available when it counts, and the automation exists so that person isn't drowning in the tasks a machine could have handled.
Peyman Khosravani makes a similar point in an IntelligentHQ piece on how how AI gives a small law firm an edge gives a small law firm an edge without turning the practice into something clients don't recognize. Use the tool to buy back the hours. Spend those hours on the client.
