If you run a small firm or a solo practice, you already know the real bottleneck isn’t legal skill — it’s everything around it. The intake calls, the first-draft engagement letters, the calendar math, the invoices that quietly go unpaid. This is exactly the kind of repetitive, deadline-driven work that AI agents are built to handle. But law is also a field where a careless tool can get you sanctioned, so it’s worth being precise about where agents earn their keep and where they don’t belong.

If you’re new to the term, an “agent” is just software that can take a multi-step task off your plate — not merely answer a question, but actually do the sequence of steps. We covered the broader shift in The Agentic Wave Is Here — and It Is Not Just for Tech Companies.

Where agents genuinely help

Client intake and screening. An intake agent can run your website chat or after-hours line, ask the standard qualifying questions, flag conflicts, and hand you a tidy summary before you ever pick up the phone. Think of it as a very organized receptionist who never sleeps — not a lawyer. The hard line: it can gather information and route it, but it must not give legal advice. That distinction matters legally, which we’ll get to. For a deeper dive into the document side of intake — extracting names, dates, and clauses from retainers and discovery — see Why solo and small firms are the right target.

First-draft document work. Retainer agreements, demand letters, discovery request shells, routine correspondence — an agent can produce a solid first draft from your templates in seconds. You still edit and sign. Used this way, it’s a faster typewriter, not a substitute for your judgment.

Deadline and calendar management. An agent can read incoming court notices and scheduling orders, calculate response deadlines, and drop them into your calendar with reminders. For a solo juggling twenty matters, catching a missed deadline is arguably the highest-value thing AI can do all week — as long as you spot-check the math against the rules.

Billing follow-up. Getting paid is where small firms bleed. An agent can track outstanding invoices, send polite, on-brand reminders on a schedule, and escalate the ones that go cold — the dull, awkward chasing that partners rarely do consistently.

If you want a feel for how to separate real wins from wishful thinking in a service business, our companion piece AI Agents for Realtors: Where They Actually Help walks through the same “help vs. hype” filter for a different profession.

The limits you cannot hand-wave

Hallucinations are not a rare glitch. AI tools invent citations that look real. In the now-famous Mata v. Avianca case, lawyers Peter LoDuca and Steven Schwartz submitted a filing containing fake cases that ChatGPT had fabricated. In May 2023, Judge P. Kevin Castel dismissed the case and ordered the plaintiff’s attorneys to pay a $5,000 fine, noting numerous inconsistencies and describing one legal analysis as “gibberish.” This isn’t a one-off. A legal researcher’s public tracker of court decisions involving AI-hallucinated material has identified more than 1,600 such cases so far, and it keeps growing. Never let an agent’s cite reach a court unverified.

Confidentiality is your problem, not the vendor’s. In July 2024, the ABA issued its first formal ethics guidance on generative AI, Formal Opinion 512. It states that lawyers and firms using these tools must fully consider their ethical obligations, including duties to provide competent representation, protect client information, communicate with clients, and charge reasonable fees. On confidentiality specifically, lawyers are responsible for knowing how a tool uses data and for securing clients’ informed consent before feeding client confidences into it — and the opinion says boilerplate consent buried in an engagement letter will not be adequate. Translation: don’t paste privileged material into a free consumer chatbot.

Unauthorized practice of law. This is the bright line for client-facing agents. Allowing AI to provide legal advice directly to clients without attorney review constitutes unauthorized practice of law, and submitting AI output to courts without verification violates candor obligations. You remain on the hook for supervision — Model Rule 5.3 places responsibility on the supervising attorney for the conduct of non-lawyer assistants, including AI tools.

A low-risk first step you can try this week

Pick the one task with the least confidentiality exposure and the most repetition: billing reminders or intake summaries. Take five real (but anonymized) past intake calls, paste your notes into a reputable AI tool, and ask it to produce a clean one-page summary in your preferred format. Compare its output to what you’d have written. You’ll learn its strengths and its blind spots in an afternoon — with zero risk to a live matter.

Start there. Once you trust the output on low-stakes work, you can decide, deliberately and with your ethics rules open, where to let an agent do more.