Small and mid-sized law firms have always competed against bigger rivals with one hand tied behind their backs. Big Law throws armies of associates, paralegals, and e-discovery vendors at a matter; a ten-lawyer shop throws whoever is free after 6 p.m. Agentic AI — software that doesn’t just answer questions but plans, executes, and follows through on multi-step work — is the first technology that genuinely levels that field. But a recent federal ruling, United States v. Heppner, shows exactly how not to use it. This article covers what agentic AI can do for your firm in the courtroom and the office, and the practical steps that keep attorney-client privilege intact while you do it.
What “Agentic” AI Actually Means for a Law Firm
A chatbot answers a prompt. An agent takes a goal — “prepare me for Thursday’s deposition” — and breaks it into steps: pull the pleadings, summarize the witness’s prior statements, flag inconsistencies, draft an outline of questions, and check the local rules for exhibit deadlines. Modern agentic systems can browse documents, call other software, draft work product, and loop until the job is done, with a human reviewing at the checkpoints that matter. For a small firm, that’s the equivalent of hiring a tireless junior associate who works in seconds and never bills overtime — as long as you deploy it correctly.
In the Courtroom: Where Agents Earn Their Keep
1. Trial prep that used to take a weekend, done by lunch
Feed an agent the case file and it can build a chronology of events, a cast-of-characters memo, and a theme-and-theory outline — then pressure-test your theory by role-playing opposing counsel and attacking it. Before oral argument, it can generate the fifty hardest questions a panel is likely to ask, drawn from your briefs and the cited authorities, so you walk in pre-rehearsed rather than hopeful.
2. Real-time transcript analysis and impeachment
With a live or same-day transcript feed, an agent can compare a witness’s testimony against their deposition and prior statements while they’re still on the stand, flagging contradictions with page-and-line cites. What once required a second chair with a photographic memory becomes a sidebar note: “Deposition p. 112, line 8 — he said the opposite.”
3. Instant exhibit retrieval
Mid-cross, you don’t have three minutes to dig through a banker box. An agent over your exhibit database retrieves “the email where the defendant acknowledges the debt” in seconds, with the Bates number ready. Small firms suddenly look like they brought a war room.
4. Voir dire and jury research
Agents can organize juror questionnaires, flag biases against your profile criteria, and keep a running strike sheet updated in real time — work that large firms outsource to jury consultants at five figures a day.
5. Witness preparation
Agents make relentless mock-cross examiners. They can run a witness through hostile questioning, review the video or transcript afterward, and produce a coaching memo on evasive answers and inconsistencies — letting your actual witnesses practice against something that never gets tired or polite.
In the Office: The Compounding Wins
1. Client intake that never sleeps
An intake agent can run conflict checks against your matter database, collect facts through a structured interview, score the matter against your acceptance criteria, and book the consultation — before a prospect who called at 9 p.m. phones your competitor in the morning.
2. Drafting and document review
First drafts of engagement letters, discovery requests, motions, and closing binders assembled from your own templates and matter files. In diligence and e-discovery, agents can triage tens of thousands of documents, cluster them by issue, and surface the hot documents for human review — work that used to price small firms out of document-heavy cases entirely.
3. Deadline and docket management
Agents watching your docket can extract every date from a scheduling order, compute rule-based deadlines, calendar them, and nag the responsible attorney until each one is confirmed. Malpractice carriers should pay you to do this.
4. Billing and collections
Agents that draft time entries from your activity, audit invoices against client billing guidelines, and run polite-but-persistent collections sequences routinely recover 5–10% of revenue small firms currently write off or never capture.
5. Client communication
Status-update agents that pull matter milestones and draft plain-English client updates turn “my lawyer never calls me back” into five-star reviews — the cheapest marketing a small firm can buy.
The Heppner Problem: How One Defendant Lost Privilege Over His Own Strategy Notes
In February 2026, Judge Jed Rakoff of the Southern District of New York decided a question of first impression in United States v. Heppner. Bradley Heppner, facing a fraud indictment, had used Claude — a consumer AI chatbot — on his own initiative to prepare roughly thirty-one documents outlining his defense strategy, using information he had learned from his attorneys. He shared the outputs with his lawyers, and they influenced counsel’s strategy. When the FBI seized his devices, the government moved to strip those documents of protection. The court agreed: no attorney-client privilege, no work-product protection.
The reasoning should be taped to every lawyer’s monitor:
- An AI is not your attorney. Communications between a client and a chatbot are not communications between client and counsel.
- No reasonable expectation of confidentiality. Heppner had consented to Anthropic’s privacy policy, under which user inputs and outputs could be collected, used for training, and disclosed to third parties — including the government. Typing privileged facts into a consumer chatbot, the court found, is like saying them in front of a third party.
- Not at counsel’s direction. Because Heppner used the tool on his own initiative rather than at his lawyers’ direction, the documents were neither prepared to obtain legal advice nor work product prepared at the behest of counsel. The court pointedly noted it could have been different if counsel had directed the use — the classic Kovel doctrine that protects a client’s communications with an attorney’s agents, like accountants and interpreters.
The uncomfortable corollary for firms: if one of your clients is pasting the facts you gave them into ChatGPT at home, they may be waiving privilege over your own advice. And if your associates are pasting client confidences into consumer AI tools, the firm may be doing the waiving.
How Small Firms Get the Benefits Without the Waiver
The good news: Heppner reads less like a ban and more like an instruction manual. Six rules cover nearly all of it:
- Counsel directs the use — on paper. AI analysis of client matter should be initiated and directed by the attorney, and the engagement letter or matter file should say so. That’s the Kovel lane the Heppner court explicitly left open: the AI functions as the lawyer’s agent, within the privilege, the way an accountant or interpreter does.
- Never put client confidences into consumer-tier chatbots. The confidentiality problem in Heppner flowed from the provider’s privacy policy: data collection, training on inputs, and third-party disclosure rights. Consumer free tiers are where privilege goes to die.
- Use enterprise or API tiers with contractual protections. Every major provider offers business arrangements with zero data retention, no training on your data, and confidentiality commitments. Get it in writing, file the DPA, and re-verify it when policies change — providers update them every few months.
- Consider private deployment for the most sensitive work. Open-weight models running on your own infrastructure (or a locked-down private cloud) keep client data inside your perimeter entirely. For criminal defense, family law, and trade-secret work, this is increasingly the belt-and-suspenders answer.
- Adopt a written AI use policy — for staff and clients. Define approved tools, prohibited inputs, and mandatory human review. Then educate your clients: “Please don’t discuss our advice with any chatbot” should be standard engagement-letter language now, because a client’s well-meaning DIY strategy session can waive privilege over your work.
- Keep the lawyer in the loop and verify everything. Privilege aside, courts have sanctioned lawyers for hallucinated citations. Agentic output is a first draft and a research accelerant — the signature on the filing, and the judgment behind it, stay human.
The Bottom Line
Agentic AI lets a fifteen-lawyer firm prepare for trial like a hundred-lawyer firm: faster research, live impeachment support, instant exhibits, tireless drafting, and an office that runs itself in the background. Heppner doesn’t stand in the way of any of that — it simply draws the boundary. Use AI at counsel’s direction, through contractually confidential channels, with human judgment at every checkpoint, and the privilege follows you into the machine. Ignore those lines, and your strategy memos are one subpoena away from the government’s exhibit list.
This article is for informational purposes and is not legal advice.
