The fastest way to turn an AI experiment into a malpractice claim is to give it open-ended authority and hope for the best. A general-purpose legal assistant that can “help with anything” is a liability waiting to happen. A document-drafting agent that follows a 12-step checklist, flags exceptions, and stops when it hits a boundary is a tool you can defend in front of the bar.

The difference isn’t the underlying model. It’s the mandate you give it.

Law firms adopting AI right now face a choice. You can deploy agents that do narrow, rule-bound work with explicit guardrails, or you can hand over broad discretion and cross your fingers. The first approach scales your team without adding risk. The second one scales your exposure.

Most partners I talk to want the productivity gain but can’t stomach the idea of an AI making judgment calls on client matters. They’re right to worry. The solution isn’t to avoid AI. It’s to design mandates so narrow that the agent can’t wander into trouble.

Why Open-Ended AI Mandates Fail in Legal Practice

A general-purpose AI assistant sounds appealing. One tool that handles intake, drafts correspondence, reviews contracts, and answers client questions. In practice, it’s a disaster.

The problem is scope creep. An agent with broad authority will eventually encounter a situation outside its training. It will guess. It will hallucinate a citation, misstate a deadline, or draft a clause that contradicts your standard terms. When that happens, you own the mistake.

Malpractice carriers are already asking pointed questions about AI use. They want to know what the tool does, who supervises it, and what happens when it gets something wrong. “We use an AI assistant for general legal work” is not an answer that inspires confidence.

Compare that to a narrow mandate. A document review agent that performs first-pass review on discovery batches, flags privileged material using a predefined taxonomy, and produces a memo for associate review. The agent doesn’t make judgment calls. It follows a checklist. It stops when it hits an exception and escalates to a human.

That’s defensible. You can document the process, audit the output, and show the bar that you maintained supervision. The agent didn’t practice law. It performed a bounded clerical task under attorney oversight.

Firms leak between $80,000 and $250,000 a year to inefficiency in document review, intake delays, and unbilled admin time. Most of that leakage sits in work that’s repetitive, high-volume, and rule-bound. Perfect for a narrow-mandate agent. Terrible for a general-purpose one.

The Three Mandates That Work

We build AI agents for law firms every month. The ones that stick share a pattern. They do one thing, they do it the same way every time, and they hand off when they reach the edge of their authority.

Here are the three mandates that consistently deliver ROI without adding risk.

Intake Voice Agent

An Intake Voice Agent answers every call that comes into the firm. After-hours, lunch, weekends. It doesn’t matter. The agent picks up, introduces itself, and walks the caller through a conflict check and intake script.

The mandate is narrow. Capture the caller’s name, matter type, and contact details. Check for conflicts using the firm’s database. If the matter fits, book a consultation directly into the partner’s calendar. If it doesn’t, provide a referral and end the call.

The agent doesn’t give legal advice. It doesn’t assess the merits of the case. It doesn’t promise outcomes. It follows a script that you write, and it stops when the script ends.

Most firms lose 30 to 40 percent of after-hours intake because no one picks up. Those calls go to the next firm on the search results. An Intake Voice Agent captures that revenue without adding headcount.

One litigation boutique in our network ran the numbers. They were missing an average of 18 calls a week outside business hours. Conversion rate on those calls, when they did connect, was around 22 percent. Average matter value was $8,500. That’s $33,000 a month walking out the door because no one answered the phone.

They deployed an Intake Voice Agent in October. It’s been running for four months. Capture rate on after-hours calls is now 94 percent. Conversion rate held steady at 21 percent. They’re booking an extra $28,000 a month in new matters, and the agent cost them less than a part-time receptionist.

The mandate is the reason it works. The agent doesn’t try to be helpful beyond its script. It doesn’t improvise. It collects the information, books the meeting, and gets off the call.

If you want to see how this fits into a broader intake workflow, the AI audit for law firms walks through the full sequence in about 60 minutes.

Matter Triage Agent

A Matter Triage Agent sits between your intake forms and your partners. It reads every submission, classifies the practice area, scores fit based on your criteria, and routes the matter to the right attorney with a one-paragraph brief attached.

The mandate is classification and routing. Not evaluation. The agent doesn’t decide whether to take the case. It decides where the case should go and what information the partner needs to make a fast decision.

Here’s what that looks like in practice. A potential client fills out your contact form at 11 PM. They describe a commercial lease dispute, mention a $40,000 claim, and attach a lease agreement and three emails.

The Matter Triage Agent reads the submission, extracts key facts, identifies the practice area as commercial litigation, scores the matter as a B-fit based on claim size and complexity, and routes it to your commercial partner. The brief includes the claim amount, the opposing party, the key dates, and a two-sentence summary of the dispute.

Your partner sees that brief at 8 AM the next morning. They can decide in 90 seconds whether to take the call. If they do, they’re already up to speed. If they don’t, the agent routes a polite decline and a referral.

Without the agent, that submission sits in a shared inbox until someone has time to read it. That’s usually 24 to 48 hours. By then, the prospect has called two other firms. You’ve lost the matter before you knew it existed.

Triage agents don’t eliminate the decision. They eliminate the delay. And they do it without making judgment calls that could expose you to liability.

Document Review Agent

A Document Review Agent performs first-pass review on contracts, discovery batches, and matter files. It flags clauses, summarises positions, and produces an associate-grade memo.

The mandate is narrow and explicit. Review this document against this checklist. Flag anything that matches these criteria. Summarise findings in this format. Stop.

Let’s say you’re reviewing a 200-page commercial contract. You want to know about indemnity clauses, limitation of liability, termination rights, and non-compete provisions. You also want a summary of payment terms and dispute resolution.

You hand the contract to the Document Review Agent with a checklist. The agent reads the contract, extracts the relevant clauses, flags two indemnity provisions that shift risk to your client, notes a 90-day termination window, and produces a four-page memo.

An associate reviews the memo, verifies the flagged clauses, and adds their own analysis. Total time: 45 minutes. Without the agent, that same associate spends four hours on first-pass review before they even start writing.

The agent doesn’t decide whether the indemnity clause is acceptable. It doesn’t negotiate. It doesn’t advise. It flags and summarises. The attorney makes the call.

Document review is one of the biggest sources of unbilled time in mid-sized firms. Associates spend 200 to 400 hours a year on first-pass work that never makes it onto a client invoice. You can’t bill it because it’s too junior, but you can’t skip it because the work has to get done.

A Document Review Agent turns that unbilled time into capacity. Your associates spend their hours on analysis and strategy. The agent handles the grunt work.

Book a 60-min Omni Audit and we’ll map out which documents in your practice are ready for this kind of automation.

How to Write a Defensible Mandate

A good mandate has three parts: input, process, and output. You define all three before the agent touches a single file.

Input is what the agent receives. A phone call. A form submission. A contract. Be specific. If the agent is reviewing discovery documents, define what counts as a discovery document. If it’s triaging intake, define what fields are required before triage starts.

Process is what the agent does with the input. This is your checklist. Step one, read the document. Step two, extract clauses that match these keywords. Step three, flag exceptions. Step four, summarise findings in this format. The process should be mechanical enough that you could hand it to a paralegal and get the same result every time.

Output is what the agent produces. A memo. A calendar invite. A routed email. A flagged document. The output should never be a final decision. It should be a work product that a human reviews before it leaves the firm.

Here’s a real example. We built a Document Review Agent for a corporate firm that handles a lot of M&A work. The mandate was simple. Review purchase agreements for 14 specific risk provisions. Flag any provision that deviates from the firm’s standard language. Produce a two-page memo with excerpts and page references.

The input was a signed purchase agreement in PDF format. The process was a 14-item checklist with exact keyword matches and deviation thresholds. The output was a memo formatted as a Word document with a table of flagged provisions.

The agent doesn’t decide whether a deviation is acceptable. It doesn’t redline. It doesn’t negotiate. It flags and reports. The partner reviews the memo, decides what to do, and moves forward.

That’s a defensible mandate. You can show the bar exactly what the agent did, how it did it, and who supervised the output. You can audit the process. You can prove that no legal judgment happened without attorney oversight.

Compare that to a mandate like “review this purchase agreement and tell me if it’s a good deal.” That’s not a mandate. That’s an invitation to liability.

What Happens When the Agent Hits a Boundary

The best agents know when to stop. They don’t guess. They don’t improvise. They escalate.

An Intake Voice Agent that can’t complete a conflict check doesn’t make one up. It tells the caller, “I need to connect you with someone who can help with that,” and transfers to a human or schedules a callback.

A Matter Triage Agent that can’t classify a submission doesn’t pick a random practice area. It flags the matter as unclassified and routes it to a senior associate for manual review.

A Document Review Agent that encounters a clause it doesn’t recognise doesn’t skip it. It flags the clause as an exception and includes it in the memo with a note: “This provision doesn’t match any item on the review checklist.”

Escalation is part of the mandate. You design it in from the start. You define the boundaries, and you tell the agent what to do when it reaches one.

This is the difference between a tool and a risk. A tool that knows its limits is predictable. You can train your team on it. You can document it. You can defend it.

A tool that guesses when it doesn’t know is a time bomb.

The Practical Workflow

Here’s what a narrow-mandate agent looks like in daily practice.

Let’s say you’re a five-attorney firm doing estate planning and probate. You get 40 to 50 intake calls a week. About half come in after 5 PM or on weekends. You’re missing most of those calls, and you know it’s costing you.

You deploy an Intake Voice Agent. The mandate is: answer the call, introduce the firm, ask four qualifying questions, check for conflicts, and book a consultation if the caller qualifies. If the caller doesn’t qualify, provide a referral and end the call.

The agent goes live on a Friday afternoon. Over the weekend, it takes 11 calls. Nine qualify. The agent books consultations for Monday and Tuesday. Two don’t qualify because they’re outside your service area. The agent provides a referral and ends the call.

Monday morning, your calendar has nine new consultations. Each one includes a brief with the caller’s name, matter type, and the answers to your four qualifying questions. You didn’t lift a finger.

You take the consultations. Six convert. Average matter value is $3,200. That’s $19,200 in new business from calls you would have missed.

The agent didn’t practice law. It didn’t give advice. It didn’t make promises. It followed a script, captured information, and booked meetings. You can show that process to your malpractice carrier without breaking a sweat.

That’s the workflow. Narrow mandate, clear boundaries, human oversight. It scales your capacity without scaling your risk.

If you want a practical tool to map out your own intake process before you deploy an agent, download the AI Client Intake Checklist for Law Firms. It walks through the questions you need to answer before the first call.

Why Most Firms Get This Wrong

The mistake most firms make is trying to automate judgment instead of process.

They want an agent that can “handle client questions” or “draft correspondence” or “review contracts.” Those aren’t mandates. Those are job descriptions. And they’re job descriptions for roles that require discretion, context, and professional judgment.

An agent can’t do that work safely. It can do the mechanical steps that lead up to that work. It can capture information, flag exceptions, summarise findings, and route tasks. But it can’t make the call.

The firms that succeed with AI are the ones that break their work into two categories: mechanical and judgmental. Mechanical work follows a checklist. Judgmental work requires a licensed attorney.

Agents handle the mechanical work. Attorneys handle the judgmental work. The line between the two is sharp, and you enforce it with mandates.

Here’s a test. If you can write a checklist that a paralegal could follow to complete the task, an agent can do it. If the task requires reading between the lines, weighing competing interests, or applying professional judgment, it stays with an attorney.

Document review is mechanical up to a point. Extracting clauses, flagging keywords, and summarising terms is checklist work. Deciding whether a clause is acceptable is judgment. The agent does the first part. The attorney does the second.

Intake is mechanical up to a point. Capturing contact details, checking for conflicts, and booking a consultation is checklist work. Deciding whether to take the case is judgment. The agent does the first part. The attorney does the second.

Most firms blur that line because they want the agent to do more. They want it to be helpful, to add value, to impress the client. That’s when things go wrong.

The agent that tries to be helpful is the agent that oversteps. It answers a question it shouldn’t. It makes a promise it can’t keep. It drafts a clause that contradicts your standard terms. And you’re the one who has to explain it to the bar.

The ROI You Can Measure

Narrow mandates deliver ROI in three places: captured revenue, freed capacity, and reduced risk.

Captured revenue is the easiest to measure. It’s the intake calls you weren’t answering, the form submissions that sat for two days, the after-hours leads that went to a competitor. An Intake Voice Agent or Matter Triage Agent captures that revenue without adding headcount.

Typical numbers for a firm doing $2M to $5M a year: 15 to 25 missed intake opportunities a month, conversion rate around 20 percent, average matter value between $3,000 and $8,000. That’s $9,000 to $40,000 a month in revenue you’re leaving on the table.

Freed capacity is harder to measure but more valuable. It’s the associate hours you’re not spending on first-pass document review, the paralegal time you’re not spending on intake triage, the partner time you’re not spending on routine correspondence.

A Document Review Agent typically saves 3 to 5 hours per matter on first-pass work. If your associates are reviewing 20 matters a month, that’s 60 to 100 hours back. At $250 an hour, that’s $15,000 to $25,000 in capacity you can redeploy to billable work or new matters.

Reduced risk is the hardest to measure but the most important. It’s the malpractice claim that doesn’t happen because your agent followed a checklist instead of guessing. It’s the bar complaint that doesn’t get filed because you can document your supervision. It’s the client relationship that doesn’t sour because your intake process was fast and professional.

You can’t put a number on that until it saves you. But every partner who’s been through a malpractice claim knows what it costs. The deductible is the smallest part. The real cost is the time, the stress, and the reputational damage.

A narrow-mandate agent reduces that risk because it can’t make the kinds of mistakes that lead to claims. It doesn’t give advice. It doesn’t make promises. It doesn’t exercise judgment. It follows a checklist, and it stops when the checklist ends.

What the Omni Audit Looks Like

An Omni Audit takes 60 minutes. We sit down with you, map your current process, identify the mechanical work that’s eating your capacity, and design a narrow-mandate agent that handles it.

You walk out with three things. A process map that shows where your time is going. A mandate document that defines what the agent will do and where it will stop. And a 90-day implementation plan that gets the agent live without disrupting your practice.

We don’t pitch you a platform. We don’t sell you a subscription. We build the agent, deploy it into your workflow, and train your team to supervise it. The agent runs on Omni, which means it integrates with your existing tools: your phone system, your practice management software, your document storage, your calendar.

Most firms see ROI in the first 30 days. Not because the agent is magic. Because the work it’s doing was already costing you money, and now it’s not.

Book my Omni Audit and we’ll map out which parts of your practice are ready for this kind of automation.

The Line You Don’t Cross

The rule is simple. If the work requires a law license, an agent doesn’t do it.

An agent can capture intake information. It can’t assess the merits of a case. An agent can flag clauses in a contract. It can’t decide whether the contract is acceptable. An agent can draft a memo summarising discovery findings. It can’t decide what to do with those findings.

The line is bright, and you enforce it with mandates. You write the checklist. You define the boundaries. You supervise the output. The agent does the mechanical work. You do the legal work.

That’s how you deploy AI without adding risk. That’s how you scale your practice without scaling your exposure. And that’s how you sleep at night knowing your malpractice carrier isn’t going to ask uncomfortable questions.

Narrow mandates work because they respect the line. General-purpose AI doesn’t. Choose accordingly.

For more on how firms are building AI into their operations without crossing that line, explore the Omni Ops platform and see what rule-bound automation looks like in practice.