On October 15, 2026, Microsoft launched Copilot for Legal. It ships more than 20 legal skills that live inside Word and Cowork, the agentic workspace Microsoft has been building. Bob Ambrogi at LawSites counted it as the fourth major AI company in five months to build a legal vertical. Anthropic went first in May, Google followed in August, and OpenAI came in September. All four aimed at the document: redlines, research, chronologies, obligations tracking. Five days later, on October 20, Smith.ai named Suresh Khanna its new CEO. Khanna spent six years at Google and led its North America SMB ads teams, and in the announcement he said what most managing partners already suspect about their own firms. Law firms, he said, "often treat intake as an administrative task rather than a revenue engine."
The month's announcements leave a gap, and I want to argue from it. If you run a firm of 8 to 40 people, the most valuable AI agent you can deploy this winter will never draft a clause. It books time. It handles consultations, rescheduling, reminders and no-show follow-up, and it never gives a client advice. Working out how a small law firm can have an AI agent handle scheduling is a smaller problem than the one the giants are solving. For a firm your size it pays far better.
Four giants looked at the same desk
Ambrogi's summary of the four launches is blunt. "All four follow essentially the same model, providing a set of legal-specific skills or agents, a roster of connectors into the specialist products lawyers already use, and a promise that lawyers can get more work done without leaving the platform."
The dates matter, because they show how fast the money moved. Anthropic fired the starting gun May 12, releasing more than 20 MCP connectors and 12 practice-area plugins for Claude. Google followed Aug. 25 with Gemini Enterprise for Legal, launched in preview with Cleary Gottlieb, Freshfields, Weil Gotshal and Williams & Connolly as launch customers. OpenAI launched Astra for Law Sept. 17, a configuration of its GPT-6 Astra model with a dedicated legal search index covering U.S. case law, statutes, regulations and court rules. Then Microsoft.
Look at who those launch customers are. Freshfields. Cleary. Weil. Clyde & Co, whose CIO told Microsoft that Copilot adoption at the firm is "now approaching 90%". These are firms with thousands of lawyers and a building full of people whose job is answering the phone. Their bottleneck really is the document. For them, a redline that takes minutes instead of an hour is real money.
Microsoft also published its scorecard. Since April, it said, more than 1,800 users have generated nearly 100,000 edits and kept roughly seven in ten of the suggestions. That is a fine number for a contract team. It tells a managing partner of a 15-lawyer family law practice almost nothing, because her contracts are not where the firm loses money.
Read the partner list. Microsoft named legal tech partners in five categories: legal research, document and matter management, agreements and e-signature, specialized legal AI, and e-discovery. No intake vendor. No calendaring vendor. Nobody whose product answers a ringing phone at 7:40 in the evening. The biggest software companies on earth spent five months building tools for the lawyer at the desk and none for the person trying to reach her.
Where an 8 to 40 person firm actually bleeds
Clio has been publishing the same uncomfortable figures for years, and it repeated them on August 20, 2026, when it launched Grow AI. As uncovered by Clio's Legal Trends Report, 79% of prospective clients expect a reply within 24 hours. However, 67% of firms don't respond to email inquiries and 48% don't respond to phone inquiries. Clio also put a name on the cost. Those missed opportunities can lead to higher acquisition costs; firms increase ad spend to attract customers they should have already captured.
That last sentence describes most small firms I talk to. They pay Google to make the phone ring, and then they let it ring.
The timing makes it worse. Clio says data from Grow AI shows nearly 50% of leads captured with the tool occurred outside normal business hours. People call a lawyer after work, after the kids are asleep, or after the car accident that happened on the drive home. Your receptionist leaves at five, your associates are billing, your partners are in court, and the person with a real problem reaches a voicemail box and then calls the next firm on the search results page.
The arithmetic is simple enough to do on a napkin, so here is an invented version. Suppose your firm spends $6,000 a month on ads and that buys 120 inquiries. Suppose half arrive after hours and you return a third of those by noon the next day. For illustration, if your normal rate of turning a consult into a signed client is one in four, every lost evening caller is a quarter of a fee walking away, and a firm with an average matter worth $4,000 loses real money in a single month. Put in your own numbers. The shape will hold, and it is always worse than the partners guess.
A paralegal at one Clio customer described the change in plain terms. "We've hired 1-2 additional clients a month just by having Grow AI handle after-hours calls that otherwise would have gone to voicemail," said Christian Moreno, Intake Paralegal, Rosenberg Perry & Associates. One or two clients a month is less than a bar-journal headline. Over a year it pays for the tool, a paralegal raise and some ad spend besides.
How a small law firm can have an AI agent handle scheduling
Ask the right first question. Your firm already has a scheduling policy. It just lives in four people's heads and a sticky note on the receptionist's monitor. An agent forces you to write it down, and writing it down is the actual project. The software is the easy half.
I break the work into four jobs, because each one fails in its own way.
Booking the consultation
The agent answers the phone, the chat window and the web form. It collects the facts that decide who should see this person: practice area, county, opposing party, how urgent it is. Then it puts a real slot on a real lawyer's calendar.
Two rules matter more than any others. The agent runs the conflict check before it books, never after, because a consult with the other side's spouse is a disaster you cannot undo with an apology email. Clio built this order into its own product: per the launch, Grow AI checks the lead against the firm's own case lead scoring and conflict rules before anything else happens. The second rule is that the agent knows which lawyer takes which kind of matter, and at what hour. Somebody in your firm knows that the senior partner only takes estate consults in the afternoon and never on court days. Today that knowledge lives with one person. It needs to live in a file.
Rescheduling
Rescheduling is where cheap tools break. A client who moves a meeting is giving you information. They may be losing interest, or something may have gone wrong, or they may be talking to another firm. A human receptionist hears that in a voice. An agent needs rules for it.
Say your policy lets a prospect move a consultation twice, and the third request sends a note to the responsible attorney instead of simply offering new times. Say a paid consultation keeps its deposit across one move and releases it after the second. You can pick different numbers. You do have to pick, because the agent will do exactly what you wrote and nothing you meant.
Reminders
Reminders are the dull part, and they do the most work. A booked consult that nobody reminds is a coin toss. Say the agent sends a confirmation at booking, a text the day before with a one-tap confirm or move link, and a short note two hours ahead with parking directions and the documents to bring. Each message asks for a reply, so a silence the day before becomes a signal your staff can act on while the slot can still be filled.
No-show follow-up
This is where small firms leave the most on the table, because following up with someone who stood you up feels awkward, and nobody on staff owns it. Clio's product chases exactly this. The launch lists SMS and email to chase outstanding intake forms, unsigned agreements, and unscheduled consultations. Cold leads are flagged for follow-up or closure.
The design question is tone and stopping. Say the agent sends one warm message an hour after the missed slot, offering two new times, and one more three days later, and then it stops and hands the record to a human with a note. People miss consultations for reasons that make them better clients later: a hearing moved, a shift changed, they got scared. A firm that follows up kindly and then lets go will sign some of them. A firm that sends seven texts will get a bar complaint.
The calendar is a policy document
This is the part most vendors skip in the demo.
Your calendar decides who matters to your firm. The ten o'clock slot with your best litigator is the scarcest thing you sell. Today, who gets it depends on whoever happened to pick up the phone, how tired they were, and whether the caller sounded like money. The decision is made hundreds of times a year, by people with no instructions, and it sets your revenue more directly than your rate card does.
When you put an agent in front of that calendar, every one of those choices becomes explicit. Do high-value matter types get same-day slots while the rest wait three days? Do paid consultations get the morning? Does a referral from a former client skip the queue? Does an urgent criminal matter outrank a scheduled estate review? Your firm has answers to all of these. You have never seen them written down, and once you do you will want to change several.
That is why I tell managing partners that the scheduling agent is a strategy project dressed as an admin purchase. Whoever writes the booking rules is pricing your firm. If you accept a vendor's default template, a product manager in another city is deciding which of your callers get your best people. Write the rules yourself, in plain language, in a document the firm owns, and treat any software as the thing that carries them out.
There is a second reason to own the rules. Vendors change. Smith.ai has a new CEO, a new chief product officer and a new chief customer officer as of this week. Clio, 8am and a dozen startups are all shipping intake agents. If your booking policy lives only in one vendor's settings screen, switching costs you the policy along with the contract. If it lives in your own file, switching costs you a weekend.
The line it does not cross
A scheduling agent is one of the few AI agents a law firm can run with almost no professional risk, and the reason is the rule I put in the opening: it books time and never advises. That rule has to be built into the agent, because the callers will push against it constantly.
The ethics bodies have started to say this out loud. In February, the Oregon State Bar issued Formal Opinion 2026-208 on autonomous AI agents for client intake. Its answer to whether firms can use them was two words: "Yes, qualified." The qualifications are the useful part. The opinion counsels that lawyers need to understand how the agent reacts to particular prompts and monitor its activities so that it does not, for example, appear to create an attorney-client relationship, promise services that the law firm can't deliver, or guarantee particular results. The opinion also frames the duty to supervise under RPC 5.3 from the perspective of supervising the service provider rather than treating the AI agent itself as an "assistant" for purposes of the rule.
Read that last point as a managing partner. Oregon is saying you supervise the vendor. That means you need to know what the vendor's agent says when a caller asks whether they have a case, whether they should sign the settlement offer, or whether they can skip tomorrow's hearing. Those questions will come in through the scheduling line, every week, because a frightened person calls the number they have.
So most of the agent's script is refusals. It needs a short list of trigger phrases that send a call straight to a human: a hearing in the next 48 hours, an arrest, a protective order, a filing deadline. It needs a plain sentence it repeats when asked for advice, offering the earliest slot with a lawyer instead. It needs to avoid words like "your case" before anyone has signed an engagement letter. Test it the way a hostile caller would. Have your newest associate spend an afternoon trying to get legal advice out of it. Every answer it gives that it should not have becomes a rule.
Buy it, wire it, or staff it
Three paths showed up in the market this quarter, and each one decides where your booking rules end up living.
The first path is the suite. Clio's Grow AI, launched August 20, puts phone, email and chat agents inside the practice management system you may already pay for. Automated lead intake that collects information and schedules qualified leads into a consultation. Clio makes a big claim for it: Clio data shows matters are three times more likely to convert to hired when Grow AI is used. Treat that as a vendor figure until your own month of data agrees. The advantage is that the booked consult, the conflict check and the eventual matter all live in one record.
The second path is wiring a general model into your existing system. On September 23, at its Kaleidoscope conference in Las Vegas, 8am launched an MCP connector that lets Claude act inside MyCase. 8am says it is the first Claude MCP connector built specifically for legal practice management. The connector supports 73 different actions across cases, clients, documents, billing and intake, the company says. That opens a path where your own scheduling logic, written in your own words, runs on a model you choose and acts on the records you already keep. It is more work. In return you own more of it.
The third path keeps humans in the loop at the front door. Smith.ai's pitch is a blend: it describes itself as combining AI efficiency with a network of more than 500 live, North America-based agents, built on more than 30 million calls and a decade of experience with over 5,000 small to mid-sized law firm clients. For a firm whose callers are often in crisis, such as criminal defense or domestic violence work, a human backstop on the urgent calls may be worth the price difference.
None of the three paths makes the policy work go away. They differ only in whose screen your policy ends up on.
What a managing partner should do before January
Start with an audit you can finish in two weeks. Pull every inquiry from the last 90 days, from every channel, and mark three things for each one: when it arrived, how long until a human responded, and what happened next. Most firms find that their response time after five o'clock is measured in days.
Then write the booking policy. Keep it to two pages. It covers matter types and which lawyers take them, consult lengths, paid versus free, conflict-check order, the urgent triggers that go to a human, reschedule limits, the reminder sequence and the no-show sequence. Have the partners argue about it in one meeting. The argument is the most useful hour of strategy your firm will have this year, because it is the first time anyone has asked out loud who deserves the ten o'clock.
Then pick a path and run it on one practice area for 60 days before rolling it out firm-wide. Measure booked consults, kept consults and signed engagements. Ignore the vendor's dashboard of calls answered, which measures activity. You want to know how many people sat down across from a lawyer and then signed.
The work the giants skipped
Microsoft, Google, OpenAI and Anthropic will keep competing for the big-firm document desk, and they should. The document is where a 2,000-lawyer firm makes and loses money. A 20-person firm makes and loses money in the 90 seconds after a stranger calls, and nobody on that October 15 partner list is working on those 90 seconds.
That gap is your opening, and it will not stay open. Smith.ai just hired an ad-tech operator to run legal intake as a sales funnel. Clio is selling conversion. The firms that write their own booking rules now will move from vendor to vendor as the tools get cheaper and keep their policy each time. Firms that skip the writing will rent somebody else's judgment about who gets seen, and they will pay for it in ad spend every month.
This is the work Agor AI Advisory does. We sit with your partners, get the unwritten scheduling rules out of their heads and into a policy you own, pressure-test the agent's refusals against the ethics guidance in your state, and pick the path (suite, wired model or staffed hybrid) that fits your practice areas instead of a vendor's quarter. Buying a scheduling tool takes an afternoon. Designing the rules that decide who walks through your door is the strategic decision, and it is too important to leave as a factory setting.
Sources
- LawSites, "Microsoft Launches Copilot for Legal, Becoming the Fourth AI Giant in Five Months to Build Out a Legal Vertical," October 15, 2026
- LawSites, "Smith.ai Names Suresh Khanna CEO, Expands Executive Team to Accelerate Growth in Legal Intake," October 20, 2026
- LawSites, "At Its User Conference in Las Vegas, 8am Announces MCP Integration and New AI Tools for MyCase," September 23, 2026
- Clio, "Introducing Grow AI, 24/7 Client Acquisition In Clio Grow," August 20, 2026
- NWSidebar (WSBA), "New Oregon State Bar Opinion on Using AI Agents for Client Intake," March 23, 2026