Every small law firm gets the same phone call. A client who hired you four months ago, paid a retainer that hurt, and has heard nothing since the engagement letter calls the front desk and asks where things stand. The paralegal puts them on hold, opens the matter, finds the last docket entry, walks down the hall, interrupts an attorney mid-draft, and comes back with a sentence like "we're waiting on the other side." The client says thanks. Nobody bills for it. It happens again the next week with a different client, and again the week after that, until the firm has a part-time job that exists only to translate silence into reassurance.
On October 26, 2026, at ClioCon in Boston, Clio shipped a pile of software aimed at that front desk. There is an AI phone system, an assistant that answers inbound calls after hours, a unified inbox, and a redesigned client portal that shows the client which stage their matter is in. All of it is useful. Almost all of it is built to answer the where-are-we call faster. My argument runs the other way. The call is a symptom of an update that should have gone out and didn't. A small law firm can have an AI agent send clients plain-language status updates on their matters, so staff stop answering where-are-we calls, on one condition that most vendors skip past. The agent relays only what the attorney has approved. It never characterizes an outcome and it never gives legal advice. That condition is the design, and the firms that get it right will have a quieter phone and a cleaner disciplinary record than the firms that buy the fastest answering machine.
What Boston shipped
The ClioCon keynote by founder and CEO Jack Newton was dense. According to LawNext's wrap-up of the October 26 announcements, Clio merged Grow, Manage and Work into a single interface (Newton called the new logo's clear center square "a single pane of glass"), with rollout starting in November. Clio Phones is a business phone system built into the platform that shows the caller's matter when the phone rings, and with recording on it transcribes the call, summarizes it, extracts tasks and prepares a time entry. Clio says its survey found lawyers lose an average of 4.6 billable hours a month to unlogged calls. Grow AI answers inbound calls, screens leads, runs conflict checks and books consultations, and Clio says 44% of client inquiries arrive after hours.
Then there are agentic workflows. Legal IT Insider's coverage describes firms writing a process in plain language or uploading an existing standard operating procedure, after which Clio builds the steps. LawNext reports that firms can inspect those steps and add approval points before switching the workflow on. Clio also introduced "playbooks" that define how a whole matter type is handled, including where the AI must stop for lawyer judgment. The Clio for Clients portal was redesigned so firms can show clients their matter stage and assign them tasks.
The same week brought Clio's 2026 Legal Trends Report, with numbers that frame the problem better than any keynote slide. Lawyer utilization is projected at 42% of an eight-hour day, up from 39% in 2025. 85% of firms now use AI. 46% of U.S. adults would consult AI before contacting a lawyer, and 82% want to know when their lawyer uses AI.
Read those together. Lawyers bill less than half the day. Clients are already asking chatbots legal questions before they call. They also want to be told when the firm's software is talking to them. The tools exist to automate client communication, and the clients have opinions about how it's done.
The call is a debt
A where-are-we call has a specific cause. The client has a mental model of their matter that has gone stale. They know the firm filed something in July. They don't know that opposing counsel asked for an extension in August, that the court granted it, and that nothing will happen until November. From the client's chair, silence and neglect feel the same, and the client has no way to tell which one they're getting.
Bar regulators have counted this for decades. The Florida Bar's review of 2024 discipline found inadequate communication was the fourth most common complaint against lawyers, behind interference with the administration of justice, neglect, and trust accounting. Elizabeth Tarbert, director of the Bar's Lawyer Regulation Division, said that "communication is always in the top 10." Neglect sits right above it, and in the client's eyes the two are the same complaint. A lawyer who is working hard on a matter and says nothing looks, from the outside, exactly like a lawyer who forgot.
So every unsent update is a small debt. The client pays interest on it in anxiety. The firm pays when the client calls and a staff member spends ten unbilled minutes reconstructing what happened, and pays more if the anxiety curdles into a bad review or a grievance. An answering machine, however intelligent, settles the debt one call at a time. Sending the update before the client feels the gap means the debt never builds up.
This is why I am less excited by AI phone systems than the vendors want me to be. Logging calls to the right matter and capturing the 4.6 hours is good accounting. It still treats the inbound call as the unit of work. The better unit is the matter event, the moment something changed, because that is when the client's mental model goes stale. If an update leaves the firm within a day of every event that matters to the client, most of the calls never get placed.
How a small law firm can have an AI agent send clients status updates
The idea sounds simple, and the naive version is dangerous. A naive version points a language model at the matter file and asks it to write the client a friendly summary. The model reads a deposition transcript, decides the other side's witness "struggled," and tells the client their case "is looking strong." The client screenshots it. Eight months later the case settles for less than the client hoped, and that screenshot reappears in a fee dispute.
The version that works splits the job in two. The attorney owns the facts. The agent owns the timing, the plain language and the delivery. Everything else is plumbing.
The attorney writes the fact, the agent writes the timing
Every status update has a payload, the one or two facts that changed, and an envelope, which covers who gets told, when, through which channel, in what reading level and with what reminder of next steps. Attorneys are bad at envelopes and indispensable for payloads. Software is the reverse.
So the workflow starts with a matter event. A filing lands, a deadline moves, a hearing gets scheduled, an offer arrives, a document comes back signed. The agent notices the event from the docket feed, the document system or the inbox, then drafts a candidate payload from a fixed menu (more on that below) and puts it in front of the responsible attorney as a single approval card. The attorney sees something like "Court granted the other side's request for more time. New deadline for their response: November 14. Nothing needed from you." They tap approve, edit a word, or kill it. Only then does the agent wrap the payload in an envelope and send it.
The approval card has to cost the attorney almost nothing. Clio's own framing of agentic workflows with approval points before activation is the right primitive, and so is the playbook idea of marking where AI must stop for lawyer judgment. The client update is the clearest case for both. Every outbound status message is a stop point, and the stop should take seconds. If approving an update takes as long as answering the call, the attorney will stop approving, the agent will stop sending, and the phone starts ringing again.
A vocabulary with walls
The agent should not be free to compose. It should pick from a controlled vocabulary of status statements written once per matter type by the firm's attorneys and reviewed like any other form. A family law firm might have forty of them. A personal injury practice might have sixty, covering medical records requested, records received, demand letter sent, insurer acknowledged receipt, offer received and so on. Each statement describes a procedural fact. None describes what the fact means for the client's chances.
This sounds rigid, and it is. The rigidity is what makes the agent safe to let out the door. "The insurance company made an offer. Your attorney will call you this week to discuss it" is a fact plus a handoff. "The insurance company made a lowball offer" is a characterization, and an agent should have no word in its vocabulary for "lowball." The language model's job is to choose the right statement, fill in dates and names from the matter record, and translate the result into the client's preferred language and reading level. It gets to be fluent. Inventing anything is off the table.
The vocabulary is also where the firm's judgment goes to scale. A senior partner who has spent twenty years learning how to tell a nervous client that discovery takes months without sounding evasive can put that skill into a dozen sentences. Every client hears the partner's phrasing, even when the partner is in trial.
What the agent refuses to say
The refusal list is the other half of the vocabulary, and it should be written down. The agent never predicts an outcome. It never compares the client's matter to other matters. It never says whether an offer is good, whether a ruling helps, whether the client should do anything beyond the specific task the attorney approved. When a client replies with a question that drifts into advice ("should I take it?"), the agent's only move is a handoff: it confirms receipt, says the attorney will respond, creates a task on the matter, and flags it on the attorney's queue with the client's exact words.
This is where a lot of client-facing AI products fail in practice. They are built as chat interfaces, and chat invites the client to keep talking. A status agent should behave more like a good legal secretary of the old school, warm and fast and completely unwilling to speculate. The client learns within two exchanges that the agent tells them what happened and when the lawyer will call, and that it does nothing else. Clients find that reassuring, because it is honest.
The ethics are already written
Firms tend to treat the regulatory picture as unsettled and use that as a reason to wait. On this narrow use case the rules are clear enough to build against today.
Oregon's Formal Opinion 2026-208, issued in February 2026, addresses chatbots and AI agents directly. According to the summary published by Ezel, it treats a chatbot vendor as a nonlawyer assistant under Rule 5.3, says the chatbot "must not give legal advice," and notes that autonomous actions such as sending an engagement letter should require lawyer approval. It also cites In re Weidner for the point that "the existence of the relationship turns on the client's reasonable belief," which is why a chatbot that sounds like a lawyer can create obligations its builder never intended. The opinion recommends clear disclaimers that the chatbot is not a lawyer.
That opinion is about intake chatbots talking to prospective clients, and a status agent talks to existing clients, where the relationship already exists. The principles carry over cleanly. The agent is a nonlawyer assistant the firm must supervise. It must not give advice. Anything with legal consequence gets a lawyer's approval before it leaves the building. The approved-payload design satisfies all three by construction, because no message reaches a client without a lawyer having seen its substance.
Alabama's Formal Opinion 2026-01, covered by the National Law Review on August 4, 2026, takes the same line: existing professional conduct rules apply to AI, lawyers must supervise the people and tools using it, and they should communicate with clients about it when appropriate. Pair that with Clio's finding that 82% of consumers want to know when their lawyer uses AI, and disclosure becomes easy to settle. Say so in the engagement letter, say so in the footer of every automated update, and tell the client how to reach a human. A firm that discloses up front loses nothing, while one that gets caught hiding it loses trust it may never get back.
Rule 1.4 itself requires lawyers to keep clients reasonably informed about the status of their matters and to promptly comply with reasonable requests for information. A firm that runs an approved-update agent is arguably meeting that duty more reliably than it did with humans alone, because the agent never forgets to send the update after a busy week. The legal risk of the agent is real and manageable. The legal risk of silence is already in the discipline statistics.
Why the giants won't build this for you
The last five months saw every large AI company plant a flag in law. LawNext's October 15 report on Microsoft's Copilot for Legal lists the sequence: Anthropic's Claude for Legal on May 12, 2026, Google's Gemini Enterprise for Legal in preview on August 25, OpenAI's Astra for Law on September 17, and then Microsoft's launch, with more than 20 legal skills in Word and its Cowork agentic workspace. Those products cover contract review, drafting, research and obligations tracking. They are built for the lawyer at the desk, and the client calling from the parking lot of their kid's school is not in the picture.
That is a reasonable product decision for them. Large firms bill corporate clients whose general counsel reads the docket themselves. The where-are-we problem is overwhelmingly a consumer-law problem. It lives in family law, immigration, personal injury, estate planning, criminal defense and bankruptcy, where clients are individuals, often frightened, often reading at a different level from their lawyer, and often in a second language. The firms serving those clients are small. They have one paralegal, maybe two. They don't have a knowledge management department to write status vocabularies.
Clio comes closest, because its portal now shows matter stages and its workflows can carry approval points. A stage indicator, though, is a lagging, coarse signal. "Discovery" can last nine months and tells a client nothing about whether anyone touched their file this week. The value sits in the specific, event-driven, approved sentence, and in the vocabulary underneath it. No vendor can write that vocabulary for your practice, because it encodes how your attorneys talk about your matters to your clients.
The arithmetic of silence
Firms underweight this because the cost of status calls is spread thin. No line on the P&L reads "reassurance." It shows up as a paralegal who can't finish document prep, an attorney who loses a train of thought, and a client who leaves a three-star review that says "hard to reach."
Suppose a four-lawyer family law firm carries 160 open matters, and the average client calls once a month to ask where things stand. Suppose each call takes twelve minutes of combined staff and attorney time once you count the hold, the file lookup, the hallway interruption and the callback. That comes to 32 hours a month, nearly a full work week, spent restating facts the firm already knew. Suppose an approved-update agent cuts those calls by two-thirds, and each approval card takes the attorney fifteen seconds across perhaps 300 matter events a month. The firm trades about 21 hours of interruption for about 75 minutes of tapping approve. Those figures are an illustration, and your own numbers will differ. The shape of the trade won't change much, because a pushed fact is far cheaper to deliver than a fact someone has to pull out of the file.
The second-order effects matter more. A client who gets a short, plain message within a day of every event stops wondering whether the lawyer is working. Fee disputes get rarer because the client watched the work happen. Referrals improve, because "they always kept me in the loop" is the review consumer firms want and rarely earn. The Legal Trends numbers point at the growth path too. Clio reports that firms using its AI tools record about 8% more revenue per responsible lawyer than comparable non-adopters, and utilization at 42% means more than half the lawyer's day is still unbilled. Status calls are one of the cleanest places to get some of it back.
Building it in the right order
The order of construction matters more than the choice of model.
First, write the vocabulary. Pull a year of sent emails and call notes for one practice area, have the attorneys mark every sentence they would be comfortable seeing in front of a disciplinary panel, and reduce those to a list of approved status statements with slots for dates and names. Write the refusal list next to it. This is a few days of partner time, and it is the most valuable asset in the system.
Second, map events to statements. For each matter type, decide which events trigger an update and which don't. A routine internal task should never ping the client. A filing by the other side always should, even when the update just says nothing is needed from them.
Third, put approval where the attorney already is. If the attorneys live in Clio, the approval card belongs in Clio's workflow queue. If they live in Outlook, it belongs there. An approval that requires opening a new app will be ignored by the second week.
Fourth, handle replies by handing them off. Every client reply that isn't a pure acknowledgment becomes a task for a human, with a promised response window that the agent states honestly.
Fifth, measure the phone. Count where-are-we calls per open matter before launch and every month after. If the number doesn't fall, the vocabulary is too vague or the events are firing too late, and you fix those before you touch the model.
Notice what's missing from that list. There's no step for picking the smartest model, and the system doesn't need one. A modest model choosing from a vetted list and translating into Spanish at a sixth-grade reading level is safer and cheaper than a frontier model improvising. The intelligence lives in the vocabulary, the triggers and the approval gate, all of which the firm owns.
Architect it, don't install it
Buying the phone system is the easy part, and it will make your firm better at answering a call that shouldn't have been necessary. The work that matters is deciding what your firm is allowed to say to a client without a lawyer in the room, writing it down, wiring it to the events that change a client's world, and putting a fifteen-second lawyer's approval between every one of those sentences and the client's inbox. That is architecture. It touches your ethics posture, your engagement letters, your practice management configuration, your staff roles and your partners' habits, and it fails if any one of them is left to default.
Firms that treat this as a software purchase end up with a fluent chatbot that one day tells a client their case looks strong. Firms that design it end up with clients who stop calling because they already know, and with paralegals who spend their afternoons on the work clients pay for. The vendors shipped the parts this October. Agor AI Advisory helps firms design the vocabulary, the approval gates and the event map, configure them inside the platforms you already run, and measure the phone until the calls fall. Clients who hear what happened before they think to ask don't need to call. Schedule a strategic consultation with us today.
Sources
- LawNext, ClioCon 2026: A Quick Wrap-Up of Everything Clio Announced Today, October 2026
- Legal IT Insider, Clio unveils unified platform with agentic workflows, AI phone system and new DMS, October 2026
- LawNext, Microsoft Launches Copilot for Legal, October 15, 2026
- Ezel, Oregon State Bar Formal Opinion 2026-208: Chatbots and AI Agents
- National Law Review, Alabama State Bar Issues AI Ethics Guidance for Lawyers, August 4, 2026
- The Florida Bar News, 2024 Florida Bar discipline trends, February 17, 2025