All insights
    LOG/012Oct 13, 2026Thought PiecesShadman Hossain

    AI agents for small law firms: five questions to ask first

    A small firm does not lose to bigger firms on skill. It loses on hours. The partner who should be thinking about the case is answering scheduling email at 9 PM.

    An AI agent can take that work. Most small firms still say no, and the reason is trust. The longer argument is in Why SAIN is building the trust infrastructure for AI. This piece is the practical version for a firm of two to twenty people.

    What an agent can take off your desk

    Start with the work around the work. A title lawyer with four inboxes does not need an agent to practice law. They need the document chase to continue after hours. They need the closing calendar to stay current. They need the client who asked a question on Friday to get an answer on Friday.

    That is the right first job for an agent in a law firm:

    • Intake follow-ups and scheduling
    • Document collection and reminders
    • Status updates to clients, drafted for your approval
    • Meeting prep and notes
    • First drafts that an attorney reviews and signs

    None of that is legal judgment. All of it eats the week.

    The line that keeps it safe is simple. Reading, sorting, and drafting can run on their own. Anything that leaves the firm under your name waits for a person. Run one evening both ways:

    Five questions to ask any vendor

    1. What leaves the firm without my approval? The right answer is nothing. An agent should draft freely and send only when a person says yes.

    2. Can I show what it did? Ask to see the record of one task from start to finish. Every action, every approval, every draft. If the vendor shows you a chat history, keep asking.

    3. Can anyone edit that record? You want a tamper-evident audit trail. If the vendor, the agent, or a staff member can change an entry and leave no sign, the record will not settle a question when a client or a carrier asks.

    4. Can I show the record without handing over the file? Privilege and confidentiality do not pause because someone asks a question. The record should show who acted, who approved, and when, and it should let you check that a document is the one that was sent without putting the document itself in front of anyone. You decide which file, if any, to show.

    5. What happens when it does something unusual? Someone should find out fast, and the agent should wait until a person decides. If the answer is "you can review the logs," then you are the supervisor, and you were asleep.

    Take these into your next sales call. Pick the answer closest to what the vendor said:

    What the bar will care about

    Guidance on AI in practice keeps landing in the same place. You are responsible for the work. You must supervise it. You must protect client information. You must be able to explain what was done.

    An agent with no record makes all four harder. An agent with approval gates and a tamper-evident audit trail makes all four easier. The record becomes your supervision file.

    Where to start

    Pick one workflow. The best candidates are repetitive, low risk, and visible: the document chase on a closing, the follow-up after a consult, the weekly status note. Run it for two weeks with approvals on everything. Then read the record and ask three things:

    1. Did any draft need heavy editing? That tells you where the agent needs better instructions.
    2. Did anything wait too long for approval? That tells you who should be the approver.
    3. Could you explain every entry to a client? If not, that workflow is not ready.

    We built Aura for this work. It runs in iMessage, where you already are, and a new agent sets up in under 10 seconds. Anchor'd is our tamper-evident record. There is more on the firm-specific setup on our legal page, and on files that should never leave the office in Privacy, transparency, intelligence.

    If you want to think through which workflow to start with, we are glad to listen first.

    Talk to us.