A small firm: client drafts back in hours, every send still reviewed

The job

The firm barely used AI. A lawyer might paste something into a chat window now and then, but the daily work ran the way it always had. Assistants keyed client and matter details by hand. A lawyer's notes went into an assistant's queue, and the draft came back when the queue allowed.

Nobody was doing anything wrong. The work was simply bigger than the hours.

What I built

• AI for the lawyers and staff on the firm's business plan, set up in person at each desk.
• Access to the firm's own precedents, so drafts start from the firm's own work rather than a blank page.
• Shared instructions for the firm's most common documents and tasks, so everyone gets the same quality of draft.
• Email drafting inside the email the firm already uses, plus dictation (you talk, it types).
• A connector to the firm's practice software that can find files, create records, add notes and log time. I tested every action before the team relied on it.
• A log of every change the AI makes, and drafts by default.

I worked alongside the firm's own IT provider, who handled the network side.

What changed

• A client emailed a lawyer with a question. A draft reply was ready for the lawyer to review within two hours of her email.
• In his first week, one lawyer cleared about three pieces of work that had been sitting on the back burner. His first case summary came back in about 15 minutes.
• One assistant used it to read and summarize dozens of client emails on a file. When she tested it with the wrong file number, it caught the mismatch and asked if she meant the other file.

What stayed in human hands

Nothing goes out without a person reviewing it and pressing send. That includes documents sent for signature. I tested it: when the AI tried to send an email, it stopped and waited for a person, and the send was refused.

Honest limits

The practice-software connector cannot edit or delete records, and it cannot read documents stored there. Building a larger package of documents is still the slow part of the day. Every connector has limits like these, and the team knows them.

Written by Devan Borsboom. Client names and identifying details (city, team size and anything specific enough to place them) are left out on purpose.