AI agents for law firms: 7 workflows, explained
From a conflict check run before the first call to a research memo where every citation has been opened and verified: what an AI agent can do in a law firm, and what stays with the lawyers.
Lawyers bill for judgement, but a large part of the week goes to reading, checking and chasing: running conflicts, comparing a counterparty's draft to the firm's positions, finding the order that resets a deadline, reconstructing the day's time. An AI agent takes that work across the firm's systems and hands the lawyer something to decide on.
Here are the workflows, told through a mid-sized commercial and litigation practice.
What makes it an agent, not a chatbot
Legal AI tools usually answer questions in a chat window. An agent does the work inside your firm's systems, and shows you its sources. Three things set it apart:
1Client intake with the conflict check done before the first call
An enquiry comes in, and before anyone can even talk to the client we need the parties, a conflict check across every matter we've ever opened, and the limitation date.
When a new enquiry arrives by email, web form or a call note, Meher's agent:
- Extracts the facts: who the client is, the other side, related companies and directors, what happened and when.
- Runs the conflict check against every client, opposing party and related entity in the matter system, including name variations and group companies.
- Flags time limits such as limitation periods and notice deadlines from the dates in the story.
- Drafts the engagement letter with the scope and fee structure the partner usually uses for this kind of matter.
| Party | Found in our records | Assessment |
|---|---|---|
| Orbit Foods Pvt Ltd | No prior matters | Clear |
| Coldchain Logistics LLP | No matters | Clear |
| Coldchain's parent, Polar Holdings | Former client: 2023 lease matter, closed | Partner to decide |
| Director R. Menon | Witness in a 2021 arbitration | Note only |
2Contract review against your playbook, with fallbacks already drafted
A counterparty sends a 40-page MSA on Friday afternoon. I know our positions; what takes time is finding every place their draft departs from them.
Aditi's agent compares the incoming draft with the firm's playbook for that contract type and the client's own preferences, then produces a marked-up draft and a one-page issues list.
| Clause | Their draft | Our position | Proposed |
|---|---|---|---|
| 12 · Liability cap | Uncapped for supplier | Cap at 12 months' fees | Cap at 12 months' fees, carve-outs for confidentiality |
| 14 · Indemnity | Customer indemnifies all IP claims | Mutual, limited to own materials | Mutual indemnity redline |
| 22 · Governing law | Singapore, SIAC | Indian law, Mumbai seat | Fallback: SIAC acceptable to the client |
| 8 · Payment | Net 60 | Net 30 | Net 45 (client's stated fallback) |
3Research memos where every citation has been opened and checked
The danger with AI in research is a confident citation to a case that doesn't say that, or doesn't exist. I won't put my name to anything I can't trace.
Karan's agent drafts research memos from the firm's research database and precedent bank, and treats citations as claims to verify, not text to generate. Every authority it cites, it has opened; every quote is matched against the source text; anything it can't verify is marked.
| Authority | Proposition | Check |
|---|---|---|
| Supreme Court, 2015 (from the research database) | A genuine pre-estimate of loss is enforceable | ✓ Quote matched at para 43 |
| High Court, 2019 | Burden on the party claiming a penalty | ✓ Verified, still good law |
| Arbitral award, 2022 (firm precedent bank) | Applied to delay damages in supply contracts | ✓ Internal, privileged |
4Every order read the day it's uploaded, and every deadline in the calendar
Missing a deadline is the one mistake you can't fix. Orders come out late in the evening, and each one can reset three dates.
Joseph's agent checks the court and tribunal websites and cause lists for every live matter daily, reads new orders, updates the next hearing date and computes the filing deadlines that follow from them. Then it drafts the client update.
| Matter | Order | Deadline set |
|---|---|---|
| Orbit v. Coldchain | Reply to be filed within 30 days | Fri 14 Nov (calendar + reminder at T−7) |
| Rao Estates v. City Bank | Next hearing 3 Dec | Brief to counsel by 26 Nov |
| Arbitration 17/2025 | Tribunal extended evidence deadline | Moved 20 Oct → 10 Nov |
5M&A due diligence: the data room turned into a disclosure-ready issues list
In a deal, the work is reading 1,500 documents to find the 30 that matter, and then telling the client what they mean.
The agent reviews every contract, licence and board resolution in the data room against the due-diligence checklist: change-of-control clauses, assignment restrictions, exclusivity, unusual indemnities, expiring licences, litigation. Each finding links to the exact page.
| Issue | Documents | Severity |
|---|---|---|
| Change of control needs consent | 11 customer contracts | High |
| Trade licence expired, renewal pending | Dubai branch | Medium |
| IP created by a contractor without assignment | Core software module | High |
6The day's time entries drafted from what the lawyer actually did
Lawyers lose ten to fifteen percent of their time because they fill in timesheets on Friday from memory.
Each evening the agent drafts time entries from the lawyer's emails, calls, calendar and the documents they edited, coded to the right matter and task, in the narrative style the client's billing guidelines require.
| Matter | Entry | Time |
|---|---|---|
| Orbit v. Coldchain | Research on liquidated damages; draft memo sections 2–3 | 3.1 h |
| Northwind MSA | Review supplier draft against playbook; issues list | 2.2 h |
| Rao Estates | Call with client re hearing strategy (calendar, 40 min) | 0.7 h |
7When an associate leaves, the matter knowledge stays
When an associate moves on mid-matter, the new person spends weeks catching up on what was agreed in calls nobody wrote down.
The departing lawyer's work memory for each matter moves to whoever takes it over: the history, the client's preferences, the open points, the drafts in progress. Personal notes stay private.
Where people stay in charge
In a law firm, responsibility can't be delegated to software. Every workflow above ends with a lawyer reviewing, deciding and signing.
- Nothing goes to a client, a counterparty or a court without a lawyer's approval.
- Every source is linked and every citation verified, or clearly marked as unverified.
- Privilege and confidentiality are respected. Each firm's agents and firm brain are separate, and access follows your matter permissions.
How to start
- Start with intake and conflicts or contract review: high volume, clear rules.
- Connect the matter system, email, the document system and your playbooks.
- Run in shadow mode for two weeks and compare with your team's work.
- Turn on review mode once the team trusts the output, then add the next workflow.
Frequently asked questions
Can an AI agent run conflict checks for a law firm?
Yes. It extracts the parties and related entities from a new enquiry and checks them against all past and current clients, opposing parties and group companies in your matter system, including name variations. A partner decides on every hit.
Can it review contracts against our playbook?
Yes. It compares the counterparty's draft with your positions and the client's preferences, proposes redlines and fallbacks, and produces an issues list for the lawyer to review.
How do you avoid fake citations in legal research?
The agent only cites authorities it has opened in your research sources, matches each quote against the source text, and marks anything it couldn't verify. The lawyer reviews the reasoning.
Can it track court dates and deadlines?
Yes. It reads new orders and cause lists for each live matter, updates hearing dates, calculates the deadlines that follow and puts them in the calendar with reminders.
Will AI replace lawyers?
No. It takes over reading, checking and first drafts so lawyers spend their time on judgement, strategy and clients.
In a 30-minute call we'll run an agent on a sample contract against your playbook.
The people, companies and numbers in this article are illustrative.