# Legal AI agent for law firms: what "agentic" should mean

By Josef Long. Published October 17, 2026. General information, not legal advice.

"Agentic AI" now describes everything from a chatbot with a button to software that works through a task on its own. For a law firm the useful question is narrower: what is the agent allowed to do, and who decides?

## Read, draft and act are three different permissions

- Read: search the meetings, documents and connected sources this lawyer may see. Nothing changes, so nothing needs approval.
- Draft: prepare an email, memo, letter in your template, time entry or digest. A draft stays a draft until a person acts on it.
- Act: anything that leaves the firm or cannot be undone: sending, deleting, sharing, exporting, filing. This is where the agent stops and asks.

## A plan you can read before it runs

Design preview (sample data): "Find every deadline in the depot agreement and Amendment No. 1, check them against our deadlines, and draft an email to the client listing the open ones." The plan has five steps, written by a model running on the firm's own server, and every step is labeled before anything runs:

1. Find both documents in the matter. Runs without asking.
2. Pull out every deadline, each with its citation. Runs without asking.
3. Check them against the matter's deadlines. Runs without asking.
4. Draft the email to the client. Drafts here, sends nothing.
5. Save the draft in your mail. Asks you first.

## The agent works as you, not as a superuser

The agent should inherit exactly what the lawyer using it may open. If a matter is walled off from that lawyer, it is walled off from the agent too, and walled material is dropped before any model reads it.

## Several agents, one set of walls

An outside assistant is a guest: an administrator approves each one, it sees only the matters the firm chooses, access expires, every write asks first, and it cannot reach raw audio. The built-in agent and every guest answer to the same walls and the same log.

## What a responsible agent proposes and does not decide

- Deadlines: it can do the date arithmetic and propose; a lawyer confirms the rule before docketing.
- Time: it can draft an entry; it should never post time on its own.
- Conflicts and privilege: it can flag a possible issue; it should not declare a conflict cleared or a document privileged.
- The record: a machine transcript is not a certified record.

The ABA's Formal Opinion 512 (July 29, 2024) says lawyers may not hand tasks that call for their own judgment to an AI tool: https://www.americanbar.org/content/dam/aba/administrative/professional_responsibility/ethics-opinions/aba-formal-opinion-512.pdf

## Six questions to ask any legal AI agent vendor

1. Can I see the plan before it runs, and edit it?
2. Which steps can change something outside the system, and does each one ask first?
3. Does it act as me or as a service account?
4. What happens when a document or email tells it to do something?
5. Is every action in a log I can read?
6. Who can connect another AI to it, and for how long?

## Questions

**What is a legal AI agent?** Software that takes a directive from a lawyer and carries out the steps across the firm's own sources. A good one shows its plan, works only with what that lawyer may see, and stops before any step that sends, deletes, shares or files.

**Is agentic AI the same as a chatbot?** No. A chatbot answers a question. An agent carries out a multi-step job using tools, which makes permissions the central issue.

**Can a legal AI agent act without a lawyer's approval?** It can read and draft without asking. Steps that send, delete, share, export or file should wait for a person, and a lawyer should review what it produces.

**Can other AI assistants connect to it?** In VaultScribe's design, yes, only when an administrator approves them, for the matters the firm chooses, for a limited time, with approval before each write.

## Where VaultScribe fits

VaultScribe is a private AI workspace for law firms with an agent built around these rules. Founding design partners sign a non-binding letter of intent, with terms agreed on a call: hello@vaultscribe.app.

Related: https://vaultscribe.app/private-legal-ai.md and https://vaultscribe.app/ai-for-law-firms.md.
