
Why AI drafts and attorneys decide
AI can draft, compare, check and organize legal work. What it cannot carry is the professional duty attached to advice and to a signature. That is why the line sits where it does: AI drafts, attorneys decide.
The short answer
AI should draft legal work while attorneys decide because the duties attached to legal work, to advise, to sign and to stand behind a position, belong to a licensed person and cannot be handed to software.
An agent can do the drafting, comparing, checking and organizing. The attorney reads the result, judges it and makes every call.
Definition
AI drafts, attorneys decide: a division of legal work in which an agent prepares and checks material, and a named attorney makes every judgment, gives every piece of advice and signs everything that carries legal effect.
What goes wrong when the line is not drawn?
Two failures show up. In one, a team lets an AI tool produce something that goes straight to a client or a court, and nobody with the professional duty read it closely. In the other, a team bans AI entirely and keeps spending attorney hours on comparison, formatting and citation checking that a well-governed agent could do. Both come from treating the question as all or nothing, when the work divides cleanly.
Who needs to decide where the line sits?
General counsel, managing partners and owners who are deciding how far to let AI into legal work. It matters most where legal work is high-volume and repetitive, such as contracts, NDAs, research and status reporting, because that is where both the time savings and the risks concentrate.
Which legal work can an agent take, and which stays human?
| Kind of work | An agent can do it | An attorney must do it |
|---|---|---|
| Intake and routing | Log, classify and route to the named reviewer | Decide priority when it matters |
| Comparison | Mark every departure from your playbook | Decide which departures matter |
| Research | Draft and verify every citation against its source | Judge the analysis and what to rely on |
| Client communication | Draft a status report from the record | Give advice and send it |
| Signature | Prepare the document | Sign, and carry the duty that comes with it |
How it works, step by step
The division is applied the same way to every legal workflow.
Name the decision in each workflow
Give the agent everything before the decision
Keep drafting and checking separate
Human gate: the attorney decides, advises and signs
Record who decided what
Revisit the line as the work changes
What keeps the line in place?
One name per decision
Each decision has one attorney who owns it, as approval gates describes.
Checks against sources. Agent output is verified against your sources before anyone relies on it.
Nothing outbound without an attorney
No advice, filing or signature leaves the firm from the agent.
Rules that differ by state
Professional duties are set where the attorney practises: the ethics rules that govern lawyers take effect state by state, as states choose to adopt them.
A record. The file shows who prepared and who decided.
Where this fits
This principle sits under every legal workflow in the library and on the Platform page that describes how gates are built. Before choosing a first legal workflow, the readiness checklist helps show where the decision in each one sits. The broader version of the argument, for work beyond law, is when not to automate, and the research discipline it relies on is in the governed research brief. The use case library shows the workflows built on it.
How does this differ from banning AI in legal work?
| Question | Banning AI | AI drafts, attorneys decide |
|---|---|---|
| Attorney time on routine checking | All of it | Shifted to review of checked work |
| Who decides | The attorney | The attorney |
| Risk from unchecked output | Pushed into unofficial use | Handled by checks and a gate |
| A record of the work | Whatever the attorney kept | Who prepared and who decided, every time |
Related reading
- Approval gates: how one named person owns each decision.
- When not to automate: the general case for work that stays human.
- Shadow mode: how an agent earns trust before anyone relies on it.
