Which use case should we actually start with?

Usually NDAs, because you issue the template, the positions are well understood and the volume makes the result obvious within weeks rather than quarters.

Do we need to load our contract history first?

No. The agent works from the mandate you write, not from a corpus of past deals. Historical paper is useful context later; it is not a prerequisite.

What if the agent's first proposals are wrong?

That is what the lowest autonomy level is for. On observe, nothing leaves without a person approving it, so a wrong proposal is a correction to the mandate rather than an incident.

Can we run two use cases with different autonomy levels?

Yes, and most teams do. Autonomy is set per agreement type, so NDAs can execute inside the mandate while MSAs stay on propose-and-wait.

Start with one agreement type, not the whole function

Rolling a platform across every contract at once is how rollouts die. Each page below takes a single agreement type with enough volume to matter, and shows the agent running it from intake to signature — so the pattern is proven on paper you already understand before it is widened.

Workflows worth handing over first

Each page takes one agreement type and follows it from intake to signature, including the points where the agent stops and asks.

NDAs at scale — Standard paper, high volume, handled unattended
MSA negotiation — Long agreements, many reviewers, one thread
Vendor renewals — Never renew by accident again
Employment agreements — Offers, amendments and jurisdiction-correct terms
Property leases — LOI to executed lease, dates included

What makes a good first workflow

The first agreement type you hand over decides whether anyone trusts the second. These are the properties worth optimising for.

Enough volume — Something that happens weekly, not quarterly. A workflow you see often is a workflow you can judge, and the time saved is visible without a spreadsheet.
Paper you control — An agreement where you issue the template. Your approved position already exists, so the mandate is a transcription job rather than a negotiation about the negotiation.
One clear owner — A person who can say what the walk-away point is without convening a committee. Gates need an owner more than they need a policy.
A visible finish — A signature, a counter-signature, a renewal date. Something that either happened or did not, so the result is not a matter of opinion.

How a first workflow normally goes

Nothing here requires a migration. The sequence below is the one most teams follow on their first agreement type.

Write the mandate — Your approved position, the clauses that cannot move and the walk-away points, in your own words.
Run it on observe — The agent drafts and proposes while a person approves every outbound move, so you see its judgement before it has any.
Raise the dial — Once the proposals stop surprising you, let it execute inside the mandate and keep the gated clauses coming back.
Close the loop — Signature, agreement history, and a record of which deviations were accepted and by whom.
Widen it — Take the next agreement type. The machinery is already configured; only the mandate is new.

Questions about getting started

Which use case should we actually start with? Usually NDAs, because you issue the template, the positions are well understood and the volume makes the result obvious within weeks rather than quarters.
Do we need to load our contract history first? No. The agent works from the mandate you write, not from a corpus of past deals. Historical paper is useful context later; it is not a prerequisite.
What if the agent's first proposals are wrong? That is what the lowest autonomy level is for. On observe, nothing leaves without a person approving it, so a wrong proposal is a correction to the mandate rather than an incident.
Can we run two use cases with different autonomy levels? Yes, and most teams do. Autonomy is set per agreement type, so NDAs can execute inside the mandate while MSAs stay on propose-and-wait.