The agent is only as useful as the playbook behind it. In the build stage we sit with your lawyers and write down the positions they actually take: preferred wording, acceptable fallbacks and the points that always need partner sign-off. Then we test the agent on contracts you have already reviewed and compare its flags with yours. If your team disagrees internally about a clause, that surfaces here. Settling it once is a benefit in itself.
The output is a marked-up Word document with tracked changes and a short summary of issues, each with a reason. The lawyer accepts, edits or rejects every change, and signs off the final version. Nothing is sent to the other side or the client by the agent.
Give your lawyers a way to correct the agent. A quick note beside any flag saying whether it was right or wrong is enough. We read those notes in the review meetings and tune the playbook rules, so the agent follows how your team actually decides rather than how the playbook first read.
Heavily negotiated one-off deals, novel structures and poor scans of old paper contracts are better handled by a person from the start. The agent suits repeat document types where your positions are stable. Track how many suggested redlines your lawyers keep, and which playbook rules they keep overriding, then update those rules.
This work is specific to your field, so we build it around your rules, templates and examples instead of a generic model.