Sullivan & Cromwell's Agreement Analyzer brings legal expertise to contract review!
Hi everyone, it's Shii-chan! Today I'm sharing a story about a law firm building its own AI tool.
OpenAI News
openai.com
What was announced?
OpenAI News describes how the law firm Sullivan & Cromwell worked with OpenAI to build "Agreement Analyzer", an AI tool that finds issues in a draft agreement. It brings the firm's precedents, experience, and client priorities into one workflow.
Why it matters
Partner John Savva first built it as a single-user proof of concept using Codex. He then worked with OpenAI to improve output quality, add parallel processing, and let many lawyers use it at once. It's a nice example of a domain expert shaping an AI tool around their own work.
What changes
Lawyers can do deeper analysis and find more authority and precedent. The article frames it as a way to extend lawyers, not replace them.
Dive Deep
The flow looks like this:
- A lawyer uploads a draft agreement
- They add selected precedents, a playbook of firm and client positions, and market-practice information
- The tool runs multiple review cycles from different perspectives, then synthesizes and flags issues
- It returns a memo and proposed tracked changes, which the lawyer can review, adapt, or reject
It combines the firm's workflows, curated precedents, negotiating playbooks, and ChatGPT Enterprise. It also compares matching provisions across precedents and tracks how provisions interact.
Savva says trustworthy output needs visible sources. In the adversarial setup described:
- Specialized agents review in parallel for market deviations, risk allocation, clause interactions, and ambiguities that could affect litigation
- Other agents challenge their conclusions
- Findings are sorted into supported after challenge, dismissed, or unverified and kept with qualifications
Lawyers can inspect each issue, its explanation, and the relevant precedents before deciding what goes into the final markup. Savva credits improved reasoning in GPT-6 Astra with addressing an earlier bottleneck in analysis quality. The article does not report metrics such as time saved or accuracy.
Wrap-up
- Sullivan & Cromwell built Agreement Analyzer with OpenAI
- It combines precedents, playbooks, and ChatGPT Enterprise to review agreements
- Multi-perspective review and challenge sort findings into a memo and tracked changes
- Lawyers make the final call
A great read if you're curious how professional AI workflows get built!
