- Home
- AI agents
- Enterprise AI agent examples
- Legal and contracts
Plan · Enterprise AI agent examples
AI agents for legal and contract review: what they do and what legal teams report
A contract review agent reads a contract against the legal team’s playbook, flags clauses that depart from agreed positions, drafts redlines and comments with the reason, and answers questions about signed agreements. A lawyer decides on every change: the agent prepares the first pass and the lawyer owns the advice.
In this guide: 83 pages
Key takeaways
- Every published contract review agent assists. It drafts a first pass and a lawyer accepts, edits or rejects it; none of the contract examples we checked lets an agent send a contract or advice on its own.
- The figures measure time on a narrow document type. Proof reports 50% to 75% less time on NDA reviews with Ironclad, and Dentsu’s 4 to 10 hours saved per lawyer per week is described as an estimate.
- An agent can only flag a departure from a position someone wrote down. For many teams the first project is writing or updating the playbook.
- Wilson Sonsini’s published scores, 95% accuracy on first-party contracts against a 90% target, are evaluation results, and a good model for your own acceptance test.
What do AI agents do in legal and contract work?
They give a first-pass review of contracts against a playbook, redraft clauses, run narrow checks such as non-compete conflicts, extract and compare terms across a portfolio, translate templates and answer questions about signed agreements. In every published contract review case a lawyer reviews the output before it is used; Epiq’s claimant agent is the exception, answering claimants directly from case information.
Legal tasks by what the agent is allowed to do. "Answers" means it replies from content; "assists" means it drafts or flags and a lawyer acts.
| Task | Level | What it touches, and the control we add | Seen in |
|---|---|---|---|
| First-pass NDA review against a playbook | Assists | Counterparty paper; suggested redlines and comments that an attorney refines | Proof (Ironclad) |
| Review and redraft of complex agreements | Assists | Matter documents; a lawyer reviews each change note | Dentsu (Harvey) |
| First-pass review for a contract service | Assists | The firm’s playbooks; in-line edits with the reasoning | Wilson Sonsini (Icertis) |
| Non-compete checks and data protection impact assessments | Assists | Customer data and assessment templates; a lawyer signs | Dentsu (agents built in Harvey) |
| Extract and compare terms across loan agreements | Assists | A loan portfolio; draft work product for lawyers | A&O Shearman (Harvey) |
| Answers on signed contracts, such as notice requirements | Answers | The contract repository, with the lawyer’s access | Proof (Ironclad) |
| Claimant questions in class action administration | Answers | Case information and claim status by file number; handoff to a call centre | Epiq (Copilot Studio) |
Most of these run inside a contract lifecycle or legal AI product. That matters for controls: a feature inside the contract system usually works within that system’s permissions, which is worth confirming with the vendor, while an agent you build yourself has to be given them.
What have companies published about legal agents?
Five organizations publish something specific. Two give time figures on a document type, one gives evaluation scores, and two give a quote or a volume. All are vendor stories or the firm’s own release, so read each figure as the publisher’s claim.
Publicly reported legal agents, as published by each source and opened on 2 November 2026. None of these organizations is a QueryNow client.
| Organization | What the agent does | Published figure | Source and date | How to read it |
|---|---|---|---|---|
| Proof | Ironclad’s Contract Review Agent gives attorneys a first-pass review of NDAs against Proof’s playbook, with suggested redlines and comments | 50% to 75% less time on NDA reviews (the headline says 75%); notice-requirement lookups take minutes instead of hours | Ironclad customer story, undated, read 2 November 2026 | Vendor claim; a range on one document type |
| Dentsu (250-person legal and compliance team) | Harvey for review and redrafting of complex agreements, plus in-house agents for non-compete checks and data protection impact assessments | 93% of lawyers use it weekly; 4 to 10 hours saved per lawyer per week, given as an estimate; 75% less time on the impact assessment process | Harvey customer story, undated, read 2 November 2026 | Vendor claim; usage is measured and the hours are estimated |
| Wilson Sonsini | Icertis Vera Engage gives a first-pass review of customer agreements for the firm’s commercial contracts service, with in-line edits and reasoning | 95% accuracy on first-party contracts, 92% on third-party contracts and 94% on issue detection, against a 90% target | Icertis customer story, undated, read 2 November 2026 | Evaluation scores; no production volume or time saved |
| A&O Shearman | Harvey agents break legal tasks into action plans and assemble draft work product, for example extracting and comparing loan terms across a portfolio | No measured figure; a partner says the agents do in minutes what took several hours | A&O Shearman news release, 6 April 2025 | A quote without a measurement |
| Epiq | Copilot Studio agents answer class action claimants on eligibility, filing and status, and hand off to the call centre | One agent handled almost 50,000 messages on one of Epiq’s largest cases | Microsoft customer story, 2025 | Volume only; no resolution or accuracy figure |
Two stories are left out. Microsoft’s legal team with Icertis carries an 83% headline that the story ties to statement-of-work creation, and the automation it describes is rule-based clause work. Benjamin Moore uses Ironclad’s agent to translate contracts between English and French and publishes no figures beyond "hours and hours of time" (Ironclad, 14 October 2025).
What controls does a contract review agent need?
Legal adds two controls to the usual set: a playbook a named lawyer owns, and a rule that no contract or legal advice reaches a counterparty or a client without a lawyer. Confidentiality then decides where the agent may run and what it may retain.
- Matter access. The agent reads only the matters and contracts the signed-in lawyer can open, so ethical walls hold. Use the contract system’s or the document management system’s permissions, never a shared service account that can see everything.
- Playbook ownership. A named lawyer owns the playbook, versions it and signs off changes. Every suggested redline cites the playbook position it applies, so a reviewer checks the rule as well as the edit.
- No automatic sending. Drafts stay drafts until a lawyer sends them, which is how every published contract review example works.
- Confidentiality. Check where prompts and documents are processed and retained and whether the provider may train on them. Client terms and data residency rules can rule out a model region.
- Acceptance test. Score the agent on 50 to 100 contracts the team already reviewed, clause by clause: issues found, issues missed and wrong edits. Wilson Sonsini’s scores against a 90% target show the shape of such a test.
Watch two signals in the first months. If the share of agent edits that lawyers keep climbs toward 100%, check whether they still read the comments or accept redlines in bulk. And when the team agrees a new position in a negotiation, update the playbook the same week, or the agent keeps flagging a clause the team now accepts.
Where should a legal team start?
Start with the highest-volume standard document, usually the NDA, against a playbook you already trust. Measure time per review before and after, and the share of the agent’s edits the lawyer keeps.
- Count last year’s contracts by type and by time spent. NDAs and standard supplier terms usually lead.
- Write or update the playbook for that type: preferred position, fallbacks and walk-away points, each with an owner.
- Run the agent on 50 past contracts and compare its issues and edits with what the team did. Where it disagrees for a good reason, fix the playbook.
- Go live with a lawyer reviewing every draft, and track time per review and edits kept against edits discarded.
- After a quarter, add a second contract type or questions over signed contracts, such as notice periods and renewal dates.
How do Microsoft, AWS and Google implement it?
Most legal agents run inside a legal product. If you build one, this is what each cloud provides, with statuses as the vendor pages showed on 2 November 2026.
Capability, Microsoft, AWS, Google
| Capability | Microsoft | AWS | |
|---|---|---|---|
| Where the examples ran | Epiq on Copilot Studio, Power Platform and Azure OpenAI | No legal example with figures in our research set | No legal example with figures in our research set |
| Build the agent | Copilot Studio (low code) or Microsoft Foundry Agent Service (pro code) | Amazon Bedrock AgentCore Runtime and Gateway (GA 13 October 2025) | Gemini Enterprise Agent Platform with ADK and Agent Runtime |
| Identity for the agent | Microsoft Entra Agent ID, created for every new Copilot Studio agent since July 2026 | AgentCore Identity with on-behalf-of token exchange | Agent Identity (SPIFFE-based); API in preview since 18 June 2026 |
| Evaluation against known answers | Copilot Studio agent evaluation (no preview label on 5 October 2026) | AgentCore Evaluations | Gen AI evaluation service |
Harvey, Ironclad and Icertis, the vendors behind four of the five examples, sell their own legal products. Govern them with the same matter permissions, identity and logging as an agent you build.
Frequently asked questions
Can an AI agent review contracts without a lawyer?
Not in any example we found. Every published contract review agent produces a first pass, such as redlines, comments or a summary, and a lawyer accepts, edits or rejects it. The time saved comes from starting at a marked-up draft instead of an unread contract.
What is a contract review playbook?
It is the written set of positions a legal team takes on each clause: the preferred wording, acceptable fallbacks and the points where it walks away. The agent compares a contract with it clause by clause, so the review can only be as good as the playbook.
How accurate are AI contract review agents?
Accuracy depends on your playbook and your contract mix, so measure it yourself. The one published set of scores, from Icertis for Wilson Sonsini, gives 95% on first-party contracts, 92% on third-party paper and 94% on issue detection against a 90% target. Those are the vendor’s evaluation results on that firm’s contracts.
Which contracts should stay out of scope at first?
Bespoke, high-value and litigation-sensitive documents. They have the fewest past examples to test against and the highest cost of a missed issue, and Proof’s published time figure comes from NDAs.
Who owns a contract review agent?
The general counsel or the head of legal operations owns its behaviour and the playbook; IT owns its identity, connectors and logging. Dentsu’s story quotes its head of legal and compliance operations, the role that usually runs these tools day to day.
More in Enterprise AI agent examples
Sources
Each source was read on 2 November 2026. The date shown is the date on the page, or the page’s ms.date for Microsoft Learn; an undated page shows the date it was read.
- Proof cuts NDA review times by 75% and gets contract answers in minutes with Ironclad AI
- How Dentsu’s legal and compliance team turned pressure into strategic advantage with Harvey
- Wilson Sonsini accelerates commercial contract review with Vera Engage
- A&O Shearman and Harvey to roll out agentic AI agents targeting complex legal workflows
- Epiq Global customer story (Microsoft Power Platform)
- Benjamin Moore customer story
- Microsoft customer story
- Evaluate agents in Copilot Studio
- What’s new in Copilot Studio
- Amazon Bedrock AgentCore is now generally available
- Gemini Enterprise Agent Platform release notes
- On-behalf-of token exchange (AgentCore Identity)
- Agent Identity overview
- What is Microsoft Foundry Agent Service?
- Evaluations for Amazon Bedrock AgentCore
- Gen AI evaluation service overview
Planning an AI program?
Talk to the team that builds it and runs it after launch.
- Built to SOC 2, HIPAA and GDPR standards
- EU AI Act-aligned delivery
- Microsoft Solutions Partner