New AI Agents for Legal Professionals

1. Why legal work is a natural fit for specialized AI agents

Legal practice has always been text-heavy in a way that few professions can match. A single commercial transaction might produce hundreds of pages of agreements, schedules, exhibits, correspondence, and negotiation notes. A litigation matter can generate thousands of documents over its lifetime. Even a routine contract review for a mid-market SaaS vendor requires a trained reader to scan for non-standard indemnification language, check liability caps against your company’s playbook, flag missing representations, and note any jurisdiction-specific gotchas — a task that might take a junior associate two to three hours when done carefully.

What makes legal work particularly amenable to AI assistance is that it is simultaneously high-volume and highly structured. Contracts follow templates. NDAs have a predictable architecture: definition of confidential information, obligations of the receiving party, exclusions, term and termination, remedies. Patent filings have required sections. Court briefs follow local rules. That structure means AI agents can be trained to understand exactly what to look for, where to look for it, and how to flag what they find in a format that a senior attorney can review quickly. The combination of well-defined document structure, repeatable workflows, and extremely high attorney billing rates makes legal tasks an ideal target for automation.

The economics tell the story clearly. A junior associate billing at $300–$500 per hour who spends three hours reviewing an NDA represents $900–$1,500 of cost for what is often a commodity task. A specialized AI agent that handles the same review in under twenty minutes, produces a structured redline with plain-English explanations of each change, and flags the two clauses that are genuinely non-standard for attorney attention — that changes the unit economics of legal service delivery substantially. Firms and in-house teams that adopt these tools don’t eliminate lawyers; they redeploy them from document processing to higher-value judgment work.

2. The 12 one-click legal plugins

Anthropic organized the legal plugin suite into four practice area groupings, each reflecting the distinct workflows, document types, and reference materials that attorneys in that area rely on every day.

Commercial legal covers the contracts that most in-house legal teams spend the majority of their time on. The contract review assistant reads any agreement and produces a structured summary flagging key terms, non-standard provisions, and missing clauses — sorted by risk level. The vendor NDA redliner applies your organization’s standard NDA playbook and generates a tracked-changes redline automatically, with a comment on each change explaining the business rationale. The SaaS agreement analyzer focuses specifically on software-as-a-service contracts, checking for data processing terms, uptime commitments, security obligations, auto-renewal clauses, and price escalation provisions that are commonly buried in lengthy enterprise software agreements.

Corporate legal addresses governance and transactional work. The board meeting minutes drafter takes an agenda and action-item notes and produces draft minutes in the format your jurisdiction and governance documents require. The M&A due diligence checklist plugin reads the target company’s disclosed documents against a configurable checklist and produces a gap analysis showing what has been provided, what is missing, and what raises follow-up questions. The entity management summarizer digests your organization’s entity structure and produces a plain-English summary of subsidiary relationships, registered agents, and annual filing obligations — useful when anyone from finance or operations asks “what entities are we responsible for?”

Intellectual property plugins serve both prosecution and portfolio management. The patent landscape researcher searches public patent databases and produces a landscape summary for a given technology area, identifying key players, filing trends, and potential freedom-to-operate risks. The trademark clearance assistant runs a preliminary clearance analysis for a proposed mark, surfacing potentially conflicting registrations and common-law uses that would warrant a more thorough professional search. The IP portfolio summarizer reads your patent and trademark portfolio data and produces a high-level summary organized by technology area, expiration date, and maintenance status.

Litigation plugins support the pre-trial and briefing workflows that consume enormous associate time. The case law researcher takes a legal question and jurisdiction and produces a memo-style summary of controlling authority, distinguishable cases, and current circuit splits. The deposition prep assistant reads deposition transcripts, prior declarations, and the relevant pleadings to generate a preparation outline for an upcoming witness deposition. The brief summarizer reads any court filing and produces a structured summary of the claims, legal standards, key facts, and argument structure — useful for quickly getting oriented to a new matter or understanding an opposing filing.

3. What “one-click” setup actually means

The phrase “one-click” is doing meaningful work here. Legal AI tools in the past required substantial configuration: writing or adapting a system prompt for your organization’s policies, connecting API keys to legal research databases, building output templates that match how your team formats deliverables, and integrating with your document management system. For a legal team without dedicated engineering support — which describes most in-house legal departments and many boutique firms — that setup barrier was prohibitive.

Each plugin in the Cowork suite bundles all of that configuration. The system prompt encodes legal domain expertise for the specific task: the NDA redliner knows what a standard mutual NDA looks like and what deviations commonly favor the other party. The case law researcher knows how to structure a legal memo, how to format citations, and how to flag when the law is unsettled or jurisdiction-specific. The output templates match the professional formats attorneys actually use. The connections to legal reference databases — including pattern libraries based on Westlaw and LexisNexis research structures — are pre-wired. Connecting your account activates all of it simultaneously.

This is the “one-click” in the name: one click connects your workspace, and you get twelve production-ready workflows immediately. There is no prompt engineering, no template customization, no database configuration required to get started. Teams can begin with the plugin as shipped and customize it over time as they develop preferences — but the default configuration is designed to be immediately useful.

4. Legal-specific guardrails and safe use

The legal domain has essentially zero tolerance for confident errors. A fabricated case citation in a brief could result in sanctions. A missed clause in a contract review could expose the company to liability. Anthropic designed the legal plugins with guardrails that reflect these stakes explicitly.

Every plugin output includes source citations for factual and legal claims. When the case law researcher produces a memo, every case cited includes a reference that can be verified. When the NDA redliner flags a clause as non-standard, it references the specific provision in your playbook that the clause deviates from. This is not optional — the citation requirement is built into the system prompt for every legal plugin.

The plugins also explicitly flag uncertainty. When the law in a particular area is genuinely unsettled, the output says so rather than presenting one position as controlling. When a clause is unusual but the plugin’s confidence that it is problematic is low, it flags the clause with a lower severity rating and notes the ambiguity. When a situation appears to fall outside the scope of the plugin — an extremely complex cross-border transaction, a novel legal question, a jurisdiction the plugin’s training doesn’t cover well — the output escalates to human review with a plain-English explanation of why the situation requires attorney judgment.

Critically, every plugin output carries a reminder that the results are not legal advice and must be reviewed by a qualified attorney before being relied upon. This is not a liability disclaimer in the fine print; it is a prominent part of every output. The plugins are designed to function as a very capable first-pass reviewer, not as a replacement for the attorney who ultimately signs off.

🧠 Check your understanding

5 questions · your answers are saved in this browser only

1. 1. Which of the following best describes what makes legal work well-suited to specialized AI agents?

   Legal documents are short and simple, making them easy to process Legal work involves high text volumes, well-defined document structures, and repeatable workflows Attorneys prefer AI agents over junior associates in all circumstances Legal databases like Westlaw are not accessible to humans, only to AI

   The combination of large document volumes, predictable structures (NDAs, contracts, briefs all follow templates), and repeatable workflows makes legal tasks especially amenable to AI automation — without replacing attorney judgment on complex matters.

2. 2. What does 'one-click' mean in the context of the legal plugins?

   You only need to click once per document to generate a review The plugin bundles system prompt, database connections, and output templates so no configuration is needed beyond connecting your account Each plugin only performs one specific action and cannot be combined with others One click purchases a one-year license for all 12 plugins

   One-click means the entire stack — system prompt, legal domain expertise, database connections, and output templates — is pre-configured. Teams connect their account and get 12 production-ready workflows immediately.

3. 3. Which practice area does the 'deposition prep assistant' plugin belong to?

   Commercial legal Corporate legal Intellectual property Litigation

   The deposition prep assistant is one of three litigation plugins. It reads prior transcripts, declarations, and pleadings to generate a preparation outline for an upcoming witness deposition.

4. 4. How do the legal plugins handle situations where the law is genuinely unsettled?

   They pick the most common position and present it as controlling They refuse to answer and ask the attorney to research manually They flag the uncertainty explicitly in the output rather than presenting one position as definitive They default to the most conservative interpretation in all cases

   The plugins are designed with legal-specific guardrails that flag uncertainty explicitly. When the law is unsettled, the output says so — rather than presenting one view as controlling — and may escalate to human review.

5. 5. What time reduction is typically achievable when using a legal plugin for a task a paralegal or junior associate would spend 3 hours on?

   2 hours 45 minutes (5 minutes saved) About 2 hours (1 hour saved) About 20 minutes (saving roughly 2 hours 40 minutes) Instantaneous — zero human time required

   The plugins can compress a 3-hour paralegal task to approximately 20 minutes — not zero, because attorney review of the output is still required, but dramatically faster. The saved time is redeployed to higher-value legal work.

Reset quiz

🛠️ Build it yourself

Follow these exact steps to reproduce it yourself

• Connect your Claude Cowork account at cowork.claude.ai if you haven’t already
• Navigate to the Plugins section and locate the “Legal” category
• Click “Connect” next to the Vendor NDA Redliner — this activates the plugin with your account
• Find an NDA your team recently received from a vendor (use one that is already executed so there’s no risk in testing)
• Open a new Cowork session and attach the NDA document
• Type: “Review this NDA using the Vendor NDA Redliner plugin and flag any provisions that deviate from our standard playbook”
• Review the output: check that each flagged clause has a source citation and a plain-English explanation
• Compare the plugin’s redline to a redline your team previously did manually on a similar agreement
• Note where the outputs agree (high confidence areas) and where they differ (areas to calibrate the playbook)
• Document two or three playbook additions that would improve the plugin’s accuracy for your organization’s specific standards