Claude for Legal Now Has 90 Pre-Built AI Agents. Here's Where a Small Firm Starts.

Claude for Legal Now Has 90 Pre-Built AI Agents. Here's Where a Small Firm Starts.

A transactional attorney at a 10-person firm used to spend 45 minutes reviewing vendor agreements for red flags before she got to the substance. Now there's a pre-configured agent called "Vendor Agreement Reviewer" that does the first pass before she opens the document. That's one of 90+ named agents now live in Claude for Legal — deployed in under 10 minutes, no engineering team required.

On June 1, Hanson Bridgett — an AmLaw 200 firm with approximately 200 lawyers — announced firm-wide Claude deployment: attorneys, operations, HR, finance, and knowledge management, all in, with a written AI use policy. If a 200-lawyer regional firm is running Claude across every department with governance in place, the question for a 10-person firm isn't "is this ready?" It's "which three agents do I start with?"

This guide answers that question.

---

What Claude for Legal Actually Is in 2026 (Not the May 12 Version)

When Claude for Legal launched on May 12, 2026, it had 12 practice-area plugins. Three weeks later: 90+ named workflow agents. That's not an update — it's a different product.

Here's what matters for a small firm:

Where the agents live. The full library is in Anthropic's public GitHub repository at github.com/anthropics/claude-for-legal. Open, no vendor contract, no procurement process. You can browse the entire library before deciding whether to deploy anything.

What "agent" means here. A Claude for Legal agent is pre-configured for a specific legal task. It knows the expected inputs (a contract, a record, a policy document), the output format, and the legal context to apply. You don't write the prompt from scratch each time. You open the agent, feed it your document, and it runs.

The "no engineering required" reality. For query-based agents — those you run on demand — this is accurate. Paste the configuration from GitHub into Claude, and it works. Modification happens in plain language: add a line saying "flag provisions that deviate from California law" and the agent adjusts. No code.

Cost. Claude Pro ($20/user/month) or Claude Team (starting at $30/user/month, with enterprise data protections). For a 5-attorney firm, full team access runs $100–$150/month total. Enterprise legal AI platforms run significantly higher — the cost differential is real and meaningful for a firm under 25 attorneys.

For background on the May 12 launch — the original plugins and MCP connectors — see Claude for Legal: What a 5–20 Attorney Firm Should Actually Do With It.

---

The 90-Agent Library: How It's Organized by Practice Area

The agents are named for the task, not the technology. Here's a working map of the practice areas and example agents in each:

Practice Area	Example Agents
Corporate / Transactional	Vendor Agreement Reviewer, M&A Due Diligence Organizer, Term Sheet Analyzer
Employment	Termination Reviewer, Separation Agreement Reviewer, Employee Handbook Policy Checker
Litigation	Case Summary Builder, Deposition Prep Assistant, Claim Chart Builder
Compliance / Privacy	DSAR Responder, Data Breach Notification Drafter, Privacy Impact Reviewer
AI Governance	AI Policy Gap Analyzer, AI Contract Clause Reviewer
Regulatory	FCPA Checklist, Export Control Screener
IP	Patent Claim Reviewer, Prior Art Summary Builder
Legal Education / Training	Research Memo Drafter, Brief Summarizer

Not every practice area is equally developed in the current library. Corporate/transactional and employment agents tend to be more thoroughly built than some regulatory-specific agents, which may require jurisdiction-specific customization. Before building a workflow around a specific agent, run it on two or three real matters from last month and evaluate the output quality for your practice context.

One note for general practice firms: modification happens in plain language. Any agent can be adjusted for a specific firm's workflow, jurisdiction, or matter type. Start with the closest match to your primary work and tune from there.

---

The Active Agent Shift: What Continuous Monitoring Means for Small Firms

This is the capability that separates Claude for Legal 2026 from every "AI assistant" pitch of the past three years.

Most AI tools are query-based: you bring the document, ask the question, get the answer. The attorney controls the moment of interaction. That works for one-off research or single contract review. It doesn't change your baseline.

Active agents work differently. They run on a schedule or trigger without a prompt from the attorney. The most cited example in the library: an agent that "performs a weekly sweep of signed agreements for playbook deviations." You configure it once. Every week, it reviews your signed agreements against your firm's standard playbook and surfaces deviations — before a client calls you about them.

For a compliance or employment practice, the implication is direct: configure an agent to run a periodic check of matter files for approaching deadlines, or monitor incoming client communications for high-priority items. The attorney still reviews the flags and decides what to do. Active agents surface items; attorneys resolve them.

The honest setup note. Active agents are more complex to configure than query-based agents. The pattern matching and trigger logic require careful initial setup to avoid false positives or missed items. If you're new to Claude for Legal, start with query-based agents. Build a baseline for output quality in your specific practice context. Layer in active monitoring once you know what to expect from a given agent.

The active/query distinction also changes how you measure ROI. A query-based agent saves time on a specific task each time you run it. An active agent removes an entire category of manual review from your weekly workload. That's a structural change in how the firm operates, not a productivity increment.

---

The Three Agents to Deploy First (by Practice Type)

Don't start with 90 options. Start with three.

Transactional / Corporate practices:

1. Vendor Agreement Reviewer — immediate time savings on routine contract first-pass before attorney review
2. Term Sheet Analyzer — faster turnaround on deal intake and kick-off documents
3. M&A Due Diligence Organizer — structure before substance on any transaction with a large document set

Employment Law practices:

1. Termination Reviewer — catch procedural issues in termination documentation before they become claims
2. Separation Agreement Reviewer — consistency across separation matters, especially for multi-location employers
3. Employee Handbook Policy Checker — governance check when state employment law updates require policy review

Litigation practices:

1. Case Summary Builder — prep from large record volumes when you receive discovery
2. Deposition Prep Assistant — witness outline from documents already in your possession
3. Claim Chart Builder — particularly useful for IP litigation and patent-focused work

General / Mixed practices:

1. Vendor Agreement Reviewer — applies to any firm managing contracts on behalf of clients or receiving vendor agreements of its own
2. DSAR Responder — any firm handling client data under CCPA, GDPR, or state privacy law equivalents
3. AI Policy Gap Analyzer — governance check every firm needs now, regardless of practice area

The AI Policy Gap Analyzer deserves emphasis across all firm types. Hanson Bridgett deployed Claude firm-wide with a written governance policy, not without one. Freshfields did the same. The firms running AI at scale are not operating without guardrails — they built the policy first and expanded from there. If your firm doesn't have a written AI use policy, the AI Policy Gap Analyzer is a defensible starting point for building one.

---

The Hanson Bridgett Signal: What a 200-Lawyer Firm's Decision Means for Your 10-Person Firm

On June 1 — the same day the Claude for Legal library hit 90+ agents — Hanson Bridgett announced firm-wide Claude deployment. Not just for attorneys: operations, marketing, HR, finance, and knowledge management, all running Claude as part of their regular workflows. Specific uses: summarizing depositions and lengthy records, drafting routine correspondence and memos, comparing document versions, supporting due diligence in corporate transactions.

What they implemented alongside the rollout: a written AI use policy specifying what information can and cannot be used in AI systems, enterprise-grade data protections, and ongoing internal workflow reviews. COO Laura Long's stated goal was "building long-term capability across the firm" — not running a pilot.

The small-firm read: Hanson Bridgett is not BigLaw in the Kirkland or Skadden sense. It's a 200-lawyer regional firm — sophisticated, but operating closer to the mid-market than to the top of the Am Law 100. When a firm of that profile goes all-in across every department with governance in place, the "is Claude ready for serious legal work" debate is over.

Two firms in three weeks — Freshfields (UK Magic Circle) and Hanson Bridgett (AmLaw 200) — committed to firm-wide Claude deployment with written policies. The mid-market validation has arrived.

The governance piece is the lesson that transfers directly to small firms. Deploying Claude without a written policy creates operational risk: attorneys use it inconsistently, confidential matter information may be processed outside approved data boundaries, or client-facing documents contain AI-assisted content without disclosure. The legal AI pricing and platform context shows where Claude sits in the cost stack — but policy is what makes firm-wide adoption defensible. A 10-person firm can draft a one-page policy in an afternoon. That's the governance foundation the 90-agent deployment runs on.

---

How to Get Started: A 30-Minute Deployment Path

No engineer. No contract. No implementation timeline.

Step 1 — Open the repository. Go to github.com/anthropics/claude-for-legal. Browse by practice area using the folder structure. You don't need a GitHub account to read it.

Step 2 — Pick three agents. Use the practice area map above. Find three agents that match your most common matter types. Open each agent's configuration file. It's written in plain English — the task description, expected inputs, and output format are all readable without technical knowledge.

Step 3 — Open Claude.ai. You need Claude Pro ($20/month) or Claude Team. In a new conversation, paste the agent configuration from GitHub. That's the setup. No installation, no coding.

Step 4 — Run it on a real matter from last week. Take a contract, record, or policy document you worked on recently. Run it through the agent. Compare the output to what you would have done manually. Note specifically: what did the agent catch that you'd have caught anyway? What did it miss? What would you adjust for your jurisdiction?

Step 5 — Document what works. If the output is useful after two or three test runs: write one paragraph in your firm's AI policy draft noting this as an approved workflow. You're building institutional knowledge, not just running a test.

The honest expectation. The first run will not be perfect. The agent may flag items you'd dismiss or miss something specific to your jurisdiction. That's calibration. The question after three test runs is: does this agent save time on this task with the adjustments I've made? If yes, it belongs in your workflow. If no, move to the next agent on the list.

---

What to Watch For: Honest Limitations

Output quality varies by practice area. Some areas in the library are more fully developed than others. Corporate/transactional and employment agents have more depth; some regulatory-specific agents will require more customization for jurisdiction-specific work.

"No engineering required" means plain-language modification, not guaranteed accuracy. Every output requires attorney review before it goes into a client communication or filing. Claude produces confident-sounding output that can miss jurisdiction-specific nuance. The 90-agent library accelerates deployment — it doesn't replace professional judgment on the output.

Active agents require careful initial setup. If you configure an active agent to monitor your agreement stream and the pattern recognition doesn't align with your firm's standard documents, you'll get noise instead of signal. Build the baseline on query-based agents first; add active monitoring when you understand what a specific agent surfaces in your practice.

The library is actively growing. Anthropic moved from 12 to 90+ agents in three weeks. Check the GitHub repository monthly. If your practice area is underserved today, it may not be in 60 days.

---

One Action

Pick one agent from the three-agents-per-practice-area list above. Open the Claude for Legal GitHub repository, find it, paste the configuration into a Claude Pro or Team conversation, and run it on a real matter from last week.

You're not committing to a platform. You're finding out whether there's a workflow in those 90 agents that saves you time on work you're already doing. That's a 30-minute test with no downside.

If the test works: add a single paragraph to your firm's AI policy draft documenting it as an approved workflow. If it doesn't: pick the next agent on the list.

The firms running AI across every department right now — Hanson Bridgett, Freshfields, Grant Thornton on the accounting side — started with one workflow that worked. The second workflow followed from the first. The firm-wide rollout followed from the second.

Pick the first one.

---

Last verified: June 4, 2026. Agent library and pricing subject to change — check github.com/anthropics/claude-for-legal for current documentation. Sources: Artificial Lawyer (June 1, 2026); Anthropic GitHub github.com/anthropics/claude-for-legal; Research Brief #73 (June 4, 2026).

---

The Crossing Report tracks AI moves in professional services every week, filtered for firm owners who don't have time to follow every announcement themselves. Subscribe free — get the top 3 insights from every issue.