The Legal AI Playbook (Choose Your Own Adventure)
Inspired by notes from the Anthropic Claude for Legal Webinar 2 (May 2026), originally captured and published by Helen Fan, Chief AI Officer and creator of Legal AI Value. You can find Helen's full four-page summary on her LinkedIn and it is well worth your time.
I did not make it to the webinar. Aussie hours being what they are, I was asleep when Anthropic's Mark Pike and Harry Liu walked a room of lawyers through what is now, I think, a genuinely watershed moment for our profession.
But I caught the notes the next morning, courtesy of Helen Fan, and then I spent the better part of the day playing around with claude-for-legal myself. One thing to say before we get into the substance of all of this.
This just fundamentally changed the game.
Not in a hypey, AI-bro, everything-is-disrupted kind of way. In the quiet, structural, you-will-not-be-able-to-unsee-this kind of way.
Here is the underlying premise I keep coming back to, the one that orients everything I am going to share with you below.
This is just the internet again.
Or email. Or the fax machine. Or the telephone before that. At every lockstep evolution in technology, our profession has faced the same choice: adapt, or slowly become irrelevant. Not overnight. Not dramatically. Just slowly, incrementally, and then all at once.
The lawyers who resisted email in the 1990s did not lose their jobs the next week. They just got gradually harder to work with, harder to compete against, harder to justify hiring at the same rate. The same thing is happening now, just faster.
Now, I want to say something clearly before we go any further, because I mean it: you do not have to do any of this. If you are at the tail end of your career and comfortable where you are, good for you. Genuinely. If you are in a position where you do not need to remain competitive in the market, that is a valid place to stand. Not every person reading this needs to become an AI power user by Tuesday.
But for everyone else? Let's break it down. Big Sis style. No fear, lots of curiosity, and a clear-eyed view of where you actually sit right now and what is genuinely useful for you to know.
Here is your choose-your-own-adventure. Find your lane. Some of these sections will overlap, because the fundamentals do not change much, but the emphasis and the action items are different depending on where you are standing.
🎓 You Are a Law Student
What this means for you right now
You are entering a profession that is mid-transformation. That is not a problem. That is actually an extraordinary piece of luck, because you have not yet built the habits that need unbuilding. You get to learn it the right way from the start.
Here is what Helen's notes from the webinar surfaced that matters most for you: Anthropic has built a specific plugin for law students and bar prep. It covers Socratic drills, case briefs, outline generation, IRAC grading, legal writing feedback and cold-call prep. This is not a novelty. This is a study partner that does not get tired, does not judge you for not knowing something, and is available at 2am before a tutorial.
But the deeper thing to understand, which Mark Pike kept emphasising across both webinar sessions, is this: the plugins are the starting point, not the finished product. The tool only becomes genuinely useful once it knows who you are, what you are trying to do, and how you want to work.
The cold-start interview is the key mechanism. When you first set up Claude for Legal, it asks you five categories of questions: your jurisdictions, your risk posture, your house style, your connected tools and your routing preferences. It writes a practice profile that every subsequent command reads before responding. For a student, this means telling it which country you are studying in, which subjects you are focusing on, what kind of lawyer you want to become, and what materials you already have access to.
Once it knows that context, it stops being a generic chatbot and starts being something much closer to a well-briefed tutor.
What to actually do
| Action | Why It Matters |
|---|---|
| Set up the cold-start interview properly | Everything downstream depends on this. Do not skip it. |
| Use it for IRAC drilling before tutorials | The feedback is immediate, specific and brutal in the best possible way. |
| Feed it your past essays and ask it to identify your patterns | Not to rewrite them. To show you where your reasoning consistently gaps out. |
| Practice cold-call prep by describing the fact pattern and letting it Socratic-question you | It is a much lower-stakes version of the real thing. |
| Pay attention to how it flags uncertainty | Watch when it says it cannot verify something. That skill transfers to how you read your own work. |
The one thing I would add that the webinar did not cover: your AI literacy is now a genuine differentiator when you graduate. The students who understand not just how to use this but why it works the way it does will be the ones partners actually want to invest in. Learn the methodology, not just the tool.
📋 You Are a Law Graduate (Admitted, Pre-Clerkship or Early Post-Admission)
What this means for you right now
You are in one of the most interesting positions of all. You are not yet embedded in a firm's way of doing things, which means you are not yet stuck doing things the old way. But you also do not have enough seniority to introduce change from above.
What you do have is time and access and, honestly, the most powerful thing of all: you are not threatening to anyone. A junior lawyer who quietly becomes the most AI-fluent person in the team does not get replaced. They get relied upon.
The most useful frame from the webinar for you is Mark's own story. He did not set out to become a builder. He pointed Claude at his legal ticket queue in Jira and Prospect, had it analyse 742 tickets to identify bottlenecks, and then started creating repeatable skills from what he found. He did not ask permission. He did not announce it. He just made the work better.
His colleague Pamela, who is not technical at all, asked Claude to build her a custom regulatory newspaper: summarising the publications she subscribes to, filtered by what matters to her practice area, delivered every morning. She is not a technologist. She just identified a pain point and asked for help solving it.
That is the move at your stage. Identify the repetitive, time-consuming, standardisable parts of your work. Then ask Claude to help you build a repeatable process for them.
What to actually do
| Action | Why It Matters |
|---|---|
| Map the Commercial Legal plugin to your practice area | Vendor agreement review, NDA triage, SaaS agreement review, renewal tracking. If any of this is in your day job, this is your starting point. |
| Start documenting your own patterns | Every time you review a contract, note what you are looking for. That is the raw material of a skill. |
| Learn what the tabular review workflow actually does | It ran through 30 contracts, extracted key terms, flagged severity levels and outputted everything to Excel with source citations in every cell. Understanding this workflow will make you a better reviewer, AI-assisted or not. |
| Ask about the privacy settings at your firm | On commercial plans, there is no training on your data by default. But consumer plan users need to toggle this off manually. Know which one you are on. |
| Be the person who reads the CLAUDE.md | If your firm is implementing any of this, that practice profile is where the institutional knowledge lives. Know what it says. |
One flag worth noting from the Q&A in Helen's notes: if you are dealing with clients or colleagues in other jurisdictions, the plugins are currently US-centric out of the box, but the cold-start interview can configure Claude to local law. For Australian practitioners, this is worth pushing on early.
⚖️ You Are an Early-Career Lawyer (Two to Six Years PQE)
What this means for you right now
You are at the career stage where efficiency and output quality are the two metrics that define your trajectory. You bill by the hour, your work gets reviewed, and the gap between what you produce and what a senior lawyer would produce is visible to everyone. AI does not close that gap automatically. But used well, it compresses the time it takes to close it yourself.
The architecture that Helen's notes pulled out from the webinar is worth understanding at this stage, because it maps directly to how good senior lawyers actually think.
Anthropic built claude-for-legal on four layers:
| Layer | Question It Answers | What It Does |
|---|---|---|
| Practice Profile | Who are you? | Writes a CLAUDE.md file that every command reads. Stops the AI from guessing and starts it behaving like a well-briefed colleague. |
| Skill | How to do it? | Decomposes a legal task into a step chain, with defined inputs, judgment criteria, output formats and gates between steps. |
| Agent | When to do it? | Background monitors on schedules: contract renewals, court dockets, regulatory feeds, product launches. The things that currently only live in a senior lawyer's calendar reminders. |
| Connector | Where is the data? | Links to your document management system, CLM, research tools and collaboration platforms. Knows where to look before you ask. |
What strikes me about this is how precisely it mirrors the knowledge a junior lawyer is trying to build. When you are two years in, you are learning the how. When you are four years in, you start to understand the when. The where takes longer because it is institutional, relationship-based and almost never written down anywhere.
Claude for Legal is an attempt to make all of that explicit and executable. That is useful for you because it names the thing you are trying to learn.
What to actually do
If you are in commercial law, look at the Commercial Legal and Corporate Legal plugins first. Tabular due diligence, disclosure schedules, closing checklists and NDA triage are all in there. If you are in employment, the Employment Legal plugin covers offer letter review, termination risk checks, wage and hour Q&A and new-country employment setup.
Regardless of your practice area, use the tabular review skill to understand what good review looks like structurally, not just to generate output. Read the flagged cells. Ask why something was flagged medium versus high.
Start paying attention to your hallucination literacy. The three mitigation layers from the webinar are: the cold-start interview providing firm-specific context upfront, built-in verification steps in every skill, and guardrails that alert you when Claude is making assumptions and prompt you to manually verify citations. Mark's framing: "We want to make sure you're not the next lawyer who ends up on the naughty list of people who submitted hallucinated citations in a courtroom filing."
The last point is the one I cannot say loudly enough. AI does not remove your professional responsibility. It changes the nature of the supervision you need to apply.
🏢 You Run Your Own Practice or Work in a Small Firm
What this means for you right now
This is where I think claude-for-legal is most transformative, and it is also where the adoption curve is steepest. Because the people who need this most are often the ones with the least time to set it up.
The access to justice section of the webinar is the part that I keep sitting with. Free Law Project, Descrie and Courtroom5 are now live as free law connectors, because Mark's point was unambiguous: "Frontier AI for legal shouldn't only belong to firms with huge innovation budgets." Over 80% of civil litigants show up without a lawyer. That statistic is not just a social problem. It is a market signal.
For small practice operators, here is the thing: you are already doing the work of three or four people. The question is not whether you can afford to implement this. The question is whether you can afford not to compress the time you spend on standardisable work.
The plugins most immediately relevant to small practices:
| Plugin | What It Does |
|---|---|
| Commercial Legal | Vendor agreement review, NDA triage, business-readable contract summaries, renewal tracking. If you are advising SMEs, this is your day job. |
| Litigation Legal | Matter intake, case portfolio dashboard, demand letters, subpoena triage, litigation holds, fact timelines, brief section drafting. |
| Privacy Legal | DPA review, privacy impact assessments, regulatory gap analysis. If you have clients asking about data compliance, this is the plugin that makes you look like you have a whole privacy team. |
The Q&A from the webinar surfaced something specifically relevant here: Mark acknowledged single-seat commercial licensing is in development. He said "I hear you loud and clear" on the demand, which is a reasonable signal that it is coming. Worth watching.
In the meantime, the Team plan is where the privacy-by-default settings apply, there is no training on your data, and you get access to the full plugin suite. For a small firm, even if it is just you and a paralegal, this is worth the investment.
What to set up first
The cold-start interview is more important for a small firm than anyone else, because you are the firm. Your risk posture, your preferred clause positions, your escalation thresholds, your document management system: all of this needs to be in that practice profile. Take the time to do it properly. Everything the AI produces downstream will be shaped by what you put in there.
Connectors second. If you are using iManage, Box, DocuSign or Westlaw, these are all available as MCP connectors now. The point of the connector layer is that it knows where your data lives without you having to paste it into every conversation. That is the practical time saver.
🏗️ You Are In-House Counsel
What this means for you right now
The demo from the webinar was built for you. Harry Liu played the role of in-house counsel at a fictional company called Pluto Co., acquiring a target (Acme) in a virtual deal room with 30 contracts in Box. The tabular review skill ran through all 30 contracts, extracted key terms, flagged severity, and outputted everything to Excel with source citations in every cell. If a cell could not be verified, it was flagged rather than fabricated.
Then Claude created the Excel file in Cowork, Harry opened it in Excel, and the same Claude session continued with full context. No copy-paste. No re-explaining. Mark called it the "Food Network moment": watching the ingredients being assembled, then pulling the finished dish out of the oven.
Two contracts were flagged as blocking for data rights issues. Harry asked Claude to pull the exact clause language, insert it into a new column, and add deep links back to the source documents in Box for manual verification. Then, from the same session, Claude drafted a deal-lead briefing with a summary action table and posted it to a private Slack deal channel, but only after asking permission first.
Mark's framing for what this actually means: "It's still me doing the legal work. Claude handles the tedious formatting and synthesis so I can focus on strategic decisions."
That is the in-house value proposition in one sentence.
The five managed agents worth knowing about
These are background monitors that run on schedules. If you are an in-house lawyer, these are the things that currently live only in your head, your calendar reminders and your anxiety.
| Agent | What It Watches | Why It Matters |
|---|---|---|
| renewal-watcher | Contract repository for renewal/cancel-by deadlines | Missed cancel windows = auto-renewed contracts and wasted spend |
| docket-watcher | Court dockets for new filings, deadlines, deliverables | Missed court deadlines = malpractice exposure |
| reg-monitor | Federal Register, agency RSS, regulatory feeds | Regulatory changes caught early = proactive compliance |
| diligence-grid | Virtual data rooms for new uploads | New documents auto-extracted into due diligence grid |
| launch-radar | Jira/Linear/Asana for upcoming product launches | Products shipping without legal review = risk |
The launch-radar agent is the one that made me stop and sit still for a moment. If you work in tech or product-adjacent environments, you know exactly what this is solving for. Legal being the last to know a product shipped is one of the oldest, most exhausting structural problems in in-house practice.
What to prioritise
If your firm uses Microsoft 365, this is particularly relevant: Claude now works inside Word, Excel, PowerPoint and Outlook, carrying matter context across all apps. For Word, this means drafting, redlining and clause-by-clause comparison against your playbook, all in tracked changes. For PowerPoint, it means turning legal analysis into board presentations. For Outlook, it means communicating legal guidance in your voice and tone, without starting from scratch every time.
Mark's key point: "We're not asking you to learn a new tool. We're putting Claude inside the tools you already use."
The privacy note for in-house teams is also worth flagging explicitly: on commercial plans (Team, Enterprise), there is no training on your data by default. But if anyone on your team is using the consumer plan, they need to toggle training off manually. As Mark put it: "As a lawyer, you have an ethical obligation to understand how these tools work."
The Part That Stays With Me
Helen Fan's notes pull out something from the webinar that I keep returning to, and I want to close with it because I think it is the most important thing in this whole document.
The plugins are not the product. The methodology is.
Four layers, four questions. Who are you? How to do it? When to do it? Where is the data?
Traditional knowledge management tried to make tacit legal knowledge explicit for twenty years and largely failed. Lawyers do not write SOPs because billable hours are better spent reviewing contracts. This system takes a different approach: instead of asking lawyers to write documentation, it asks them to answer questions and feed in seed documents. The AI extracts the patterns from actual work product. The result is not a static manual. It is a living, executable system that updates as practice evolves.
The twelve plugins are also, as Helen put it, a map of where AI will hit legal practice hardest. Anthropic built what it built first because it has the data on where lawyers spend the most time on standardised, repeatable work. That map is worth reading regardless of whether you ever open a plugin.
Don't be scared. Stay curious.
If you want to go deeper on any of this, you know where I am.
This briefing was written by Mel Storey, founder of Counsel Media and Career Big Sis. It was inspired by notes taken by Helen Fan (Chief AI Officer, Creator of Legal AI Value) from the Anthropic Claude for Legal Webinar 2, held on 16 May 2026. Helen's original four-page summary is available on LinkedIn and is the kind of note-taking that makes the rest of us look like we are paying attention in meetings when we are not. Full credit to her.
