Working Like a Lawyer with Claude

1. How Lawyers Actually Start Using Claude

Mark Pike is Legal Counsel at Anthropic, leading the team that reviews enterprise products including Claude Code. Anna Gressel is Global Co-Head of AI at Freshfields, one of the world’s leading law firms — with roughly 5,700 staff and lawyers worldwide already using AI tools. Their conversation is valuable precisely because it is between a sophisticated internal user of Claude (a lawyer at the company that makes it) and a sophisticated external advisor (a lawyer who guides enterprise clients on AI adoption).

Their observation about how lawyers typically start is revealing: “A lot of lawyers, when they first start using Claude, are just using it for these sort of chat based question and answer experiences.” This is the entry point — ask a question, get a useful answer. It is valuable but it is also the surface of what is possible. The analogy would be getting email access for the first time and using it only to send short messages without ever attaching a document, sharing a calendar, or connecting it to other tools.

The more powerful capabilities that lawyers are slower to adopt involve Claude acting as an agent on their local environment: creating Word documents, Excel spreadsheets, and PowerPoint presentations directly. Instead of asking Claude a question and then manually creating an artifact based on the answer, a lawyer can ask Claude to create the artifact itself. This changes the unit of work from “generate text I will then format” to “produce the finished document I will then review.” The workflow becomes materially faster.

2. The Context Gap: The Hardest Problem in Legal AI Adoption

Gressel identifies “the context gap” as one of the stickiest and trickiest areas for lawyers to navigate. The context gap is the challenge of keeping up with rapid changes in the AI products and capabilities that legal clients are using and asking about. “What does the spec look like today? And the fact that it might look different tomorrow and your engineering teams are moving at the speed of light?”

For a lawyer advising an enterprise client on AI risk — what they can do, what they cannot do, what exposure they are taking on — the foundational challenge is that the underlying system keeps changing. A legal analysis completed three months ago may no longer be accurate because the model, the product, or the regulatory environment has shifted. Legal advice is traditionally stable: a contract interpretation does not change week to week. But AI product advice exists in a domain where the facts of the matter are in continuous motion.

This context gap problem also affects internal legal teams at technology companies. Anthropic’s own legal team reviews new features and products as they come out — meaning the lawyers need to develop genuine technical fluency, not just legal knowledge. Mark Pike’s team is in the business of reviewing things that did not exist a year ago, using frameworks that are still being developed, with regulatory clarity that is still emerging. The context gap is not a temporary onboarding problem; it is a structural feature of the legal environment for AI.

3. Where Claude Adds the Most Legal Value

Both Pike and Gressel point to the same capability as the highest-leverage: synthesizing large bodies of information and spotting themes across them. “If there’s one thing that I feel like AI is very good at, it’s making sense of large bodies of information and spotting themes across them.”

The practical legal application is uploading multiple source documents — existing memos, product roadmaps, new legislation, regulatory guidance — and asking Claude to create a new artifact based on what it sees across all those surfaces. This is a task that previously required a junior associate to spend days reading and synthesizing, producing a draft that a partner would then review and substantially revise. With Claude, the same task can produce a draft in minutes that a lawyer can then refine.

The value here is not just speed. It is also coverage. A human reviewing fifty documents under time pressure will inevitably miss some things, prioritize recent over older material, and be influenced by the order in which they encountered the documents. Claude applied to the same corpus does not have these biases — it can hold all fifty documents in its attention simultaneously and surface patterns that a human reading serially would be likely to miss.

4. The Rudder Stays With the Lawyer

The most important structural point that both Pike and Gressel make is about what AI does not replace. Gressel’s metaphor is precise: “The role that we play, ideally for our clients, is to help them sail a ship. Sometimes they’re sailing through a really rocky strait and there might be, you know, rocks to the left and rocks to the right. And they’re being slammed by, you know, hail and all of these things. But at the end of the day, our judgment is the thing that keeps the rudder straight.”

In this metaphor, AI might help gather information about where the rocks are, produce analysis of previous voyages through similar straits, and draft the navigation plan. But the lawyer’s judgment — the decision about which way to steer when the situation is uncertain, the advice about whether to proceed or turn back, the liability assessment about what exposure the client faces — remains with the human. AI can make the lawyer faster and better-informed; it cannot substitute for the professional judgment that is the actual product of legal services.

This is consistent with the observation about junior associates: “Sometimes they have such an inherent understanding of, you know, the benefits, but also the limitations and the right process for checking and being in the loop.” The lawyers who are best positioned to use AI effectively are those who understand not just what AI can do but what it cannot do — who maintain appropriate skepticism, run appropriate verification checks, and stay in the loop on outputs rather than treating AI as a black-box authority.

Gressel closes with an Anthropic value that she finds apt for the legal context: “holding light and shade.” The technology is both genuinely risky and genuinely promising. Lawyers who can hold both realities simultaneously — who can help clients capture the promise without ignoring the risk — are the ones who will provide the most value in an AI-shaped legal landscape. The ability to see both sides clearly, rather than being either an AI pessimist or an AI enthusiast, is itself a kind of professional judgment.

🧠 Check your understanding

5 questions · your answers are saved in this browser only

1. 1. According to Pike and Gressel, what is the most common entry-level way lawyers use Claude?

   Automated contract review and red-lining Chat-based question and answer experiences Generating complete legal briefs from scratch Real-time legal research with citation checking

   They specifically note that 'a lot of lawyers, when they first start using Claude, are just using it for these sort of chat based question and answer experiences' — the entry point that is valuable but represents the surface of what's possible.

2. 2. What is the 'context gap' in legal AI adoption?

   The gap between what AI can do and what lawyers think it can do The challenge of staying current with rapidly changing AI products and capabilities when providing legal advice about them The difference in context window sizes between different AI models The gap between how lawyers and non-lawyers understand AI risk

   The context gap refers to the difficulty of advising clients on AI when the underlying products, capabilities, and regulatory environment change faster than traditional legal analysis cycles — 'what does the spec look like today and it might look different tomorrow.'

3. 3. Where does Claude add the most value in legal work according to Gressel?

   Providing definitive legal opinions that reduce the need for partner review Automating standard contract templates for routine transactions Making sense of large bodies of information and spotting themes across multiple documents simultaneously Conducting real-time analysis of ongoing court proceedings

   Gressel specifically identifies this capability: 'If there's one thing that I feel like AI is very good at, it's making sense of large bodies of information and spotting themes across them' — synthesizing across multiple documents to create a new artifact.

4. 4. What does Gressel's 'sailing through rocky straits' metaphor mean for the role of lawyer judgment?

   Lawyers should be cautious about recommending AI tools that could be risky AI provides the map, but the lawyer's judgment is the rudder that steers the client through uncertainty Legal AI is still too early-stage to use in high-stakes situations Clients need to be protected from their own decisions about AI adoption

   The metaphor illustrates that AI can help gather information and prepare analysis, but 'at the end of the day, our judgment is the thing that keeps the rudder straight' — the professional judgment about how to advise remains with the lawyer.

5. 5. What does 'holding light and shade' mean in the context of lawyers advising clients on AI?

   Using both light and dark mode when working with AI tools Being able to hold both the genuine risks and the genuine promise of AI simultaneously, rather than being purely pessimist or enthusiast Balancing time between AI-assisted work and traditional legal research Maintaining both confidentiality and transparency when using AI for client matters

   Gressel borrows this Anthropic value to describe the orientation she thinks best serves lawyers: 'there's so much risks in this technology and that as lawyers, we often help our clients navigate that risk. But there's also a lot of promise.' Holding both realities is the professional stance.

Reset quiz

🛠️ Build it yourself

Follow these exact steps to reproduce it yourself · estimated time: ~40 min

Prerequisites

• Access to Claude (claude.ai or API)
• Three to five real documents from a single subject area (memos, reports, articles, regulatory text)

Step 1 — Start with a single-document Q&A

Upload one document and ask Claude a specific factual question about its contents. This is the entry-level use case. Notice the quality and speed of the response compared to reading the document yourself.

Step 2 — Expand to multi-document synthesis

Now upload three to five related documents on the same topic. Ask Claude to identify the key themes across all of them, note where they agree and where they differ, and flag anything surprising. This is the synthesis capability Gressel identifies as highest-leverage.

Step 3 — Ask for an artifact, not just an answer

Instead of “what are the key points,” ask Claude to draft a one-page summary memo as if you were preparing it for a client or colleague. Specify the format, tone, and what to include. Evaluate the draft against what you would have written yourself.

Step 4 — Test the context gap problem

Find a document about an AI product or regulation from six months ago. Ask Claude to assess what has likely changed since then. Notice where it is helpful (flagging areas of uncertainty) and where it is limited (it cannot know what has actually happened since its training cutoff).

Step 5 — Identify where your judgment adds the most value

After working through the above steps, write down: what did Claude do that genuinely freed you to do something else? And what decisions required your professional judgment that you would not trust Claude to make alone? This is your personal map of where the rudder stays with you.