Of all the teams at Anthropic, the Legal team might be the most surprising adopter of Claude Code. Legal professionals are trained to be methodical, risk-aware, and precise — qualities that sometimes exist in tension with the experimental, rapid-iteration ethos that tools like Claude Code tend to encourage. And yet, Anthropic's legal team has become genuine advocates for the tool, using it to build custom applications, automate internal workflows, and even create accessibility tools for family members.
Their path to Claude Code started with personal curiosity and experimentation, not a top-down directive. And the perspective they bring — as lawyers who think carefully about security, privacy, and risk — adds an important dimension to how Anthropic thinks about Claude Code adoption at enterprise scale.
Custom Accessibility Solution
The most personally significant Claude Code project from the legal team didn't start as a work project at all. A team member wanted to build a communication assistant for family members with speaking difficulties — a predictive text application with speech-to-text capabilities and personalized voice banks.
Commercial accessibility tools exist, but they often don't fit specific individual needs — too complex, missing specific features, or not customizable enough for particular communication styles. Building a custom solution would traditionally require either significant technical expertise or expensive custom software development.
Using Claude Code, the legal team member built a working application in approximately one hour. The app includes predictive text, speech-to-text input, and voice bank functionality — features that genuinely fill a gap in existing commercial tools for this family member's specific situation.
This experience — building something genuinely useful for a real person in an hour — created a fundamentally different understanding of what's possible. It's one thing to read about non-technical users building applications with AI assistance. It's another to actually do it.
Legal Department Workflow Automation
Inside the legal department, the team has built tools that address the specific coordination challenges that legal work involves.
One significant project: a "phone tree" system to connect team members with the right lawyer at Anthropic based on the nature of their question. In a growing legal department covering many different specializations — employment law, intellectual property, privacy, regulatory compliance, contracts — figuring out who to contact for a specific question is itself a friction point. The phone tree system routes requests automatically based on the type of question, reducing the overhead of getting to the right expert.
Building this would traditionally require either engineering resources or using a third-party workflow tool that may or may not fit the specific requirements. Claude Code made it buildable by the legal team itself, without either dependency.
Team Coordination Tools
The team has also built G Suite applications that automate recurring coordination tasks:
A weekly team update system — instead of manually collecting updates from team members and assembling them into a report, a button click triggers collection, aggregation, and formatting. The legal team members provide structured input; the system handles assembly and distribution.
A legal review status tracker — instead of maintaining a spreadsheet manually and updating it after each review, the system tracks review status automatically as changes are made. Status visibility that previously required spreadsheet discipline now happens as a byproduct of the existing workflow.
These aren't flashy applications. But they represent real weekly time savings for a team that has better uses for its time than manual administrative tasks.
Rapid Prototyping for Solution Validation
Lawyers working on novel problems — which describes most interesting legal work — often face the challenge of proposing solutions to domain experts before knowing whether those solutions are technically viable. The traditional approach is verbal or written proposals, which can be hard to evaluate without seeing something working.
The legal team has started using Claude Code to build functional prototypes quickly, specifically for the purpose of showing domain experts something tangible to respond to. An example: building a prototype accessibility application to show specialists at UCSF what the tool could do, getting their expert feedback on the interaction design and feature priorities before investing more development time.
This rapid prototyping workflow changes the feedback cycle from "here's what I'm thinking about building" to "here's a working version — what would you change?" The latter produces dramatically better input.
Planning in Claude.ai, Building in Claude Code
The legal team has developed a two-phase workflow that reflects their methodical professional background and that other non-technical users may find particularly useful.
Phase one: spend time in Claude.ai brainstorming and planning. Work through the full design of what you want to build — the user flow, the requirements, the edge cases, the data it needs to handle, the security considerations. Don't write any code. Just think.
Phase two: ask Claude.ai to distill that planning conversation into a comprehensive, step-by-step prompt for Claude Code. Then take that prompt into Claude Code and implement one step at a time — not all at once.
The "one step at a time" constraint is deliberate. Legal professionals are trained to understand that mistakes compound, and that catching problems early is much cheaper than fixing them late. Implementing incrementally means each step can be verified before building on it, which produces more reliable results than trying to implement everything at once.
Security Awareness and Enterprise Implications
As product lawyers, the legal team brings something to the Claude Code conversation that most technical teams don't: an acute awareness of the security and compliance implications of the tools they're using.
The team has been explicit about their concerns with deep MCP integrations — scenarios where Claude Code has broad access to sensitive systems. As AI tools access more sensitive data (contracts, financial records, personnel files, privileged communications), the security posture of those integrations matters enormously. Conservative security postures will create real barriers to adoption, and those barriers are appropriate given the sensitivity of the data involved.
The legal team has become advocates internally for building compliance tools quickly as AI capabilities expand — recognizing that the governance frameworks need to keep pace with the technology. Their perspective as both users and lawyers gives them credibility on these questions that purely technical advocates may lack.
Team Impact
- Non-technical professionals building production-quality applications in hours — a fundamentally different relationship with technology
- Accessibility solutions filling specific gaps that commercial tools don't address, built by the person who understands the need
- Legal workflow automation built and maintained by the legal team itself, without engineering dependencies
- Faster validation of ideas through functional prototypes, leading to better-informed expert feedback
- Internal advocacy for thoughtful security and compliance frameworks as AI tools gain access to sensitive systems
Top Tips From the Legal Team
Plan extensively in Claude.ai before moving to Claude Code. This is the legal team's most strongly recommended practice. Flesh out your entire idea — requirements, edge cases, user flows, potential problems — in Claude.ai before writing a single line of code. Then ask Claude.ai to summarize that planning conversation into a comprehensive step-by-step implementation prompt. Arriving at Claude Code with a clear plan produces dramatically better results than trying to figure out what you're building while you're building it.
Work incrementally and use screenshots liberally. Ask Claude Code to implement one step at a time — not the whole thing at once. After each step, use screenshots to show Claude what the current state looks like and what needs to change. Claude excels at reading visual interfaces and making targeted adjustments. This incremental visual workflow gives you checkpoints to verify progress and catch problems before they compound, and it makes the process much less overwhelming for non-technical users.
Share prototypes even when they're imperfect. The legal team has found that sharing rough, unfinished work sparks conversations and collaboration that wouldn't happen otherwise. A working prototype — even one with rough edges — communicates possibility in a way that descriptions and mockups can't. It prompts domain experts to engage concretely with what they're seeing, which produces specific, actionable feedback. Don't wait until something is polished to share it. The value of a prototype is the conversation it starts, not the artifact itself.