Six Tips for Building AI Workflows with Claude for In-house Legal Teams

Banner artwork by Samuel Boivin / Shutterstock.com

How can legal departments use Claude to improve workflows?

Experienced legal operations manager Midori Vasquez provided in-house counsel and legal operations professionals with a practical look at how Anthropic’s Claude AI tools can support day-to-day legal work during an ACC webcast, Claude Crash Course: AI Workflows for In-House Legal Teams.  

Below are six actionable tips for in-house counsel and legal operations professionals who are using Claude in their corporate legal function.

1. Know which model you are using

Not every task requires the most powerful model. Vasquez recommended that legal teams use Sonnet for most routine legal operations work to save on token costs. Reserve the most powerful models, such as Opus, for complex, multi-step analysis.

As a general rule:

• Use Sonnet most of the time. It is faster, more cost-effective, and well-suited for routine work such as contract review, email drafting, research summaries, document organization, and matter folder management.
• Use Opus selectively. Save it for work that requires deeper reasoning across multiple dependent steps, such as M&A due diligence, complex litigation analysis, or regulatory strategy.
• Match the model to the risk and complexity of the task. A first-pass summary of a vendor NDA likely does not need the same model as a cross-border regulatory risk assessment.

The key is intentional model selection. Legal departments should avoid defaulting to the most expensive option when a lighter model can produce the needed result.

2. Keep tokens in check

Claude Cowork can use more credits than standard Claude Chat. For example, Claude may run several background operations for one prompt (this may include reading files, querying connectors, and executing sub-tasks).

For legal teams experimenting with workflows, that usage can add up quickly. In-house teams should build awareness into their AI governance and budgeting practices:

• Start with focused prompts. Ask for the specific output needed, rather than loading unnecessary files or asking broad exploratory questions.
• Limit unnecessary context. Do not ask Claude to review an entire matter folder when only one contract, policy, or memo is relevant.
• Use structured workspaces. A well-organized folder system helps Claude load the right materials without wasting tokens on irrelevant documents.
• Work with administrators. On Team or Enterprise plans, legal operations professionals should coordinate with system administrators to set appropriate defaults, monitor usage, and ensure attorneys have enough credits for high-value workflows without creating runaway costs.

For legal operations teams, token management is not just a technical issue. It is part of responsible resource allocation.

3. Save rules and memory locally

One of the most practical takeaways from the session was the value of local instruction files. Vasquez explained that Claude Cowork can use markdown files to preserve rules, preferences, and institutional knowledge.

Two files are especially important:

• CLAUDE.md: The “constitution” for the legal AI workspace. This file should contain standing instructions, routing logic, tone preferences, ethics guardrails, and department-wide rules.
• MEMORY.md: A file for evolving facts, such as matter status, stakeholder preferences, recurring project updates, or attorney drafting preferences.

The distinction matters. Rules should live in CLAUDE.md; changing facts should live in MEMORY.md.

That separation keeps the system clean and prevents standing instructions from becoming cluttered with outdated matter details.

Examples of useful local rules include:

• “Always label AI-drafted legal work product as ‘DRAFT — Attorney Review Required.’”
• “Never send emails or messages without explicit approval.”
• “Use plain English and avoid cliché wording.”
• “Flag non-standard indemnity, limitation of liability, assignment, confidentiality, and governing law clauses.”
• “Cite source documents when summarizing case law, policies, or contract terms.”

Legal teams should also avoid overloading these files. A bloated instruction file can slow the system and waste tokens.

The goal is not to document everything; it is to document the rules and context that repeatedly improve the quality of outputs.

4. Leverage connectors and plugins

Connectors help Claude link to the systems where legal work already happens. These may include emails, calendars, document management systems, collaboration tools, matter management platforms, and shared drives.

For in-house teams, the practical benefit is that Claude can move from a standalone drafting assistant to a workflow tool connected to live business context. Plugins can also accelerate adoption by packaging commands, skills, and recommended connectors for specific roles or workflows.

A legal plugin, for example, may include pre-built support for NDA review, matter intake, billing audits, compliance tracking, or precedent research.

The governance point is just as important as the productivity point: Start with the least access necessary.

A connector used for inbox triage may only need read access. A connector used for document review may not need permission to create, rename, or delete files. Legal teams should define permission levels intentionally before scaling connected workflows.

5. Level up with skills

Skills allow legal teams to turn repeatable work into reusable playbooks.

Instead of prompting from scratch every time, attorneys and legal operations professionals can create standardized workflows that run through a slash command or scheduled task.

A practical approach is to build skills in four steps:

1. Do the work manually first. Run the workflow in chat and refine the prompt until the output is reliable.
2. Capture the workflow. Ask Claude to turn the refined process into a repeatable skill.
3. Trigger it with a command. Use a slash command such as /nda-review, /matter-intake, /billing-audit, /board-minutes, or /daily-legal-brief.
4. Automate it when ready. Once the workflow is stable, schedule it to run on a defined cadence.

This approach is especially valuable for legal playbooks that already exist. A department’s contract review checklist, billing guidelines, board minutes template, or matter intake protocol can become the foundation for a skill.

The best practice is to refine before automating. If a workflow is inconsistent in chat, it will be inconsistent as a skill.

Legal teams should identify edge cases, test outputs, and define review points before relying on automation.

6. Schedule repeatable tasks

Once a workflow is reliable, scheduled tasks can help legal teams keep work moving without waiting for a manual prompt.

Vasquez highlighted several examples that could be useful for in-house counsel and legal operations teams.

Potential cadences include:

• Hourly: Inbox triage, draft responses for review, monitor new matter assignments, or check billing exceptions.
• Daily: Prepare a morning legal brief combining calendar items, emails, matter updates, and relevant legal news.
• Weekly: Generate outside counsel spend reports, matter status summaries, compliance checklist updates, or legal operations dashboards.

The right cadence depends on the workflow, risk level, and budget. An urgent inbox triage process may justify hourly review.

A spend report may only need to run weekly. A matter update summary might be most useful ahead of a recurring legal department meeting.

Legal teams should also remember the practical requirements: Scheduled tasks depend on the local setup.

If the workflow runs through the desktop app, the computer must remain on, and the app must remain open. Enterprise teams should coordinate with IT on power settings, access controls, and approved deployment models.

Most importantly, scheduled tasks should not remove attorney review. For any workflow involving legal analysis, privileged material, outbound communication, or file changes, human approval should remain built into the process.

More AI resources for in-house counsel

Claude’s value for in-house legal teams will not come from one-off prompts alone. The greater opportunity is in building repeatable, governed workflows that improve over time.

Check out more in-house insights in the ACC AI Center of Excellence for In-house Counsel, including AI Use Cases from companies like Elastic. And download the Claude Crash Course white paper in the ACC Resource Library.

This article was developed with the assistance of Generative AI.