Enterprise Legal Project

Cowork Legal Plugin

AI-powered legal assistant embedded directly in the company's Cowork workspace. Answers legal and policy questions in context, connected to the company's knowledge base. Department-level access controls enforce data boundaries.

Status: Configured
Platform: Glean + Cowork
Departments: All 4 (access-controlled)
Users: Enterprise-wide
The Problem

Why it was built

Employees across the enterprise had legal and policy questions — about contract terms, compliance requirements, company policies, data handling rules. The answers existed in the knowledge base but finding them required knowing where to look — which document, which system, which version. Most people didn't know. They'd email Legal, wait days, and get a response that could have been self-serve.

The Cowork workspace was where people already worked. Embedding legal AI there meant meeting users where they already were rather than asking them to navigate to a separate legal portal. The access-control challenge was nontrivial: employment legal data needed to be invisible to commercial legal users, and vice versa.

Before
Employee has a legal question. Searches the intranet. Finds outdated policies. Emails Legal. Waits days. Response may or may not cite the authoritative source. Legal team spends hours on answerable-by-document questions.
Email → wait → hope
After
Employee types the question in Cowork. Plugin queries the appropriate knowledge sources — scoped to their department's access level. Returns answer with source citation. Legal team focuses on judgment questions, not lookup questions.
Ask → answer in seconds
The Solution

What it does

A Glean-powered plugin embedded in the Cowork workspace interface. Users ask legal and policy questions in natural language. The plugin queries the company's knowledge base — policies, playbooks, contract templates, compliance docs — and returns answers with source citations. Access controls at the department level ensure employment legal data stays walled off from commercial legal, and vice versa (ABA Model Rule 1.6).

Unlike the task-specific agents (separation agreements, metrics), this is a horizontal capability — any legal question, any department, any policy. The plugin routes the question to the right knowledge sources based on the user's department and access level. Legal teams can update the knowledge base without touching the plugin configuration.

Why this mattered: This was the pattern for scaling legal AI across the enterprise. Task-specific agents solve point problems. A workspace-embedded plugin with access controls solves the distribution problem — getting AI into the hands of every employee, in the tool they already use, with the data boundaries Legal requires. The plugin model means Legal owns the knowledge, not the implementation.
The Impact

Metrics & improvements

4
Departments covered
Self-serve
No Legal email needed
RLS
Department-level access
In-context
Where users already work

Improvement: Common legal and policy questions answered in seconds within the existing workspace, not days via email. Legal team reclaims hours spent on lookup questions. The access-control architecture proved the pattern for horizontal AI deployment — any tool, any department, one knowledge base, enforced boundaries. The plugin model is the distribution strategy.

← Back to Index