Perplexity has released Computer for Counsel, an agentic AI system designed to automate the administrative and research work that consumes roughly three-quarters of lawyers' time. The product, available now to Perplexity Enterprise and Max subscribers, doesn't try to replace legal research databases like Westlaw or LexisNexis. Instead, it sits as an orchestration layer that connects to the tools lawyers already use, routing tasks across multiple AI models and grounding every answer in verifiable sources.

Why Are Lawyers So Interested in AI Agents Right Now?

A Thomson Reuters survey found that nearly 75% of lawyers identify administrative tasks as a major time challenge. Computer for Counsel targets that exact problem. The system decomposes a legal task into smaller subtasks, routes each one to the best-suited AI model and data source, then assembles the results into a brief, memo, or deal summary. Every output links back to its source, so attorneys can verify citations in seconds before the work reaches clients.

The architecture is deliberately multi-model. Computer for Counsel selects from 20 or more frontier AI models, choosing the best one for each subtask automatically. Research, reasoning, and contract work can each use a different model, which removes the pressure on legal teams to bet their entire workflow on a single AI vendor. This approach contrasts with traditional legal research platforms, which rely on proprietary stacks and lock teams into one vendor's ecosystem.

How Does Computer for Counsel Actually Work in Practice?

The system operates through the Model Context Protocol (MCP), an open standard for linking AI systems to external tools and data. Administrators can install custom MCP connectors for internal systems, and the platform ships with 400 connectors that reach Microsoft 365, Google Workspace, and other common business tools. Inside Microsoft 365, Computer drafts in Word, retrieves files from SharePoint, and references context from Outlook or Teams conversations.

Three concrete workflows demonstrate how the agentic pattern works in real legal practice:

• Third-party NDA intake: The system reviews third-party non-disclosure agreements for red flags, fills in entity and signatory information, prepares clean copies, and routes them for approval and signature via DocuSign.
• Regulatory monitoring: Computer builds a shareable dashboard for US state privacy and adtech laws, showing which states have laws in effect and citing relevant cases from Midpage.
• Case research with citation review: The system researches non-compete enforceability following the FTC's 2024 ban, summarizes key cases, flags unsettled ones, and exports a PDF with citations.

What Legal Sources Does the System Connect To?

Computer for Counsel integrates premium legal sources that ground answers in authoritative data. Midpage provides US case law at the federal and state appellate level, statutes, regulations, and a citator to verify whether a case is still good law. Deel offers compliance data on worker classification, employment rules, immigration, and cross-border payroll across 150 countries. LegalZoom provides contract templates for customer agreements, employment contracts, and NDAs.

Additional connectors include DocuSign for contract workflows, NetDocuments and Box for document management, DeepJudge for institutional intelligence grounded in a firm's prior work, and Clio for legal research across more than one billion legal sources in over 100 jurisdictions. Carta and Ironclad connectors handle equity data and contract repositories. Several connectors, including Clio and Ironclad, are still listed as coming soon, indicating the platform is still expanding its integrations.

The system explicitly does not train on company data at the Enterprise tier, meaning connected files stay under firm control. Early adoption has been strong; at Gunderson Dettmer, 80% of lawyers actively use Perplexity Enterprise, generating 35,000 queries per month.

How Does This Differ From Existing Legal AI Tools?

Perplexity is not trying to out-research Westlaw or LexisNexis. Instead, it targets the work before, around, and after formal legal research. The real differentiator is reach into the open web and everyday firm tools. Multi-model routing is no longer unique; other platforms like Harvey also route across vendors. But Computer for Counsel's integration with 400 MCP connectors puts it directly inside the tools lawyers already use daily.

Every output links to a source for one-click verification, a critical feature in legal work where citation accuracy is non-negotiable. The system removes the pressure of single-vendor lock-in by automatically selecting the best model for each subtask. However, good-law checks still depend on Midpage coverage, and lawyers must verify every citation before it enters client work. Judgment and strategy remain with the attorney; the system handles the busywork.

Perplexity also exposes its cited search primitive publicly through the Sonar API, an OpenAI-compatible interface that returns sources with every answer. Domain filters let users restrict grounding to trusted sites, the way a lawyer would curate research sources.

What Are the Limitations and Open Questions?

Computer for Counsel is not a standalone citator, and good-law checks depend on Midpage's coverage of case law. Several connectors are still listed as coming soon, meaning the platform is still expanding its integrations. Perplexity also faces unresolved data-sourcing lawsuits, and web grounding can miss paywalled or unpublished opinions that traditional legal databases capture.

The product ships as a finished tool, not a software development kit (SDK), which means legal teams cannot heavily customize the underlying logic. However, the Sonar API is available to developers who want to build cited search into their own applications, offering a foundation for custom legal AI workflows.

For legal teams drowning in administrative work, Computer for Counsel represents a shift in how agentic AI can be deployed in knowledge work. Rather than replacing human judgment, it automates the routing, research, and assembly tasks that consume time without adding strategic value. The 75% of lawyers struggling with administrative burden now have a tool designed specifically for their workflow.