Introducing Agents in Lupl 

Build and deploy AI agents that work alongside your attorneys as teammates. Work on individual tasks together or orchestrate entire processes with human supervision.

AI is transforming legal work in ways we couldn’t have imagined even two years ago.

But most AI assistants share a fundamental limitation: they wait to be asked. You pose a question, you get an answer, and then the tool sits idle until the next prompt. If a new request comes in while you’re in back-to-back meetings, that task doesn’t move forward until you return to your desk.

The real opportunity is to transition from AI as a system for answering questions to AI as a system that takes action. If 2025 was the year of the Assistant, 2026 will be the year of the Agent.

This shift is coming, and it raises important questions about how to orchestrate, manage and supervise agents for legal work. This is the problem we’re solving with Agents in Lupl.

How Agents Work

A Lupl agent is an AI teammate that lives inside a matter alongside other users. Because it exists in the same workspace as your people, tasks, documents, and workflows, it operates with real context from the start. Agents can observe events as they happen, propose actions, and complete defined work within the boundaries you set.

AI teammates, natural collaboration

Agents collaborate the same way humans do inside Lupl. You can assign tasks to them, mention them in comments, or include them in workflow automations.

Proactive by default

Agents respond to events, not just prompts. So, for example, when an email arrives, an agent can:

• Classify it
• Extract instructions
• Assign tasks and create deadlines based on agreed SLAs

Agents can even be embedded within your playbooks, so that when a matter is created, the right agents and attorneys are automatically added every time.

Full matter context

Because agents live inside Lupl matters, they have full context by design.

They can see:

• The matter structure
• Tasks, phases, and deadlines
• Documents and communications
• Prior decisions and commentary

There is no need to restate background or paste content into prompts every time. And granular permissions apply to agents exactly as they apply to people, so an agent only sees what it has been explicitly allowed to access by firm administrators.

Powerful tooling

Lupl agents have an API-driven tooling architecture that allows an agent to be connected to almost any system. This will differ from firm to firm and the screenshot above is for example purposes.

Just like human teammates, agents get access to a powerful array of tools. For example, a Drafting Agent might be given partial instructions in an email thread:

• It runs a company registry search to confirm party details
• It sends an API call to your Document Automation tool to generate the drafts
• It saves them to iManage then assigns an approval task to you in Lupl

Agents are API and MCP compatible, so you can hook them up with almost any other system.

Human supervisors

Every agent in Lupl has a human supervisor who remains accountable.

Agents can draft, propose, and execute work within defined boundaries, but supervisors review outputs, approve actions, or override decisions at any point. All activity is logged for audit purposes.

This reflects how legal teams actually delegate: responsibility can be shared, but accountability cannot be handed off.

Compliant by design

Legal teams do not need more shadow AI. Admins retain complete control over agents, including:

• Who can create agents
• Which agents can be used on which matters
• What data agents can access
• What tools agents can use, and what actions they can take

Every agent action is audited and logged. Data is never used to train models. And Agents are built on an open framework in Microsoft Azure, so you can connect almost any model – even your own privately-deployed Azure service.

Use cases

Intake and triage

An agent monitors incoming requests. It extracts requests from emails or forms, classifies tasks, extracts key details, and assigns to the right teammate.

Doc automation

An agent reviews a client’s instructions on a new real estate matter, extracts the key data points, and sends them to the firm’s doc automation service, generating draft documents in the firm’s house style and automatically assigning them to a human supervisor for review.

Task orchestration

An agent watches task progress. When something slips, it proposes reassignments, updates dependencies, and alerts the right stakeholders.

Scope and budget management

An agent reviews new work requests against the agreed scope and budget, and notifies the partner if things go off track.

Why this matters now

Law is shifting from an expertise business to an outcomes business. Up to 70% of legal work can now be automated and client expectations are changing. Tremendous progress has already been made with AI, but further transformation requires AI to be integrated into processes. We believe the firms that win in the future will be those that build complete delivery systems that combine humans and AI to produce predictable outcomes for their clients. By embedding agents directly into matters, Lupl makes AI part of how work is delivered, not just how questions are answered.

FAQs

What are AI agents for law firms?

Agents are like AI teammates that combine instructions, knowledge and tooling to take action on legal matters.

When can we get access?

Agents are rolling out in private preview to selected law firm design partners. Contact us for a demo or to request access.

Are Lupl agents fully autonomous?

No. Agents operate under defined permissions and always have a human supervisor. They can propose and perform work, but accountability stays with people.

How are agents different to tools like Harvey or Legora?

AI in legal today is primarily focused on supporting tasks such as summarization and tabular review. These tasks are critical to delivery but they are still just discrete tasks. The value we are adding is integrating tasks into a complete matter process, combining agents with human teammates and a project management framework that ensures everyone has visibility, everything stays on track, and humans are in the loop. We are working with selected customers to explore integration of tools like Harvey and Legora so that an agent in Lupl can interact with dedicated tools via API/MCP.

Does it work with Outlook?

Yes! You can assign emails to an agent directly from Outlook – they get the context from the email, including any attachments, and combine it with the matter context to move things forward.

Can agents access sensitive matter data?

Only if explicitly permitted. Agents inherit Lupl’s permission model and cannot access anything outside their assigned scope. This is managed by firm administrators.

Who can create agents?

Administrators control this entirely, deciding who can create agents, where they can be used, and what capabilities they have.

What is the tech stack?

Agents are built on a model-agnostic architecture in Microsoft Azure, allowing you to switch models or even connect your own services.

What is the cost?

We are working with customers to explore pricing models during the preview release.

Can agent behavior be changed over time?

Yes. Agents are configured through natural language and can be adjusted as your workflows evolve. Agents will also iterate themselves as a particular matter evolves.

Are agent actions auditable?

Yes. All agent activity is logged in Lupl, just like human activity, providing a clear audit trail.

Ready to learn more?

Agents are rolling out in Q1. Book a demo and see them in action.