AI employees for legal teams

Most AI that reached legal teams in the last few years was a tool. You open it, type a question or paste a clause, and it answers. Then you close it, and it forgets. The work itself, the reading, checking, redrafting, filing and chasing, stays with you.

An AI employee works the other way round. You give it a task, not a prompt: review this supplier agreement against our positions, read the data room and tell me what matters, check every contract for renewals due next quarter. It does the work on its own, comes back when it needs a decision, and hands you the finished result. Then it remembers what it learned for next time.

The names vary. Some call them AI workers, some AI teammates, some AI agents. What they share is the shape of the job: a request comes in, the work goes out finished, and a person stays accountable for what is sent. That is the shape of in-house legal work already. Requests arrive from across the business, most of them variations on the same few contracts, and a small team has to get through all of them.

So the useful question for a legal team isn't whether AI can draft a clause. It can. The question is whether it can take a piece of work off someone's desk and bring it back done.

1. The request lands. Someone in sales posts a customer's paper in a Slack channel, or a lawyer asks in the app. A schedule can start the work too: every weekday at seven, or the first of the month.
2. It reads, and asks if it must. It opens the documents on its own computer, works out what is being asked and checks its notes on the people, the company and the deal. If something is unclear, it asks, as a colleague would.
3. It does the work against your positions. It reviews the contract against your playbook, applies your fallbacks and drafts the changes as tracked changes in Word, with a comment explaining each one.
4. It hands parts off. A large job splits. One envoy reads the data room while another checks the contracts in it, and a third builds the index. They pass work between them like a close-knit team.
5. It brings decisions back to you. Before anything leaves the team, a post in Slack, an update in a connected tool, a message to another envoy, it asks. You decide where it needs your sign-off and where it needn't.
6. It files the record and remembers. The signed contract and its terms go into the team's records. What it learned, a new fallback, a preference about format, goes into its memory, so you explain it once.

Claren gives a legal team envoys: AI employees, each with a name, a desk and a job. The product is Claren Envoys, and it was built for in-house legal work from the start: requests from the business, the same contracts again and again, a small team, and a budget someone signs.

Every envoy has its own computer, with a browser, files, and Word, Excel, PowerPoint and PDF, tracked changes included. It runs on the model you choose for it, from OpenAI, Anthropic or Google. It has skills, the playbooks it follows, and a memory of how your team works. You can type to it, or talk to it if you'd rather.

What makes an envoy an employee rather than an assistant is everything around the model. It works when nobody is at the keyboard. It works with other envoys. It keeps the team's records. And it belongs to a team, not to one person.

• Its own computer. A browser and files of its own, and the Office formats lawyers actually send. It works on the document itself, not on a copy pasted into a chat. More
• Memory. People and their roles, facts about your company, your preferences down to the format, decisions and why they were made, and your playbook, followed as written. More
• Routines. Work before you ask. An envoy runs a routine on a schedule, when a message lands in a Slack channel, or when you press start. Each run is its own session, so you can see what it did. More
• Delegation. An envoy can hand part of a task to another envoy, or run several at once, and brings decisions back to you before anything goes out. More
• Teams. Teams have managers and members, and can be private. Each team has its own envoys, a shared folder, its own datastore and its own Slack channels, so legal's work never turns up in sales. More
• Connectors. Envoys read from and write back to Notion, Linear, Jira and Confluence, or any tool with an MCP server. More
• Datastores. Structured records, tables and files that envoys design, fill and keep current as they work. People read them; envoys keep them correct. More
• Apps. Ask for a contract tracker, a renewals dashboard or a review screen, and an envoy builds it on your datastore. More
• Slack. Envoys take requests in DMs and channels. Link a channel to a team, and its envoys pick up what lands there. More

• Contract review. Every contract against your playbook, returned as tracked changes in Word with a comment on every change, your fallback positions applied and a summary of what's off-playbook. More
• NDAs. The highest-volume, lowest-judgement contract a legal team sees. A routine watches the requests channel, and each NDA comes back marked up to your positions, waiting for a lawyer to approve.
• Renewals and obligations. Envoys pull renewal and notice dates, payment and reporting duties, termination windows and owners out of your contracts into one tracker. A routine checks it each morning and tells you what's coming, in the app or in Slack. More
• Due diligence. The whole data room read, not sampled: an index, the key terms of each document, issues flagged with their source, and a report you can hand on. More
• Intake from the business. Requests posted in a Slack channel are picked up by the team's envoys as they land. The business asks where it already works; legal sees the work arriving done.
• Records and your own CLM. Every contract and its terms filed in a datastore as envoys work, with an app on top for the views your team wants. More
• Cover when people are away. Holidays and absences stop being a backlog. The envoys know the playbook and the history, so the work keeps moving.

On the first day, an AI legal assistant and an AI employee look alike. Both read a contract. Both suggest changes. Both write a clean summary. The difference is what happens next.

An assistant waits. It does nothing until a lawyer opens it, and what it produces is a draft for that lawyer to check, finish, save, file and send. Whatever it was told about your positions lives in that lawyer's session, or their personal settings. The next lawyer starts again, and the team's way of working depends on who happened to ask.

An employee carries on. It picks up the next NDA from the channel without being asked. It applies the fallback the GC approved last week, because it was told once. It files the signed contract, adds the renewal date to the tracker and flags it ninety days out. Its playbook is the team's, so every review reads the same.

For a single lawyer with an occasional question, an assistant is good enough. For a team with more requests than hours, the work only comes off the desk when something else takes it.

Legal teams hand Claren their NDAs and supplier contracts and get tracked-change redlines back against their playbook. Financial documents, HR policies and the long tail of small requests from the business go the same way, picked up from Slack and handed back finished.

Through an acquisition, a whole data room is read, indexed and reviewed. Signed contracts and their terms go into the datastore, renewals are flagged before they roll, and the work carries on through absences and holidays. The team gets hours back every week.

The pattern is the same in each. The work that repeats goes to the envoys. The lawyers keep the judgement calls, and the sign-off.

1. Pick the work that repeats. Start with one stream: NDAs, supplier contracts or renewals. Volume and sameness are what an AI employee is best placed to take.
2. Give it your playbook. Your positions, fallbacks and escalation points go into the envoy's memory once. No written playbook? Teach it your positions as you go, and it keeps them. See contract playbooks.
3. Put it in a team. Create the legal team, add its managers and members, and give it a shared folder and a datastore. Make it private if the work needs to be.
4. Connect where the requests come from. Link the Slack channel the business already uses for legal requests. Add connectors to the tools where your work lives.
5. Set its routines. Pick up every contract that lands in the channel. Check renewals each morning. Report open matters every Friday. Each routine runs on its own, and each run is kept.
6. Decide what needs your sign-off. By default an envoy asks before anything leaves the team. Keep that for the work that matters, and loosen it where you trust the result.
7. Review the first weeks closely. Read what comes back, correct it and say why. Each correction goes into memory, so the second month needs less of you than the first.

Claren holds a SOC 2 Type II report and ISO/IEC 27001:2022 certification. Data is encrypted at rest with AES-256 and in transit with TLS 1.2 or higher, and stored in US, UK or EU data centres, as you choose.

Each organisation's data is kept apart, with no cross-contamination between matters or organisations, and an audit trail of every user action. Teams draw the internal lines: each envoy sees what its team can see, and nothing more.

Neither Claren nor its model providers train AI models on your data. The only thing your work improves is your own envoys' memory and playbooks. Zero data retention and enterprise SSO come with the Enterprise plan. See security.

Claren is free for one person, with private envoys, a personal computer, scheduled routines and app connections. Pro is $70 per seat per month for one to three people, with $50 of usage per seat included. Team is $250 per seat per month for three or more, with $200 of usage per seat pooled across the team and shared team envoys. Enterprise is priced with you, and adds zero data retention and enterprise SSO.

Usage covers the models, the envoys' computer time and storage. On Pro and Team, anything beyond what's included is pay-as-you-go at your plan's rate. See pricing.