The Short Answer
An in-house legal department should run five AI agents: contract intake and triage, NDA turnaround, matter intake from the business, outside-counsel invoice review, and employee policy Q&A. On ibl.ai you own all the code and the data, so the department runs those agents model-agnostic inside its own perimeter, and the business that sends the requests is billed by usage, with no per-seat pricing.
The difference from law-firm AI is who the users are. A firm's AI serves its lawyers. A corporate legal department's AI serves the whole company: sales, procurement, HR and finance all send work to legal, and the agents sit at that front door.
The 2026 ACC Chief Legal Officers Survey, drawn from 1,049 CLOs in 43 countries, found 36% of legal departments already in active deployment of generative AI. This post is about what those deployments should do first.
What AI agents should an in-house legal department run first?
An in-house legal department gets the most from AI agents at the points where the business hands it work. Five agents cover most of that volume, and each one drafts, routes or flags while a lawyer keeps the decision.
| Agent | Who sends the work | What the agent does | What a lawyer still decides |
|---|---|---|---|
| Contract intake and triage | Sales, procurement, partnerships | Classifies the agreement, compares it with the playbook, flags deviations, assigns a risk tier | Whether to accept a non-standard clause |
| NDA turnaround | Anyone starting a conversation with a third party | Generates the standard NDA or redlines the counterparty's paper to your fallback positions | Anything outside the approved fallback list |
| Matter intake | Every department, by email, chat or form | Asks the missing questions, opens the matter, routes it by type and urgency | Priority, staffing and whether to send it outside |
| Outside-counsel invoice review | Law firms, through e-billing | Checks each line against billing guidelines, rates and budget, drafts the adjustment note | What to reduce, and the conversation with the firm |
| Employee policy Q&A | All employees | Answers from approved policies with a citation, escalates anything fact-specific | Any question that needs legal advice |
The ibl.ai legal agent set covers the building blocks. The Contract Review agent handles redlining and risk flags, the Client Intake agent opens matters, the Billing & Time agent checks billing-guideline compliance, and the Knowledge Base agent holds your clause library.
Those agents ship written for a firm. In-house, you point them the other way: the "client" is the business unit, and the invoices are the ones your department receives rather than sends. The Legal Assistant sits in front and routes each request to the right specialist.
Why is AI for an in-house legal department different from AI for a law firm?
AI for an in-house legal department differs from law-firm AI in three ways: who uses it, what it is measured against, and how privilege works.
The users are the business. A law firm's AI is used by its lawyers and billed against matters. A corporate legal department's agents are used by employees who are not lawyers and never will be, so the interface is a question, not a research tool.
The measure is operational efficiency, not billable output. In the 2026 ACC survey, operational efficiency (53%) was the top strategic initiative for CLOs, and budget and resource constraints (35%) were the top barrier to success. 63% of CLOs expect headcount to stay stable as AI arrives.
Privilege for in-house counsel varies by jurisdiction. In the United States, Upjohn Co. v. United States (decided January 13, 1981) held that a company's communications with its counsel can be privileged when made by employees below the control group.
The EU rule is narrower. In European Commission competition investigations, the EU Court of Justice held in Akzo Nobel Chemicals v. Commission (2010) that internal communications with in-house lawyers are not covered by legal professional privilege.
A department operating in Europe should not assume its AI logs are privileged in such an investigation.
That makes the location of the logs a legal question, not an IT preference. When the agents run on infrastructure the company controls, the department decides retention, access and deletion for every prompt the business sends.
How should an in-house legal department use AI to triage contracts and NDAs?
An in-house legal department should use AI to sort contracts by risk before a lawyer opens them, and to turn standard NDAs without a lawyer at all when the counterparty accepts your paper or stays inside approved fallbacks.
The work splits into three lanes. Your own template, unchanged, goes straight to signature. Counterparty paper that stays inside the playbook gets an AI redline that a lawyer approves in minutes. Anything outside the playbook goes to a lawyer with the deviations already listed.
The playbook is the asset. Fallback positions, approved clause language and escalation rules are the department's institutional knowledge, which is why they belong in a store the department owns rather than in a vendor's tenant.
Harvey's Contract Intelligence announcement of May 21, 2026 describes the same shape for in-house teams: intake, triage, review and playbooks that update as contracts are signed. At announcement it was being built with design partners behind a waitlist.
How can an AI agent review outside-counsel invoices?
An AI agent reviews outside-counsel invoices by reading each time entry against the department's billing guidelines, the agreed rates and the matter budget, then drafting the adjustment note for a lawyer to send.
Departments already track this spend; they lack the hours to read it. In an ACC and Everlaw survey of 284 legal leaders published November 17, 2025, 83% said they track outside counsel spend.
The same survey found 43% expect generative AI to identify cost-saving opportunities on outside counsel services, and 42% expect it to improve analysis of law firm billing patterns. Only 12% track technology return on investment.
The 2026 ACC survey adds the pressure behind it: use of outside counsel jumped to 48% as regulatory complexity rose. An agent that reads every line of every invoice is how a department with a stable headcount keeps that spend visible.
The checks are mechanical, which is why an agent does them well. Block billing, rates above the engagement letter, tasks the guidelines exclude, and time on a matter past its budget are all rules. The judgment is what to do about them.
Can employees safely ask a legal policy agent their questions?
Employees can safely ask a legal policy agent their questions when it answers only from approved policies, cites the paragraph it relied on, and hands anything fact-specific to a lawyer instead of improvising.
The volume is real and repetitive: whether the gifts policy covers a client dinner, which approval a vendor needs, what the data-retention rule says about a shared drive. The ibl.ai HR Assistant agent handles the people-policy side of the same pattern.
The safety rules are architectural.
The agent retrieves from a policy set legal has approved. Its system prompt forbids legal advice on individual facts. Every conversation is logged where the department can audit it, and the Compliance agent page lists the security guardrails ibl.ai ships with every agent.
An escalation is a feature, not a failure. The agent's job is to answer the 80th question about the travel policy so a lawyer can take the first question about a whistleblower complaint.
How much do in-house legal AI agents cost per seat compared with usage pricing?
In-house legal AI agents cost far more per seat than by usage, because the agents serve the whole business and per-seat pricing bills every employee who might send a request, whether or not they ever do.
That is the structural problem. A seat price is sensible for a tool a lawyer opens every day. It is the wrong shape for a front door that 5,000 employees use a few times a year. Microsoft 365 Copilot lists at $30 per user per month on an annual plan.
The table below uses that list price and an illustrative month: 2,000 policy questions (4,000 input and 600 output tokens each) and 400 contracts triaged (30,000 input and 3,000 output tokens each), priced at Claude Sonnet 5.5's published API rates of $2 and $10 per million tokens.
| Pricing shape | Who is billed | Monthly cost |
|---|---|---|
| Per seat, legal team only | 25 lawyers × $30 (the business cannot use the agents) | $750 |
| Per seat, every requester | 5,000 employees × $30 | $150,000 |
| Usage, on ibl.ai | 20M input + 2.4M output tokens, model cost only | $64 |
The usage row is model tokens only; hosting and the platform licence sit on top, and your volumes will differ.
The gap is the point: one line scales with headcount, the other with requests. Microsoft also sells metered agent capacity through Copilot Studio, so the seat line is not its only option.
We work the same math for firms in AI cost math for law firms. In-house, the gap is wider, because the people sending work to legal outnumber the lawyers by a factor of 200 in this example.
Where do Harvey and Microsoft 365 Copilot fit for an in-house legal team?
Harvey and Microsoft 365 Copilot are the names that come up most often for in-house legal AI, and both are real options. The question for a legal department is who holds the code, the playbook and the logs once the agents are running.
Harvey Intake captures requests from email, Jira, Slack, ServiceNow and other channels, and routes each one to an approved agent, a solution such as Contract Intelligence, or a lawyer.
Microsoft 365 Copilot puts an assistant inside Word, Outlook and Teams, where much of the business already works.
Both are services you subscribe to. Neither hands the department the source code of the system its playbook and request logs live in.
That is the axis ibl.ai is built on: the department receives the code under a perpetual licence and runs it in its own cloud, on-premise or air-gapped.
It is also model-agnostic, so the department can route NDA triage to a small, cheap model and a novel indemnity question to a frontier one, and change either without a migration. Our Harvey and CoCounsel comparison covers the firm-side version of this choice.
How should an in-house legal department start with AI agents?
An in-house legal department should start with the one agent whose volume is highest and whose rules are already written down, usually NDA turnaround or policy Q&A, and measure turnaround time before and after.
Turnaround is the metric the business feels and the one ACC respondents already value: 81% of legal leaders in the ACC and Everlaw survey cited greater speed of legal support and matter resolution as a top benefit of generative AI. Measure it from the first week.
Then add the second lane. Contract triage reuses the playbook the NDA agent already reads, and invoice review reuses the billing guidelines the department already sends to its firms. Each new agent costs requests, not seats.
The same pattern runs for university counsel in AI agents for university legal and contracts, and for firm-side review in AI contract review for law firms.
ibl.ai is family-owned and operated from New York, NY, a long-term partner for legal departments that would rather own their AI than rent it. With ibl.ai you own all the code and the data, model-agnostic across any LLM, with no per-seat pricing.