What Is an AI Lawyer? How to Divide Work Between Attorneys and AI in Corporate Legal Practice in the Age of Generative AI

Hello, I'm Noriaki Asato, Representative Attorney at LegalAgent.

Some people look for ways to handle legal consultations and contract review by searching for terms such as "AI lawyer," "lawyer AI" and "legal AI." Behind this are expectations that feeding in a contract will immediately produce proposed revisions, that AI might be enough for legal consultations, and that it might even solve the staff shortage in the legal department.

These expectations are natural in themselves. In fact, generative AI is highly capable at routine, text-heavy work such as contract review, clause comparison and organizing meeting minutes. In corporate legal practice as well, I feel there is plenty of room to rethink which preparatory work can be entrusted to AI.

On the other hand, if only the impression conveyed by the term "AI lawyer" takes on a life of its own, a risky understanding emerges. In Japanese corporate legal practice, AI does not hold an attorney's license. Nor does AI autonomously grasp and make judgments on individual circumstances such as the company's risk tolerance, its relationships with business partners and the history of past negotiations. Separate the preparatory work that should be entrusted to AI from the issues that attorneys or in-house legal staff should decide. That is the starting point for dividing the work.

The Expectations Behind the Search and the Reality

In practice, the inquiries from people interested in AI lawyers fall into several groups. One is speeding up contract review. This is the need to have AI read NDAs, service agreements, SaaS terms of use, sales agency agreements and the like, and to quickly extract risky clauses and prepare proposed revisions.

Others want an initial sorting of legal consultations. This is the use of having AI list the relevant laws and regulations, the facts to be confirmed and the issues to be decided internally when a question comes in from a business division.

In companies short of legal staff, there is also the hope that AI could take over part of contract review and consultation. Especially in organizations with a one-person legal function or where legal work is done alongside other roles, there are situations where simply having AI helps.

However, even though these are all grouped under the same term "AI lawyer," their substance is different. There is AI that creates a first draft of a contract, AI that summarizes clauses, AI that searches internal knowledge, and so on, each with a separate role.

The dividing line to look at in corporate legal practice is less "whether AI can produce an answer" than "whether AI output can be shaped into a form usable for the company's decision-making." This perspective separates standalone legal AI tools from legal services of the AI Native Law Firm type in which attorneys are involved.

Reducing Routine Work and Where Judgment Lies

There are many tasks at which generative AI excels. Summarizing the structure of an entire contract, extracting candidate risks clause by clause, organizing the changes in the counterparty's proposed revisions, preparing memos for internal explanation, creating issue checklists: the more text-heavy and relatively fixed in format the preparatory work, the more readily AI delivers value.

There are also tasks that are hard to complete with AI alone. These are judgments such as how far to request revisions in light of the balance of power with the counterparty, whether to accept a risk and proceed with the transaction, how to assess consistency with past disputes and internal policy, and at what level of detail to escalate to management. Because the material for these judgments lies outside the contract, that is, within the company's actual circumstances, a conclusion cannot be reached by feeding in the contract alone. The materials available for reference are supplemented, and humans confirm the premises and the soundness of the judgment.

In contract review practice, too, it can be hard to use AI if it merely points out that "this clause should be revised." What the business division wants to know goes beyond theoretical problems: "Can we accept this in this transaction?", "How do we communicate it to the counterparty?" and "Is there an alternative if they will not concede?"

AI has the ability to pick up a wide range of issues. Human judgment is needed to prioritize the issues picked up and to narrow them down in line with the company's decision-making. In my view, the core of using AI in corporate legal practice is not to aim for AI replacing attorneys, but to create an environment in which, by bringing in AI, attorneys and the legal department can concentrate on judgment. How our firm actually incorporates AI agents into legal work is described concretely in Why We Use Codex for Legal Work.

Background Information to Supplement Contract Review

When using AI for contract review, the first thing to keep in mind is not to give AI the contract alone.

Looking at the clauses alone, formally risky wording can be identified to some extent. However, whether a clause is really a problem varies depending on individual circumstances, such as the content of the transaction, the amount and the importance of the counterparty.

Take a clause with no cap on damages as an example: it generally looks onerous. Even so, there are cases where the transaction amount is small, the substantive damage anticipated is limited, and the clause simply has to be accepted given the relationship with the counterparty. Conversely, even in a contract with a liability cap, exceptions may be needed for personal information leaks and intellectual property infringement. A small contract amount does not necessarily mean small damages. The assessment also takes into account whether your company is the party bearing liability or the party seeking indemnification from the counterparty.

When having AI conduct a review, I think it is desirable to include at least the following information in the prompt, within the scope of internal authorization and confidentiality obligations.

• Your company's position
• The contract type
• The transaction amount and contract term
• The main deliverables or services
• Whether personal information, confidential information or intellectual property is involved
• The relationship with the counterparty
• The risks that are acceptable and the risks to be avoided
• Whether proposed revisions can be put forward firmly to the counterparty or should be presented gently

If you ask AI without this information, you will get an answer that may be correct as a general proposition but is hard to use in that particular matter. Reflecting AI output directly in the contract should also be avoided; humans need to finalize the proposed revisions and how comments are presented, in line with the business background. How to put review standards in place as an organization is explained in In the AI Era, Contract Review Standards Are Best Organized as a Legal Playbook.

Fact-Finding and Policy Decisions in Legal Consultations

When bringing AI into day-to-day legal consultations, too, an approach that separates fact-finding from policy decisions is useful. AI is good at listing, from the text of a consultation, the relevant laws, contract clauses, internal rules and facts to be confirmed. When a business division asks, "Is this campaign OK?", "Can we pass personal data to this contractor?" or "Can we use this generative AI tool for work?", AI is useful as a tool for the initial identification of issues.

The danger lies in rushing to a conclusion while the underlying facts remain unclear. Even if AI produces a plausible answer, the conclusion changes if the underlying facts are different.

In practice, therefore, rather than making AI rush to a conclusion, it is effective to have it list "the additional facts that should be confirmed." For a consultation involving personal information, AI is asked to identify the points to be confirmed, such as whose information it is, for what purpose it will be used, and whether it is a provision to a third party or an entrustment. Once the facts are in place, the legal department or attorneys consider the response policy in light of the company's risk tolerance, internal rules and relationships with business partners, and support the company's decision-making. How to accumulate past consultations and decisions in a form that AI can refer to is described in Knowledge Management for a Law Firm with Google Drive and Codex.

Points to Consider in Tool Selection and Operational Design

When considering AI lawyer or legal AI services, choosing on the basis of the number of features alone can result in not being able to make full use of the service after introduction.

The first point to check is whether contracts and consultations can be processed in line with the actual workflow. Can it return work in Word tracked changes? Can it separate comments for the counterparty from internal risk explanations? Can consultations be made on Slack or Teams? Can past decision criteria be accumulated? Usability after introduction changes greatly depending on these points.

Next, look at who checks the AI output. Avoid a practice of using AI-drafted revisions as they are: will an attorney review them before they are returned, or will in-house legal staff do the final check? The quality control mechanism changes depending on how the reviewer is designated.

Where to escalate issues that AI cannot handle is also something that should be confirmed. Contract termination, labor disputes, M&A and the like are often situations where the matter should not proceed on AI's initial sorting alone.

When people see the term "AI lawyer," their attention tends to turn to the performance of the AI itself. In corporate legal practice, however, what is at stake is the entire legal operation into which AI is incorporated. Who inputs, who checks, in what format the work is returned, and which decisions are kept on record within the company. Only once these have been decided does AI function as a legal tool. The criteria for choosing a law firm itself are laid out in detail in How Should Companies Choose a Law Firm Strong in Generative AI?.

A structure in which attorneys provide ongoing support for contract review and legal consultations while using AI is also introduced in Legal Outsourcing.

Practical Support from an AI Native Law Firm

As an AI Native Law Firm specializing in corporate legal work, LegalAgent does not leave legal judgment to AI alone; our attorneys make full use of the AI Agent to support contract review, legal consultations, internal explanations and more.

In our legal outsourcing, we receive consultations on Slack or Teams and place importance on returning contract review and day-to-day legal consultations as practical output grounded in an understanding of the business. Even if processing is faster, if the reasons for decisions are not recorded, it becomes hard to carry knowledge over to the next matter. The substance of our support is to keep a record of the shape of each decision, including which risks the company accepts and which issues it negotiates.

When considering introducing an AI lawyer or legal AI, please start by separating the work you want AI to complete on its own from the work that attorneys or in-house legal staff should decide. Once that line is drawn, it becomes easier to see where in contract review, legal consultations and knowledge management AI should be used.

• Legal Outsourcing
• LegalAgent for Word
• AI Legal Lab: A Practical Media Outlet on Legal Work and Generative AI

Frequently asked questions

Can contract review be completed with AI alone?

AI can handle extracting clauses, identifying candidate risks and drafting revisions. However, whether to accept a clause depends on the transaction amount, the relationship with the counterparty and the company's risk tolerance, so it is considered appropriate for attorneys or in-house legal staff to make the final decision.

Do so-called AI lawyer services raise issues under the Attorney Act?

As a rule, only attorneys or legal professional corporations may handle legal business concerning legal cases for the purpose of obtaining compensation (Article 72 of the Attorney Act). From this perspective, there is always debate about formats in which an AI tool alone provides legal judgments. If attorneys use AI to provide legal services, this problem is unlikely to arise.

When having AI read a contract, what should we tell it along with the contract?

By also conveying your company's position, the contract type, the transaction amount and contract term, the deliverables, whether personal information, confidential information or intellectual property is involved, and which risks are acceptable and which you want to avoid, you get output closer to one tailored to the matter rather than general observations.

Will attorneys be replaced by AI?

At present, the division of labor at the center of practice is that AI handles organizing documents, drafting and identifying issues, while judgment, negotiation and responsibility remain with attorneys. I think a change is under way in which the more deeply attorneys use AI, the more time they have for judgment.