AI & the future of legal

A legal AI assistant combines a language model with access to legislation, case law and legal literature, and sometimes with the firm's own case files and knowledge base. The lawyer asks a question in plain language and gets an answer with references to the articles and judgments it relies on. That produces a first overview faster, often including rulings that are hard to find with separate keywords. The next step is an assistant that monitors the case file and, unprompted, points out that a new judgment affects the client's position.

AI places an agreement next to the organisation's own standard or agreed negotiation policy, and marks where the text departs from it or where something is missing, such as liability, notice periods or governing law. What follows is more interesting. AI monitors a portfolio of thousands of agreements continuously: deadlines about to expire, indexation, and contracts that need revising after a change in the law. In a negotiation, it uses earlier deals to predict which clauses the other party is likely to accept and where there is room to move. The lawyer still decides which risk the organisation is willing to carry. Further ahead, AI agents negotiate a contract themselves within a mandate, as in AI agents close the deal.

AI searches the thousands of documents in a data room and gathers the same details from every contract, such as term, penalty clauses and what happens on a change of control, into a single overview. Deviations stand out sooner, and the team reviews every document instead of a sample. Firms now practise this on complete transactions, with dozens of lawyers testing in a single day how much of the preparatory work AI can handle. The lawyer judges what a finding means for the price and the warranties. The next step is due diligence that takes days rather than weeks, and that does not stop at closing but carries on after the acquisition as monitoring of contracts, permits and disputes.

Large disputes and investigations soon involve millions of emails, chats and documents that have to be collected, searched and reviewed. AI learns from documents that lawyers mark as relevant or irrelevant, and uses that to select other documents for review. That can save a great deal of reading. Judges have allowed this form of document selection in specific cases, with attention to how reliable the process is. Newer systems also build timelines, so a team sees sooner who knew what, and when. The next step is evidence that carries signed provenance data from the moment it is created, so a court can check every recorded edit made since it was signed.

AI moves the first answer to recurring legal questions to the place where they come up: the colleague preparing an agreement, processing data or checking terms and conditions. An AI assistant connected to the company's own policies and standard contracts answers the simple questions directly and passes the rest, with a summary, to the right lawyer. In-house counsel gain time for strategic advice, negotiations and the difficult judgement calls. Before long, the legal team will use AI agents to follow the processes in procurement, HR and sales, and see a risk taking shape before anyone asks about it.

AI can link a changed threshold or a new decision by a regulator to the contracts and processes it affects, even when departments use different terms for them. The system follows new rules across several countries and sectors and shows which departments will have to deal with them. A next step is continuous testing: the system lays policies and contracts beside the current rules and flags where something no longer fits. Further ahead, rules are published directly in a form machines can read, and systems process a change in the law on the day it appears.

A legal AI agent is AI that completes an assignment in several steps and decides for itself which systems and sources to consult along the way. An AI agent can, for example, gather the documents in a case, draw up a chronology, look for relevant case law and write a draft court document. In a notary's practice, an AI agent can retrieve data from registers and prepare a deed for review. Firms use and test such AI agents in clearly defined processes, with fixed checkpoints. Where a lawyer once took every step personally, the lawyer now directs a sequence of steps that the AI agent prepares.

Legal language is hard to follow for many people and small business owners: a notice of termination, a tenancy agreement, a payment demand with a deadline. A legal AI assistant is a language model set up to work with laws, regulations and legal sources. It explains in plain language what the document says and which rules apply to a dismissal, a tenancy dispute or an unpaid invoice. Users can ask follow-up questions about an unfamiliar term or a difficult passage. Legal service providers already offer such assistants as a first step, before a lawyer takes a look. The explanation helps someone understand a letter or contract, but it can contain mistakes and does not replace a lawyer's judgement.

Filing documents, consulting case files and corresponding with the court increasingly happen digitally, and depending on the procedure, hearings can take place remotely. On that foundation, AI can support court staff and judges: organising documents, building a timeline and turning judgments into language people can follow. Some courts already do this. AI that supports judges on the substance, in establishing facts, interpreting the law and applying it to a specific case, counts as high risk under the European AI Act; purely administrative help does not automatically fall into that category. The next step is a court where AI transcribes and translates live during the hearing, and where a party can follow its case file in plain language at any moment.

In the Netherlands, a private company (bv) can already be incorporated entirely remotely, over a secure video connection and with a digital identity, and notaries are pressing for changes in the law so that other deeds can be executed remotely too. The result is a practice in which signing at the table and signing remotely exist side by side. Behind the scenes, AI applications that help the notary prepare are being tested. They pull data from registers, compare it with the draft deed, check powers of attorney and point out possible contradictions. The notary then spends less time checking data and more time talking with the client.

With AI, product safety, fundamental rights and professional liability come together in files where the behaviour of AI itself is assessed in legal terms. The European AI Act classifies AI systems by risk, with prohibited uses and strict requirements for AI in, for example, recruitment, lending or healthcare. Questions about privacy, copyright, product liability and contracts with suppliers come on top. Organisations want to know which systems they use, which risk category they fall into and what they have to document. A practice of its own is emerging, in which lawyers work alongside engineers and ethicists. For firms it is a growth market; for in-house counsel, a new part of the job. The rules will be worked out and extended over the coming years.

Generative AI raises two copyright questions. The first concerns the input: whether AI companies may use books, articles, music and images to train their models, and whether they should pay for them. Courts in several countries are now giving the first answers, and publishers and creators are signing licensing agreements with AI companies. The second concerns the output: whether a text or image created by AI is protected, and if so, who owns it. In many legal systems, copyright requires a human creation. Further ahead lies a global market for licences on training data, in which creators are paid per use and provenance data travels with every image and every text as standard.

Research, first analyses, drafts, comparison with earlier work and predicting an outcome: in an AI-first firm, that work increasingly sits with AI, so lawyers no longer have to do it themselves. Professionals spend more time talking with clients, weighing risks, setting strategy and negotiating. The organisation sets itself apart through the quality of its judgement. In ten years, AI agents may do much of the operational work, and the number of entry-level positions may fall. The question is then no longer where AI helps, but where a person makes the difference.

A legal engineer translates legal knowledge into systems: templates, decision trees and workflows through which AI and lawyers work together. A knowledge lawyer looks after the knowledge base that the AI draws on. Those roles already exist. In five to ten years, others will carry more weight. The lawyer who writes and monitors mandates for AI agents, setting out what an AI agent may promise, sign or pay on the client's behalf. The product manager of a legal subscription, who sells knowledge as a service. The specialist in neurolaw and AI law. And the lawyer who uses models to predict the outcome of cases for litigation funders. The entry-level work of junior lawyers changes with it: fewer days reading documents, earlier involvement in the substance.

On more and more cases, lawyers and AI work together, each with a part of their own. AI gathers the facts, searches case law, compares contracts and delivers the draft of a memo or court document. The lawyer asks the right questions, checks what AI produces, makes the call and speaks with the client and the other side. AI takes the repetitive and the deeply analytical work: reading thousands of pages, finding patterns in case law, working through variants. The lawyer does what people are good at: judging, choosing, persuading and working with people.

With a future academy, an organisation trains its own people for the work that is coming, without waiting for universities and training providers to adapt their programmes. People who already work there learn how AI and new technology change their work and how to assess AI output critically. New colleagues learn the profession as it will be practised, not as it used to be. Leaders learn to steer an organisation in the middle of change. The organisation adjusts the lessons as soon as tasks and applications change.

An organisation rarely makes AI useful on its own. It develops new applications together with tech companies and start-ups, examines with researchers how reliable they are, and discusses with peers what good work with AI looks like. Governments publish legislation and judgments as open data and take part in trials. That ecosystem is international: partners in different countries work with the same systems and share their experience. What is tested on one continent can be tried elsewhere through partners, so innovation reaches practice sooner.

Part of the advice a firm gives again and again can be captured: a check of whether a contract meets the client's house rules, an overview of rules per country, a decision tree for a common question. Firms package that knowledge in tools clients use themselves, often on subscription. Such a product does not come out of a practice group alone. Lawyers work on it with designers, developers and product managers, and the client's in-house counsel helps test it. A product is never finished: it is updated when the law changes, and it has its own budget and its own owner. Alongside casework, product development becomes a distinct skill within the firm.