Why working with an AI agent is becoming a baseline skill for lawyers

Nearly 70% of employers and clients already prefer a lawyer who knows how to work with AI over one who ignores it. Fewer than one in five have gone as far as writing that requirement into a job posting or the terms of an engagement, but the gap between “already expected” and “not yet asked for on paper” is closing fast.

And that gap is exactly where the biggest opportunity lies for lawyers over the next couple of years, along with a way to stay in demand as AI takes over more and more of the routine work. The expectation is there, yet only a handful genuinely know how to do it.

The lawyers who are first to learn not just to ask ChatGPT and request a tweak to the wording, but to work with a full-fledged assistant that reads, edits, checks against a playbook, and lives inside their working environment on its own, will gain a noticeable edge. In this piece we’ll look at where the profession is heading according to major studies, what an agent actually is and how it differs from an ordinary chat, and which specific legal tasks it already handles.

Where the market is heading

A recent 2026 study from the well-respected Thomson Reuters reports that over the past year AI use among lawyers has doubled. But the most interesting part is that only a handful have learned to extract real value and a noticeable speed-up from it.

The thing is, “using AI” today means two entirely different things, and the difference between them is enormous. There are two formats. The first is conversational: you open a chat, paste in a chunk of a contract, explain the context, get a response, and manually move it back into the document. The second is agentic: you simply set a task, and the system finds the right document in your folders or your inbox on its own, opens it, works through its own steps, accounts for your company’s full context, your playbooks and instructions, and hands back a finished result that you only need to review. The first format is a slightly smarter search engine at your fingertips; it barely changes your actual speed. The second changes it dramatically: you start doing complex tasks several times faster while working less. And this is where the real picture of the market comes into focus.

A recent major study conducted by top American universities together with OpenAI, “The Shift to Agentic AI: Evidence from Codex” (June 2026), measured this directly. For the average lawyer, true agents that carry work from start to finish account for less than 2% of working time. In other words, almost everyone who says “oh, I’ve been using AI for ages” is in fact stuck in the first format, in a back-and-forth with a chat, and getting almost no real benefit. Among the few who have already figured it out and moved to the second type, the share of autonomous work jumps straight to 17.6%, and the payoff is incomparable. Inside OpenAI itself, within a few months the legal function reached the point where the median lawyer was doing 13 times more than half a year earlier, and taking on more complex tasks. And these lawyers spend no more time than before. Which means you can take a more senior position, take on more clients, and multiply your income. So the mass of people saying “I’m on top of it, I use AI in my work” and the ones getting a real advantage are two different camps today, and the second one is still almost empty.

Clients and employers already sense this shift. Companies broadly expect this skill from their staff (various estimates put it around 70%), yet those stating the requirement plainly and out loud are barely 20% so far. And for anyone who figures it out now, that is precisely the main opportunity. It’s a way to secure your future and stay in demand on the market five years from now, when it will be the norm for everyone. For the moment, few have truly learned it, which means those who get there first will gain a serious competitive edge: they’ll be able to grow both their income and their career right now.

It comes down to something simple. The market already needs specialists who can work with AI assistants in full and tune them to their own needs. There are still few of them. Those who learn will gain an advantage and stay in demand, while those who don’t will start running into trouble. So the whole question is who moves into this camp first and who doesn’t. And the best news is that getting in is actually very easy: the whole thing is far simpler than it seems. We’ll look at exactly how next.

What an agent actually is and how it differs from ChatGPT

Strip away the jargon and an agent is a program that doesn’t answer a question but performs a task. The difference is like the one between a consultant and a doer. A consultant tells you what needs doing, and you still do it yourself. A doer goes and does it, and you review.

An ordinary chat waits for you to bring it all the context: to copy in the text of the contract, explain which side you’re on, spell out what to look for. It produces an answer, and you manually move it back into the document. An agent is given access to the right tools once, and from then on it opens the file from your storage itself, reads it, runs it through your rules, makes edits in track-changes mode, and hands back the result. All that’s left for you is to review and confirm.

The mechanics are simple. An agent works in a loop: it receives a task, decides which tool to call (read a document, search a database, compare versions), calls it, looks at the result, takes the next step, and so on until the task is done. You don’t need to get into this any more than you need to understand how a mail server works in order to send an email.

There’s one important point, and it’s simpler than it sounds. For the agent to work by the rules of your specific practice rather than at random, you don’t have to train it by hand and spell everything out in text. Setup is more of a joint effort: you hand it your protocols, your practice’s knowledge base, contract templates, examples of past edits, your standard positions, and it works through them itself and forms the set of rules it will follow going forward. You review that and correct it where needed. From then on those rules are reused in every task, and any request is handled with the exact same consistency whether you ran it yourself or a colleague covering for you did.

Which tasks it already handles

Look at what these agents actually do for lawyers today, and a few typical scenarios emerge. Below are the most common ones, just as examples, to make it clear what we’re talking about. In reality you can automate almost any repetitive process this way; these four are chosen for illustration.

One of the most common stories is the first-pass review of contracts and NDAs against your own playbook. It’s a familiar inbound flow: standard contracts and agreements where you need to check liability, indemnity, governing law, compare against your standard, and produce a redline. You set the agent up for this once, and again not by hand: you send it your protocols, templates, and examples of past edits, and it works out for itself which clauses are mandatory for you, which fallback wordings you use, what counts as a red flag, and turns that into rules. From then on you simply send it a new contract, and it returns a marked-up document with edits in track-changes mode and a short summary of deviations from your standard for the business. You decide whether to accept or reject each edit yourself, exactly as in a manual review, so the result stays under your control. One practitioner built such a tool on an open set of 510 real contracts (the CUAD dataset, 41 categories of legal risk) and added a handy touch: you specify which side you’re on, and the review adjusts to your interests. He also candidly notes that this is a first-pass tool, not a replacement for review on significant deals. And that’s the right framing: Gartner estimates that AI in contract management can roughly halve review time, but a human still signs.

The second common scenario is finding the latest version in the archive and comparing it. A familiar situation: there are several versions of a contract, and you need to figure out what changed from the last agreed one. Since the agent is connected to your Google Drive or document management system and remembers your matters, the request is simple: pull the latest version for this counterparty, compare it with the previous one, show what changed in the liability clauses. It finds the file itself, checks it, and hands back a list of discrepancies with links to the source. No manually digging through folders.

The third is translating regulation and a heavy clause into plain language. A significant part of a lawyer’s work is explaining to the business whether something can be done or not, drawing on dense statutory text. The agent prepares a draft explanation in plain language, but with strict discipline: use only the sources you provided, flag where there’s uncertainty, and don’t invent citations. This is a direct answer to the main well-known problem, which I’ll come to below.

And the fourth scenario is perhaps the most telling in terms of impact: essentially building yourself a junior assistant tailored to your practice, even if you work solo. The idea is that the agent connects to your databases and materials and takes on the preparatory part of recurring case types. A real example is a solo immigration lawyer: he has different visa categories set up as separate scenarios, and for each one the agent pulls the needed documents itself, assembles the initial set, and updates the data, while the lawyer steps in at the review-and-decision stage. His own observation captures the essence of the shift well: he used to move data between systems by hand and kept making mistakes doing it, and once the agent got direct access to the databases, that hassle simply vanished. So the assistant handles not a single task but an entire recurring chunk of work that usually eats up an assistant’s or the lawyer’s own time.

And here it’s worth saying what makes all these scenarios truly valuable over the long run. A scenario, once set up, is essentially a written protocol of how you work. A senior counsel at one company put it this way: a saved scenario means anyone on the team runs exactly the same review she would, at the same quality, even while she’s on vacation. Your standard stops living in one person’s head and turns into something that can be reused across the whole team. Figure out once how you work, and from then on it works for you permanently, without having to explain everything from scratch each time.

What the limitations are

A couple of honest caveats; without them the picture would be incomplete. First — AI sometimes makes things up. It may cite a nonexistent case or quote, and cases of fake precedents in court filings have already made the news. That’s why the agent should be used as a first-pass doer, not as the final word. It takes on the volume and the routine, but the key decisions and final wording are checked by the lawyer, and the lawyer is the one who signs. With the right setup, the agent is explicitly given the rule to rely only on the materials you provided, to flag disputable points, and to make nothing up.

The second is confidentiality. Client documents shouldn’t leak into someone else’s cloud, so it’s wiser to keep files locally with you. That way you get the agent’s speed and still control where the data sits yourself.

Keep these two rules in mind — a human reviews and the data stays with you — and the agent stops being a risk and simply becomes a faster way to do the same work with the same level of accountability.

How to start using this

You can build such an assistant yourself. But honestly, it’s not just about spending an evening on the installation. You also have to figure out how it all works, what to connect and where, how to describe scenarios, try it, get something wrong somewhere, and redo it. That takes time, and not every practicing lawyer has the desire or the time to dive into this and redo it many times over.

So there’s an easier path: take a ready-made, pre-configured solution. There are solutions like kvelo.dev that are already set up as agents and assemble a personal workspace for you, where everything you need — including typical legal scenarios — is already tuned and configured automatically, quickly and without a long manual grind. After that you simply work. Complex tasks you can do at your computer, and simple questions you can settle straight from a messenger: you write in Telegram or WhatsApp, and the agent goes off, pulls the needed documents, compares versions, checks the inbox, gathers what you asked for on the matter, and sends back an answer. In effect, you talk to it like a colleague.

By and large, the market has already decided that AI in law is the new norm. The difference between those who stay in a back-and-forth with a chat and those who learn to delegate to agents will soon be visible in the invoices, in the speed, and in who clients want to work with. That’s why it’s worth starting to get to grips with it now. There’s still time, and the chance to be among the first is still wide open. And to figure it out as simply as possible and build it into your work right away, you can start using and experimenting with a ready-made solution like Kvelo.