I’ve been practising long enough to distrust anything sold as a revolution, and I expected to be unimpressed by this one. I wasn’t. You can watch it change how the work actually gets done, so it’s worth being clear about what’s real and what it means if you run a business and pay the legal bills.
Harvey is the name most people have heard. Early this year it raised money at an $11 billion valuation, and it says more than 100,000 lawyers at around 1,300 organizations now use it, including most of the hundred largest firms in the United States. The company doesn’t call itself a tool that helps lawyers. It calls itself the operating system that legal work runs on. It has plenty of company: Legora, CoCounsel (now owned by Thomson Reuters), Spellbook, Robin AI and the big legal-research platforms are all after the same market, and a lot of lawyers quietly run general tools like ChatGPT and Claude on the side.
There is an important difference between those two things. A general chatbot is a clever assistant that owes you no duty and keeps no wall around your documents. The legal platforms are built to sit inside a firm’s systems, trained on legal material, and set up to keep client information confidential. That is not a technicality. It is the reason pasting a contract into a public tool is something a lawyer should never do and a client should think hard about.
For a long time, legal AI mostly answered questions. Now it runs tasks. Harvey says it hosts tens of thousands of custom “agents” that handle M&A work, due diligence, contract drafting and document review from start to finish. Think about the week a first-year associate used to spend reading a data room and working through two hundred contracts to flag every change-of-control clause. An agent does that in an afternoon. Whether you find that exciting or alarming probably depends on whether you were billing for the week or paying for it.
No, the billable hour is not dead, and it will not die this year. But it is under real pressure at the low end. No client is going to keep paying six minutes of an associate’s time for something a machine finished in twenty seconds, and none should. I expect the market to split. Routine, high-volume work moves to fixed and flat fees, because it now costs a fraction of what it used to. The harder, higher-stakes work stays on the clock, or gets priced to its value, because that is where a client is paying for the thing a machine cannot give them: judgment, and a lower chance of getting it wrong. I already quote a price up front for defined work, and AI mostly makes that easier to do with a straight face.
The upside is simple. Routine legal work should get faster and cheaper, and it already has. The cautions matter just as much, and a lawyer who won’t name them is not being straight with you.
Confidentiality comes first. Paste a document into a public AI tool and you may have broken a confidentiality obligation and chipped away at the privilege that protects legal advice, before anything has gone wrong at all. Second, these tools are wrong often enough to matter, and they are wrong with great confidence; lawyers have already been sanctioned for filing briefs built on cases a chatbot invented. Someone who can be held responsible has to check the work. Third, “I ran it through AI” is not legal advice. The model does not know how much risk you can live with, who you are dealing with, or what the other side is really after. I go into where AI saves money and where it quietly costs you more in a separate piece on small-business legal costs.
The work is changing, not vanishing. The lawyer worth paying increasingly aims the tool and checks its output instead of typing the first draft by hand, and the ones who refuse to go near it will lose ground to the ones who learn to use it well. Our duties, I would add, have not moved an inch. Competence, confidentiality and the obligation to supervise are what they have always been, and the Law Society of Ontario has said as much to every licensee. New tools, same rules.
One change has made me more optimistic rather than less. AI narrows the gap between big firms and small ones. A boutique can now take on work that used to need a room full of junior lawyers, which means a client can get a senior lawyer across the whole file instead of paying for leverage and overhead. Scale used to be the large firms’ advantage. A good part of it is now within reach of anyone with sound judgment and the right tools.
It cannot be responsible. You cannot sue a model when the advice turns out to be wrong. You can sue a lawyer, and that accountability is not a flaw in the old way of doing things; it is a large part of what you are paying for. It cannot read a negotiation, judge how hard to push, or tell you that the deal you asked for is not the one you should want. It does not hold your trust. And for all the progress, it still invents things, so checking its work stays the job rather than a formality. The lawyers who do well over the next few years will not be the ones with the fanciest software. They will be the ones who know when to rely on it and when not to.
This is the biggest shift in how legal work gets done that I have seen, and I think it is mostly a good one. Routine work gets cheaper, and it frees up experienced lawyers for the parts that are hard for the right reasons. But it is a tool, not a lawyer, and the space between those two is where clients either get protected or get hurt. I use these tools in my own practice, and I offer a lawyer’s review of AI-drafted documents for the same reason: they are now good enough to be dangerous in the wrong hands. If you are using AI in your business, or trying to work out whether your lawyer should be, it is worth a conversation. Twenty minutes is usually enough to tell whether you are getting the savings without quietly taking on the risk.
Twenty minutes, no charge — a straight read on where you stand.
Book a 20-Minute Call