Legal: AI tools got cheap, and an AI-first firm arrived

Legal AI for small firms changed shape in May 2026: the capability stopped being a premium product and became a cheap add-on to software that lawyers already run. On 12 May, Anthropic launched Claude for Legal, wiring its AI into Westlaw, DocuSign, iManage and Microsoft 365 for any paying Claude customer at about $20 a seat each month. A day later, Carta turned a law firm itself into a product, launching an AI-first firm that sells legal work to investment funds at a fraction of the traditional cost.

Where legal AI actually stands

Here is the landscape a firm is choosing from. It has three layers.

First, the research tools the profession already trusts. Thomson Reuters CoCounsel, which folds in the Westlaw research database, passed one million users on 24 February 2026 and won the law librarians' 2026 New Product Award (LawSites); LexisNexis offers similar through Lexis+ AI and its Protégé assistant. These are the closest thing legal AI has to standard equipment.

Second, the premium platforms built for large firms. Harvey, used across most of the hundred largest US firms by revenue, raised $200 million in March 2026 at an $11 billion valuation (CNBC, Bloomberg); its European rival Legora reached a $5.6 billion valuation in April with backing from Nvidia (TechCrunch).

Third, and most relevant to daily work, AI that lives inside Microsoft Word, where contracts are actually written. Spellbook has led here for contract lawyers. On 30 April 2026, Microsoft shipped its own Legal Agent inside Word: it checks a contract clause by clause against a firm's playbook, its list of acceptable terms, then produces a redline, the marked-up version showing tracked changes, with citations to the source (Artificial Lawyer; Legal IT Insider). Microsoft built it on Anthropic's Claude, not the OpenAI models that run most of Copilot.

What changed this week

On 12 May 2026, Anthropic launched Claude for Legal: twelve practice-area plug-ins and more than twenty connectors that let its AI read and act inside those same programs, from Westlaw to DocuSign, iManage and Microsoft 365 (LawSites; TechCrunch). For a small firm, the price is the point: it is available to any paying Claude customer for roughly $20 per seat a month. Harvey does not publish prices, but third-party trackers put its enterprise deals near $300,000 a year; for a fifty-lawyer firm they estimate roughly $18,000 on the Claude route against $150,000 or more for Harvey. Treat the figures as estimates, but the gap is close to ten times.

A day later, on 13 May, Carta did something different. Instead of selling a tool, it bought a law firm. It acquired Avantia, an AI-native firm already serving more than 200 investment funds, and relaunched it as Carta Law: an "AI-first" firm that sells legal and compliance work to private-equity and venture funds at, in its words, a fraction of the traditional cost (Carta; BusinessWire; Law360).

What it means for your firm

The first shift is that capability is no longer a luxury purchase. The reason a small firm could once skip legal AI was cost: a serious contract started in the hundreds of thousands. That barrier is gone, and the same review-and-redline work now sits inside Word for the price of a streaming subscription.

The second is what clients will expect. "AI discounts" are already appearing in legal requests for proposals, the documents corporate clients use to put work out to bid, and in 2026 panel reviews, where a company decides which firms to keep (Fennemore; legal.io). If the firm's costs fell, the client wants a share of the saving.

Be honest about the pace. A Bloomberg Law survey found 58% of firms said AI had not changed their billing at all, and only 19% saw billable hours fall. The billable hour, charging for time rather than the result, is not collapsing this quarter. The pressure is coming, but for most firms it has not arrived.

Carta Law is the sharper signal. If your firm does fund formation or compliance work for investment managers, a well-funded competitor just turned that exact service into a packaged product. For most firms outside that niche it is a preview, not an immediate threat. It shows where this is heading: not a better tool for lawyers, but a cheaper substitute for the firm itself.

Then there is staffing. First-pass review, contract analysis and research memos are the work that used to justify a large class of junior lawyers. In the 2026 Wolters Kluwer Future Ready Lawyer survey, 51% expected that work to shift to alternative legal service providers, meaning non-traditional, often technology-driven outfits. The junior's job is moving to checking and directing the machine's first pass, and courts will hold the human lawyer, not the software, responsible for a fabricated citation.

What to do about it

If you pay for a premium legal AI seat your team barely uses, re-price it at renewal. The cheaper option is not automatically better, since premium products bundle support and security that matters, but make the comparison deliberately rather than renewing out of habit.

Pick one or two high-volume, low-judgment tasks, such as first-pass contract review or a research memo, and run them through the cheaper route with a named person responsible for checking the output. Measure how often it is wrong before you let it near client work. That error rate, not the demo, tells you what the tool is worth.

Decide your billing answer before a client asks. You need not abandon time-based billing this year, but have a clear response when the "AI discount" question appears in your next proposal or panel review, because it will.

And watch Carta Law if your clients are funds. That is the model built to take your work, and the next year will show whether the AI-first firm is a real competitor or an expensive experiment.

The pattern underneath

Era Haus argued in Defensibility in the AI era that in a field like law the durable advantage was never the AI model itself. It is the workflow your tools are embedded in, your standing in a regulated profession, and the years of reliability behind a system that works. This month makes the point twice: Microsoft put a legal agent in Word and chose Claude to run it, and Anthropic now rents the same engine for $20 a seat. The model is becoming a commodity layer, so value moves to the software it plugs into and the firm willing to put its name behind the output, the same argument we made in The model wasn't the moat.