Why Legal AI Is Moving to Consumption-Based Pricing

Hi legal tech geeks. One of the most important changes happening in legal tech this year is how legal AI is being priced. In case you haven’t heard, AI prices are generally going up, and that means the cost of your legal tech tools may soon increase significantly, if it hasn’t already.

Word on the street is that the idea we were sold—that cheap AI tools could replace expensive human labor—was never entirely true, and that investors were absorbing excess costs all along. Well, it seems they now want to see profits. The hope appears to be that companies have seen enough of the benefits of AI that they will now be willing to purchase AI tools at something closer to their “actual cost.”

For most of the past few years, enterprise software has followed a familiar model: companies buy licenses, or “seats,” for individual users. A law firm might buy 100 seats for 100 lawyers and pay roughly the same amount whether those lawyers use the software every day or barely touch it. The pricing has been relatively predictable, typically around $200 per seat for many leading legal AI products.

But legal tech companies are beginning to break from that model.

Legora, one of the fastest-growing legal tech companies, has now introduced consumption-based pricing for Agent Pro, its most advanced agentic product. Instead of charging simply for access to a seat, Legora wants the economics of the product to reflect how much work the AI actually performs.

Legora describes the shift in ambitious terms:

“From hours billed and seats licensed, to outcomes delivered.”

Legora

That sentence may tell us quite a lot about where legal tech is heading. I may be wrong, but to me, the emphasis on “the work AI actually performs” sounds fundamentally like a cost issue.

It also signals that high-quality work performed by AI may eventually carry a significant price tag, while lower-cost AI may remain available for less sophisticated work.

Aren’t we then, in some respects, back where we started? Historically, high legal costs were justified by high labor costs. We may now be moving toward high legal costs justified, at least in part, by high AI costs.

And AI was supposed to help level the playing field between law firms. If access to the most capable AI becomes increasingly expensive, however, the gap between firms may instead widen.

Why the Sudden Change?

There is an important distinction to make here, particularly when it comes to Legora. Legora has not eliminated seat-based pricing across its entire platform. It has introduced consumption-based pricing for one of its products. What the rest of the industry does remains to be seen.

In June 2026, the company announced consumption-based pricing alongside the launch of Agent Pro, which uses frontier AI models to plan, execute, review, and deliver complex legal work. Legora says customers pay according to the work Agent Pro performs and can attribute that consumption to the project or matter responsible for it.

Meanwhile, Legora says its standard Legora Agent remains available to existing customers at no additional cost and without changes to their existing contract terms.

But the significance goes beyond one product.

CEO Max Junestrand told Business Insider that the move away from the traditional software seat was essentially inevitable. According to the publication, Legora began moving new customers toward pay-as-you-go pricing in June.

The reason comes down to a fundamental difference between traditional software and AI.

A Seat Doesn’t Tell You How Expensive an AI User Is

Let’s use Microsoft Word as an example.

If Lawyer A opens Word twice this week and uses it for a total of two hours, while Lawyer B spends eight hours a day drafting documents, the marginal computing cost to Microsoft is not dramatically different.

Generative AI doesn’t work like that.

Every prompt requires computation. Longer documents require more processing. More sophisticated reasoning can require substantially more computation. And an autonomous agent carrying out a complex, multi-step workflow may make many model calls behind the scenes before returning a finished result.

Two lawyers occupying exactly one “seat” each can therefore generate radically different costs for the AI provider. Somebody has to pay for that, which helps explain why Legora’s CEO described the shift as inevitable.

Research into AI economics increasingly supports this point. One 2026 study of reasoning models found enormous variation in the computational resources consumed by different models and even between repeated executions of similar tasks. Another study of AI agents found that agentic tasks can consume dramatically more tokens than ordinary AI interactions, with consumption varying substantially between runs.

That matters enormously as legal tech moves from lawyers asking chatbots questions to agents performing entire workflows. This brings me to my next point: AI is increasingly becoming labor.

Legal AI Is Becoming Less Like Software and More Like Labor

This may be the bigger story, and this is where AI’s role in the legal industry becomes particularly interesting.

The first generation of generative legal AI largely helped lawyers perform individual tasks: summarize this document, draft this clause, research this issue, or explain this provision.

The next generation is increasingly being designed to execute workflows. Legal tech is edging closer to performing legal work, not merely assisting lawyers in performing it.

Legora itself says its customers have moved from using AI for discrete research and document-review tasks toward multi-step agentic workflows involving large document sets and structured outputs. The company says it surpassed $100 million in annual recurring revenue in April 2026 and was serving more than 1,000 customers.

That creates a financial problem for Legora and other legal tech companies, particularly those serving large law firms. Some clients are running workflows involving thousands of documents, and the cost to the vendor can be substantial. They are effectively commissioning large amounts of computational work.

It would make little economic sense to charge a client processing thousands of documents through complex AI workflows exactly the same amount as a five-person law firm making relatively light use of the platform.

The AI Company’s Cost Structure Is Changing Too

There is another side of the equation: the legal tech vendor itself. AI companies face costs that traditional SaaS companies did not face at the same scale. Each time users invoke frontier models, somebody has to pay for that inference. And legal AI usage is growing rapidly.

Harvey CEO Winston Weinberg recently said the company’s AI consumption increased from roughly one trillion tokens per month in January to an estimated 12–13 trillion in May 2026.

That is an extraordinary increase in computational consumption in only a few months. Imagine trying to price that growth using a flat fee attached to the number of employees who have login credentials. The number of seats might barely change while the amount of AI being consumed increases tenfold.

For legal tech companies, consumption pricing allows revenue to increase alongside the computational workload.

For legal tech vendors, therefore, this isn’t merely a new way to charge customers. It may also be a way to protect the economics of the business as customers move toward increasingly compute-intensive agents.

I’m still struck by the fact that we are moving from legal tech as a workload-assistance tool toward technology that may increasingly perform work that would otherwise have been carried out by people. Wait until you hear about AI-native law firms; more on those in the next articles.

It Could Also Change AI Adoption Inside Law Firms

There is an interesting flip side.

Seat pricing can discourage experimentation. Suppose a firm has 1,000 lawyers but isn’t sure whether all of them will use a new AI platform. Buying 1,000 expensive licenses is difficult to justify. So the firm might buy 100 licenses instead.

Now access to AI is artificially scarce. Lawyers compete for licenses, pilots remain confined to innovation teams, and the technology struggles to become part of everyday practice.

Consumption pricing potentially changes that.

Legora explicitly points to this benefit, saying customers can bring more users onto the platform while maintaining control over total spending.

There is an upside for law firms: this pricing model means firms pay for what they actually use, rather than what they are contractually obligated to pay under a fixed seat commitment.

The per-seat model can have the unintended consequence of requiring law firms to commit to paying for software that they may ultimately not need or use.

Consumption-based pricing could therefore remove one of the quieter barriers to enterprise AI adoption: deciding who deserves a seat.

Consumption Pricing Fits Legal Matters Surprisingly Well

There is another reason this model could be particularly important in legal services. Legal work is already organized around matters. Law firms track time, expenses, and revenue against individual client matters. In-house legal departments similarly track spending and workstreams.

Legora’s consumption system allows usage to be attributed to individual projects or matters. Its dashboard can track consumption by organization, user, or project, while administrators can establish thresholds and spending controls.

That creates an intriguing possibility.

It moves AI measurement closer to the economics of legal work itself. And once firms can calculate AI costs at the matter level, another question inevitably follows:

This Is Where Things Get Really Interesting

Consumption-based pricing could eventually collide with one of the legal industry’s oldest institutions:

Suppose an AI agent completes in 20 minutes a document-review task that previously required an associate to spend eight hours. The law firm’s software bill may now increase because it consumed significant AI resources. But its billable hours may decrease.

That creates a fascinating economic tension. The technology vendor increasingly gets paid according to the work performed, while the law firm may still get paid according to the amount of human time consumed.

Those two models cannot coexist comfortably forever.

This doesn’t mean AI will suddenly kill the billable hour. Predictions of its death have been made for decades. And law firms’ revenues continue to reach record highs—so much so that private equity increasingly wants a piece of the legal-sector pie.

Was it the legal tech boom that caught their attention? Or the potential disruption of the billable hour?

AI does, however, make the contradiction between the billable hour and consumption-based pricing more obvious. If the cost of producing legal work becomes increasingly linked to computational consumption rather than human time, clients may increasingly ask why the price of legal services should remain tied exclusively to hours.

This is going to lead to some very awkward conversations with clients.

Consumption pricing in legal tech could therefore become one small piece of a much larger transition toward fixed fees, subscriptions, matter pricing, and other value-based arrangements.

This may also help explain why private equity is increasingly interested in the industry. If the AI sector is going to capture a larger share of the economics of legal work, investors will want a piece of that opportunity too.

There Is a Catch: Nobody Likes an Unpredictable Software Bill

Consumption pricing isn’t automatically better for customers.

The biggest advantage of seat pricing is predictability.

If a firm buys 500 licenses for the year, its finance department knows exactly what the software bill will be. With consumption pricing, however, successful adoption can actually become a budgeting challenge.

The more lawyers use AI, the larger the bill becomes. A major litigation or due-diligence project could suddenly generate enormous usage. An autonomous agent could potentially consume substantial resources while completing a complex workflow.

I’ve heard people complain that an agent consumed more than $10,000 worth of tokens, and they only discovered the cost when they received the bill.

That means legal departments may eventually find themselves dealing with something technology teams already understand very well:

Just as engineering teams monitor AWS or Azure consumption, legal operations teams may need to monitor AI consumption.

Of course, that assumes a law firm is large enough to have a legal ops department. Firms that don’t may eventually need to pay someone else to provide those services.

Who knew this was where we would end up when we started law school?

But there is no need to panic. Legora appears conscious of this problem and has already built tools to assist its clients. Its system includes real-time dashboards, notifications, usage thresholds, and spending controls.

Those features aren’t incidental. They are probably essential infrastructure for making consumption pricing acceptable to law firms.

The Pricing Unit Matters Too

There is another important question:

Are they consuming tokens, credits, agent runs, documents reviewed, tasks completed, or outcomes?

These aren’t the same thing.

Token pricing is closely connected to the vendor’s underlying technical costs, but it is difficult for lawyers to understand the value of “three million tokens.”

Charging per completed workflow is easier to understand, but workflows vary dramatically in complexity.

Charging for outcomes is perhaps the most attractive conceptually, but defining a legal “outcome” can become extremely difficult.

This may become one of the biggest pricing experiments in legal tech over the next several years.

The winning model may not be pure consumption pricing at all. It could be a hybrid:

That gives customers some budget certainty while allowing vendors to charge heavy users appropriately.

This Isn’t Just a Legora Experiment

The shift also reflects something happening throughout AI software.

Legora points to companies such as Cursor, Clay, and Lovable, as well as AI model providers themselves, as examples of businesses adopting consumption-oriented economics.

The underlying reason is straightforward.

Those are fundamentally different products.

And as AI becomes more agentic, the distinction becomes harder to ignore.

Why Legora Is Worth Watching

Legora isn’t a small company testing an obscure pricing experiment.

In March 2026, it raised $550 million at a $5.55 billion valuation. The following month, an extension increased the round to $600 million and valued the company at $5.6 billion.

And the company is still moving extraordinarily quickly. The Financial Times reported that Legora is already discussing another financing at a valuation of at least $10 billion, while its annual recurring revenue reportedly reached approximately $150 million in Q2.

So when a company growing at that speed decides that the traditional software seat no longer adequately captures the economics of its most advanced AI product, the rest of the legal tech industry should pay attention.

When an industry leader changes its business model, competitors often follow. One reason is that the first mover has already absorbed much of the initial customer backlash and tested whether the market will accept the change.

If Legora can move toward consumption-based pricing without losing customers, other legal tech companies may become more comfortable doing the same.

So the interesting question is whether the seat will remain the natural unit of value in an AI-native legal industry.

From Seats to Work

For decades, software pricing answered a simple question:

AI creates a different question:

That distinction sounds small. It isn’t.

As legal AI evolves from copilots that assist lawyers to agents capable of executing substantial portions of legal workflows, charging simply for the number of humans who can log in to the software makes less and less economic sense.

Consumption pricing has problems. Customers will demand predictability. Vendors will have to make costs understandable. Firms will need new governance systems. And the industry still needs to determine the correct unit of AI consumption.

But Legora’s move is significant because it acknowledges something fundamental:

Legal tech spent the last three years debating what AI could do.

The next debate may be about who pays for the work it does—and how.