Will AI pricing become a charge-back cost to clients?

I spent a good part of last week at the ILTA Evolve conference in Colorado, and I heard it mentioned a few times: the potential for very serious price increases.

I wrote last week that the current pricing models are not sustainable. The link from Ed Zitron provides the deep analysis that led him and me to that conclusion.

This week, I found this article – The Anthropic Pricing Shift Is a Signal, Not a Surprise

This isn’t an Anthropic-specific story. OpenAI, Google, Microsoft, and even legacy Work Tech platforms like SAP are all moving in the same direction. The trigger is the same everywhere: AI broke per-seat economics. When value came from software access, per-seat pricing made sense. When value comes from outputs: decisions made, documents processed, candidates screened, and the like, the vendor’s cost is no longer fixed. It’s variable, and it scales with usage. Agentic workloads consume anywhere from 5 to 30 times more tokens per task than a standard chat interaction. The math didn’t work at the old price, and Anthropic made a choice: be honest about it.

I, sadly, don’t remember who said it at Evolve, or which session it was said in, but I heard someone almost offhandedly comment that maybe the model would be the old legal research model, where you got billed by the hour for access to the research database and bill that time back to the client.

The more I think about the way law firms budget, the more I think this makes sense.

We don’t have very much capacity to plan out for an “unknown” amount of compute power. I’m assuming there will always be a per-seat license that gives you the level of compute needed for administrative tasks, drafting emails, scheduling meetings, etc. It’s when you start getting into client work that agentic workflows and deep research really start to add up in token expenses. Those are also the areas where clients are starting to expect firms to invest in AI because it reduces their total billed hours.

Maybe we can solve both of those issues this way? We’ll bill you fewer attorney hours by leveraging AI, but you need to foot the bill for its use? Right now, with a set price per seat, the cost of using AI on a given matter is nothing, but it won’t remain that way. It can’t.

Will the legal market make this adjustment to billing back the cost of AI compute? If not, how do firms justify the undefined cost for your firm?

Follow these topics: Artificial Intelligence, LawFirms