The New Math of Enterprise Software: Seats ➡️Usage

There was a time when slapping “AI-powered” on a product was enough to get attention. Now it’s just what’s expected. Open any enterprise software website today and you’ll find an AI assistant, copilot or agent on the first page, almost without fail. AI has gone from differentiator to table stakes.

For vendors, that creates a real problem. AI is everywhere in the product now, but it isn’t free to run. Every AI interaction pulls in model inference, compute, context retrieval, tool calls — real infrastructure cost. And as vendors move from simple copilots to agents doing longer, multi-step work, that cost adds up fast.

At the same time, AI economics are shifting under everyone’s feet. New models keep getting cheaper, and companies are getting smarter about routing tasks to the right model. One recent study of frontier reasoning models found something counterintuitive: the model with the lower sticker price per token was sometimes more expensive to actually use, because it burned through far more reasoning tokens to do the same job.

So vendors are pulled two ways at once; AI is getting cheaper to run, and AI usage is exploding. That tension is where the pricing conversation gets interesting.

Everyone wants AI. Someone has to pay for it

The vendor side of the story

For now, most software companies still bundle AI into existing subscriptions. However, a different model is emerging.

Microsoft’s Copilot Cowork is a good example. The regular Microsoft 365 Copilot subscription stays in place, but Cowork adds usage-based pricing on top for complex, long-running agentic tasks — the number of Copilot Credits burned depends on model use, context retrieval, tool calls and runtime. Microsoft talks openly about cost management, spending limits and ROI tracking, which tells you something: this isn’t a footnote feature, it’s meant to be watched.

GitHub has taken a similar step. As of June 2026, all Copilot plans run on GitHub AI Credits, with a monthly allotment and the option to buy more — meaning a GitHub bill that used to be one predictable line item can now move month to month.

Salesforce is the most explicit of the three. Agentforce can be bought through traditional user licensing or through consumption models like Flex Credits and Conversations, where a single Flex Credit ties to one AI action — answering a question, updating a record, running a workflow. Buy Agentforce, and you’re no longer just budgeting for seats; you’re budgeting for actions taken.

Different models, same direction: the seat is no longer the only thing being priced.

And that’s where the bigger story starts — not for the vendors, but for the people buying this software.

The SaaS bill is about to get a lot more complicated

The buyer side

For years, buying SaaS was straightforward. A business team estimated headcount, procurement negotiated price per seat, finance approved the annual number, and everyone knew roughly what next year’s bill would look like.

AI adds a variable nobody budgeted for: consumption.

Picture a marketing team running 15 different SaaS tools — CRM, marketing automation, analytics, SEO, design, content, research, sales enablement. Most of them now have AI built in somewhere. If just five of those vendors move to usage-based AI pricing, the budget stops being a clean multiplication of Number of users × price per sea and becomes an additional problem involving Users + AI credits + conversations + actions + consumption.

That’s a different kind of homework for the buyer. Finance and Procurement team needs visibility into AI consumption, not just annual contract value. Procurement needs to actually understand credits, tokens, actions, usage caps and overage charges — not just negotiate a discount on them. Technology teams need to know which workloads are driving the spend, and whether the priciest model is even necessary for the task at hand.

The next budget won’t just count seats , it will count value.

I don’t think the seat is going away. It’s simple, predictable, easy to manage. But I do think seat-based pricing is about to sit alongside value- and consumption-based pricing — and that changes how companies build their annual technology budgets. AI spend deserves its own line in that budget, not a spot buried inside the SaaS number.

The next time a finance leader opens the software budget, the question won’t just be “how many licenses are we buying?” It’ll be closer to: what did we actually use, and what value did we get from it?

Before your next renewal, three things are worth doing regardless of vendor: ask what the AI is metered on, get a 90-day usage baseline before you commit to a credit pool, and put one named person in charge of watching the meter.

We don’t yet know exactly where AI pricing will settle. It may end up being seats plus usage. It may eventually move toward outcomes entirely. Different software categories will probably get there differently.

But one thing feels certain: AI isn’t just changing what software can do. It’s changing how software can be bought, measured and valued.

Everyone wants AI. Someone has to pay for it.