Your Software Bill Lost Its Ceiling

I’ve watched this play out at more than one company I’ve worked with. A team hits its AI usage cap halfway through the quarter. Work stops. Someone sends an urgent message. And because nobody decided ahead of time what happens next, the decision gets made in a panic, usually by whoever is loudest.

That used to be rare, because software used to be the easiest line in the budget. That’s changing fast.

The seat had one great feature: a ceiling

For about 20 years, most business software was priced per seat. Count the people, multiply by the seat price, done. A thousand people at $30 a month is $30,000 a month, whether they use the tool once or a thousand times. You knew the number in January, and it was the same number in December.

Think of it like a gym membership versus a car service. The gym charges a flat fee whether you go every day or never. A car service bills by the ride: use it more, pay more. Seats were the gym. AI is turning software into the car service.

The reason is simple. A seat measures who can log in, not how much work gets done. When an AI agent processes invoices or closes tickets, the work goes up while the number of people logging in stays flat or drops. Under seat pricing, the vendor gets paid less as its product does more. So vendors are moving the meter from people to work.

Who’s moving, from most fixed to most variable

• Figma gives every seat a monthly allowance of AI credits. Since March 2026, when a seat runs out, the paid AI features turn off until the next reset unless you buy more. The seat is still there. It just has a fuel tank now.
• GitHub Copilot moved to usage-based AI credits on June 1, 2026. The seat price didn’t change: Business is still $19 a user. But existing Business customers’ promotional allowance dropped from 3,000 credits per seat to the standard 1,900 on September 1. The credits are pooled across the company, so a few heavy users can drain them for everyone.
• Workday spent two decades pricing per employee. Its AI agents now run on Flex Credits, an annual pool drawn down each time an agent completes a task. One licensing advisory firm says about half of companies running agents in production used up their pool well before renewal.
• Salesforce Agentforce launched at $2 per conversation, then added Flex Credits at about 10 cents per action. You pay per action, not per successful action, so an agent that loops or misfires still runs the meter.
• Clay, a data tool a lot of sales and marketing teams use, dropped seats entirely. Unlimited users, two meters: data credits for lookups and actions for workflow steps. Run out mid-month and you can top up, at a premium above your plan rate.
• treg sits at the far end. It’s a single connection that lets an AI agent call more than 4,000 tools. No seats, no subscription, just a price per call. I use it, and as a one-person shop it’s been fine. On a big team, with agents making thousands of calls and people doing duplicate work, it could add up very quickly.

Same direction every time: the fixed part shrinks and the variable part grows. Kyle Poyar’s 2026 State of B2B Monetization survey found hybrid pricing, a base fee plus a metered layer, is now the most common model at 37%. Only 5% said investors prefer seat-based pricing.

Why this is a finance problem

First, your budget lost its ceiling. With a seat, the contract value was the most you’d pay. With credits, it’s the least you’ll pay. What you actually spend depends on how much people and agents use the tool.

Second, reliance. A meter is easy to live with in the pilot, when nobody depends on the tool. Once the month-end checklist runs through it and the sales team’s lead list comes out of it, hitting the cap doesn’t just slow people down. It stops work. And because nobody wants work to stop, the cap almost always gets raised, usually at a worse price and without much discussion.

Four things a tool can do at the cap

1. Block. Features turn off until the reset, like Figma. Your spend is protected, but your team is stuck.
2. Bill overage automatically. Usage keeps going and you see it on the invoice. My own business account tops up automatically in $20 increments, with a $200 monthly limit unless I raise it. That works for one person. Bigger teams with super users need much more thought.
3. Top up at a premium. Someone has to decide to buy more, at a rate above the plan price, like Clay.
4. True up at renewal. The annual pool feels fine all year, then the overage shows up when you renew.

None of these is wrong. The problem is not knowing which one you have.

Write the plan before the alert goes off

• Know your terms. Which of the four is it? What’s the overage rate? Is the reset hourly, daily, monthly or annual?
• Require alerts and limits. People should see how fast they’re using credits before they hit the cap, and no one should be able to blow past their limit.
• Set approval levels. Maybe a manager can approve a small top-up, and anything bigger goes to an AI committee. Decide it now, not at 2 a.m.
• Decide priorities. If a shared pool runs low, the month-end close matters more than someone redoing slides.
• Reforecast more often. A seat contract you forecast once a year. A meter you forecast monthly or more, because slow starters suddenly spike and heavy users drop off.

Key takeaways

• Software is moving from a fixed price per seat to a variable price per use.
• With credits, the contract value is the floor, not the ceiling. Budget off usage, not headcount.
• Reliance turns a cap into a work stoppage.
• Every metered tool blocks, bills overage, sells a top-up or trues up at renewal. Know which one you have.
• Decide who approves, up to how much, and which work comes first before the alert goes off.

Want the CPE credit? Take the full lesson on EverydayCPE and earn 0.2 CPE credits: Your Software Bill Lost Its Ceiling