Why Your AI Tools Are Switching From Flat Fees to Meters

On May 4, Notion stopped charging a flat fee for its AI agents and started charging by the credit.

The new rate is ten dollars per thousand credits, pooled across your workspace, reset every month, with nothing rolling over. GitHub did something similar on June 1, moving Copilot from a predictable subscription to metered billing. Neither company made much noise about it.

That quiet is the part worth paying attention to.

For two years, AI inside your software worked like a buffet. You paid a fixed amount per seat, and your team used as much or as little as they wanted. The bill was boring and predictable, which is exactly what you want a bill to be.

That model is ending, and not because anyone is being greedy. The companies selling AI are losing money on every flat subscription, and flat pricing can't survive when the cost of the product scales with how much you use it. So the meter is coming. Notion and GitHub are just early.

Here's where I want to slow down.

A meter isn't automatically worse than a flat fee. If your team barely touches the AI features, metered pricing might cost you less than the old per-seat charge. The problem isn't the price. It's the variance.

A flat fee is a number you can put in a budget and forget. A meter is a number that moves, and it moves based on behavior you're not watching. That's a different kind of expense, and most businesses haven't adjusted their habits to match it.

Most of the risk comes down to one word. Agents.

When you run AI manually, a person types a request and reads the answer. The consumption is visible and self-limiting, because a human in the loop eventually gets tired. When you hand work to an autonomous agent, that ceiling disappears. An agent can run hundreds of times in an afternoon, each run quietly spending credits, and nobody is sitting there watching the counter tick.

That's how a tool you set up to save time turns into a bill you didn't see coming. You're not being careless. You budgeted for a buffet and got a taxi with the meter running.

So treat this as a prompt to do something specific this week.

Pull up the two or three tools your business relies on most and find out, for each one, whether the AI is billed flat or by usage. You're looking for words like credits, tokens, or consumption in the plan details. If you find them, you've found a number that can move.

Then ask who, or what, can spend that number. A person clicking buttons is one thing. An automation running on a schedule is another. The combination to watch is a metered tool wired into something that runs on its own, because that's where the surprise bills live.

The tools are starting to help, if you let them. Notion added a dashboard that tracks credit use in real time, warns you at eighty percent, flags unusual spend, and lets you cap how much any single agent can burn. Most metered AI products now ship some version of this. The catch is that none of it does anything until someone turns it on and sets the limits. A spending cap you never configured protects nobody.

There's a tension here worth naming honestly. The whole appeal of AI agents is that they work without supervision, and metered pricing punishes exactly that. The more autonomously a tool runs, the more it can spend, and the less likely you are to notice until the invoice arrives. You don't get hands-off automation and a perfectly predictable bill at the same time. Pick which one matters more for each task, and set the controls accordingly.

For most small businesses, this is a small adjustment, not a crisis. Check how your top tools bill. Turn on the alerts. Cap the agents that run unattended. Twenty minutes of setup buys back the one thing a meter takes away, which is the ability to not think about it.

The flat AI subscription was a nice two years. It made the cost of trying things invisible, which is part of why so many of us tried so much. That era is closing. The businesses that adjust early won't be the ones using AI the least. They'll be the ones who always know what their tools are about to spend.