One of the things we are seeing in the field right now is people talking about their own tasks in terms of dollars.
Not budget in the abstract. Not the cost of the license. The actual work in front of them. Someone will describe a thing they want to try with the AI, then pause and weigh whether it is worth the spend. A few of them do the math out loud. I have sat in a lot of rooms across a lot of industries over the last few years, and until recently I had never watched a group of employees price their own curiosity in real time. Now I am seeing it.
It is happening because the pricing model underneath these tools has changed. The major labs have moved enterprise customers onto consumption billing. Instead of a flat seat fee with a generous bucket of usage attached, organizations now pay a small per-seat charge and then pay for what they actually consume, metered by tokens at something close to API rates. Anthropic made this shift for enterprise this year, and the rest of the market is moving the same direction. The meter is now running, and for the first time it is running where the people doing the work can feel it.
The business case is real
I want to be fair to the other side of this, because the other side is not wrong.
A company exists to produce a return. Boards, investors, and finance leaders have every reason to want the cost of a tool tied to the value it produces. For most of the last two years, AI spend inside companies was a flat number that nobody could connect to anything. You bought the seats, you hoped people used them, and you had almost no way to see what the usage was buying you. Consumption pricing fixes that. It turns a vague line item into something you can measure, attribute, and manage. In that sense, watching a team think carefully about what their effort costs is a healthy thing. It is the first time a lot of these people have quantified their own work at all.
So I understand it. From the seat of a CFO or a board member, this is progress. The cost finally has a shape.
What it does to the person at the keyboard
Here is the part I do not love.
The same meter that gives finance a clean number gives the person at the keyboard a reason to hesitate. I watch people hold back from trying things because they are not sure where their allotment stands, or whether the experiment is worth what it might burn. The instinct to poke at the tool, to run something twice, to throw a messy real problem at it just to see what happens, that instinct gets quieter the moment the cost becomes visible. There is a gun-shyness I have not felt before in this work.
And the message many companies are sending alongside the new pricing makes it worse. When a business moves to consumption billing, the guidance to employees often becomes some version of "learn how to use the models efficiently so you don't waste credits." I understand the impulse. But that cannot be the headline. If the first thing a frontline user hears about a powerful new tool is how to use less of it, you have told them to be careful before you have told them to be curious. Curiosity is the entire point. It is what the labs themselves say they want, and it is where the real productivity has always come from in our engagements.
This is the tension. The thing that makes the technology valuable is people exploring it without much fear. The thing the pricing model encourages is people rationing themselves. Those two pull in opposite directions, and right now the pricing is winning.
The labs are heading where they have to head
Step back and the larger move comes into focus. For the first couple of years, the frontier labs heavily subsidized the cost of compute to drive adoption. They ate the bill so people would get hooked on what the tools could do. Consumption pricing is the planned handoff. The cost of compute is moving off the labs' balance sheets and onto their customers'. The compute bill is real and enormous, and someone has to carry it, so this is the direction they almost have to go. It is also a brilliant piece of business, the kind people will study in business school years from now.
The launch of Claude's newest model this week is the clearest illustration I have seen of the pattern. Anthropic shipped it on June 9. It is priced at roughly twice the token cost of the prior flagship, and it consumes plan quotas at about double the rate. At the same time, the company is including it at no extra charge for subscription and seat-based plans for a short window, through June 22, after which continued use gets metered at usage rates. Read that as two messages at once. The first is real: come experience how capable this is. The second is also real: once a team feels what the better model does for their work, the willingness to keep paying for it goes up. Both things are true, and that is what makes it work.
So I am not saying the labs are confused about their own strategy. They know exactly what they are doing at that level, and from where they sit, the plan is sound and the adoption numbers probably look healthy. What I am less sure they can see is what the meter does once it reaches the floor. That part is hard to watch from headquarters. It is easy to watch in the field, on site with the people actually using the tools, where you see a curious employee go quiet the moment the cost becomes visible. The strategy and the downstream effect are two different things. The labs have the first one figured out. I do not think they have fully reckoned with the second, because the second only really shows up where we happen to spend our days.
There has to be a better abstraction
I do not have the clean answer to this. Anyone who tells you they do is selling something. But the shape of a better approach seems clear enough to describe.
The meter and the message do not have to sit in the same place. Frontline users should get an allotment generous enough that normal exploration never feels expensive, so the daily experience is "go try things," not "watch your credits." The cost accounting can live behind the scenes, in an admin view, where it belongs. Give the organization a real lookback, thirty days or a quarter, to see who the genuine power users are and decide which of them are worth paying for. Most companies will happily fund their heavy users once they can see the value those users produce. The rest of the usage is noise, and it should not be the thing that teaches a hesitant employee to flinch. Separate the management of cost from the experience of use. The people doing the work should feel encouraged, while the people managing the spend get the visibility they need.
That is not a finished product. It is a principle. But the principle matters, because the alternative is a generation of users who learned to treat a transformative tool like a taxi with the meter running.
Why this is on our radar
We are navigating this moment with clients right now, and this is not an isolated signal. We are seeing the same hesitation show up in more than one place. Interns and junior people run out of their allotment and stall. Teams second-guess whether a useful experiment is worth the spend. The friction is small in any single instance and significant in aggregate, because it lands hardest on exactly the exploratory behavior that separates the teams who get value from AI from the teams who bounce off it.
None of this changes the underlying truth that keeps showing up in every engagement we run. These tools are genuinely reshaping how people work. The change is hard, it takes real change management, and it is material once it takes hold. When someone crosses over and starts doing their actual job with AI in the loop, they do not go back. That is precisely why the pricing question matters so much. The meter is going to teach people something in these early months. The open question is whether it teaches them to explore or to hold their breath.