No visibility, no cap: why finance leaders lost track of AI spend
By The Lucidrail Team · 2026-06-28
In the first half of 2026, something unusual started happening inside large companies. The people who sign off on budgets began admitting, in public, that they had lost track of what their AI tools were costing. TechCrunch gathered several of these moments in a June report, and they rhyme.
The line that captures the mood came from the head of the FinOps Foundation, a group whose entire job is watching cloud spend. He was passing along what one of his member companies had told him.
Oh my god, we are 3x over our entire 2026 token budget and it's only April. — A customer quoted by J.R. Storment, Executive Director of the FinOps Foundation (TechCrunch)
Three times the whole-year plan, with two-thirds of the year still to go. But the number is not the interesting part. The interesting part is what came next: no one could say whether that spend was a problem or a triumph.
Spend with no verdict
A cost only means something next to a result. A bill is high or low depending on what it bought. The uncomfortable truth in these accounts is that the buyers could not make that call. One story, relayed by the chief executive of an analytics company, sits right on the line:
One of my engineers spent $40,000 on tokens last month, and I genuinely don't know whether I should stop him or tell everyone else to be like him. — An anonymous CTO, relayed by Vitaly Gordon, CEO of Faros AI (TechCrunch)
Stop him, or clone him. That is the whole problem in one sentence. Forty thousand dollars is either reckless waste or the best money the company spent all quarter, and the person paying cannot tell which. The dollar figure is not what is broken. The missing verdict is.
The question that follows
When leaders cannot judge the spend, they all reach for the same next question. The head of enterprise at OpenAI described the shift plainly:
Now the conversations are about, ‘hey, we’re spending so much. What visibility do you have?’ — Alexander Embiricos, Head of Enterprise at OpenAI (TechCrunch)
Visibility. Not a smaller bill, not a ban on the tools — a clear, live picture of where the money is going and why. Almost every complaint in the report reduces to that one gap. People are not asking to spend less. They are asking to be able to see.
No visibility, no cap
Here is the part that tends to get skipped. Visibility is not just a comfort. It is the thing that makes a limit possible at all. A cap is a rule that checks spend against a number and stops when the line is crossed. That check needs a live figure to read. When the only figure anyone has is the monthly statement, the cap has nothing to act on until the month is over — which is to say, until the money is already gone.
So the two problems are really one. You cannot stop what you cannot see. A team without visibility does not have a spending problem and a control problem as separate items. It has a single missing layer that would have provided both.
That layer is not exotic. It comes down to three plain abilities:
The companies in the TechCrunch report had none of the three. That is why the story was not about one bad tool or one careless engineer. It was the same admission over and over, from finance and engineering alike: the spend is happening, and we cannot see it in time to do anything about it.
This is the layer we set out to build at Lucidrail: one console that shows what every agent in your fleet is doing and spending in real time, per-company budgets with hard caps that stop a runaway before it bills, and an audit trail where each action traces back to the goal that asked for it. The aim is not to spend less on agents — it is to be able to see, judge, and stop, instead of reading the verdict a month too late.
- See: a live view of what every agent and tool is spending right now, not on a statement next month.
- Judge: enough context to tell good spend from wasted spend while it is still happening.
- Cap: a hard limit that reads that live number and stops the work before the bill lands.