Branden Jenkins was at dinner when a glance at his phone changed the evening. Checking his AI usage dashboard, the CEO discovered that his weekend coding session had cost him $1,000 — billed automatically, in $1,000 increments, to a card set to auto-renew.

Jenkins leads Maxio, a private-equity-backed software company headquartered in Atlanta that is on a path toward $100 million in annual revenue over the next couple of years. Maxio sells subscription-billing and revenue-management software to other B2B software companies. By his own admission, he also sits near the top of the company's internal AI spending leaderboard — an unusual position for a chief executive. “A thousand is not that much, I would say, but for one weekend, it's pretty annoying,” he told Fortune. Describing his AI agent as “cooking away,” he recalled his reaction: “Wow, I just got here quickly.”

Inside his company — and across corporate America more broadly — the episode has become something of a parable about how quickly “agentic” AI tools can consume money before anyone notices, until the bill lands. But Jenkins said the surprise invoice is not what keeps him up at night. The deeper problem is messier and more human: his own employees' “insecurity” about being outpaced by the technology — and by him.

How a weekend turned into a $1,000 lesson

Jenkins describes himself as a technical CEO who builds his own agents and automations. He can write code from his phone using Claude even when away from his desk — which is how he ended up debugging and iterating on a project over dinner. The token wallet he had set up to fund those sessions was configured to auto-refill by $1,000 each time it ran dry, silently recharging his card without requiring a second thought — until he saw the total.

“I don't have governors where a lot of my staff hits limits, and they have to ask for approval,” Jenkins said, describing his unlimited internal budget as both a perk and a liability. “So I started leaning in and going, ‘What does this look like?’”

AI providers such as Anthropic and OpenAI bill by the token — the chunks of text that models read and generate — so charges accumulate line by line as conversations lengthen and retries stack up. What he found, he said, is that much of the waste comes down to model selection and runaway conversational drift — an AI system wandering a user down paths they never intended to go. “A lot of times it's the agent's own mistakes that's burning your money,” Jenkins said. “You kind of find yourself just chatting, and [things] getting away from you.”

Casting his mind back to his dialogues with his bots, he said: “You're like, ‘Yeah, yeah, I like it, more of that, more of that.’ All of a sudden you end up in who-knows-where, and you're like, ‘No, I don't want that at all.’ So some of that money is just wasted because it took you there.”

His experience mirrors a pattern now well documented across the industry. Gartner has estimated that agentic AI models can require between 5x and 30x more tokens per task than a standard chatbot exchange, and a WitnessAI survey found that 68% of U.S. companies say at least some of their AI initiatives ran over budget in the past year, with a third saying overruns happen “mostly or always.” George Sivulka, CEO of Hebbia, put it memorably when he wrote that using agents means “you just hired a million bad employees.”

The fallout has even acquired a nickname: “tokenmaxxing,” industry shorthand for maximizing uncapped AI usage. Uber reportedly burned through its entire 2026 AI coding budget within four months, and Amazon reportedly spent $500 million on AI in a single month after rolling out access without usage caps — the month “tokenmaxxing” died. Jenkins's $1,000 weekend is a rounding error by comparison, but the point, he noted, is that it is real money. “There's no refund button. There's no dispute button in Claude,” Jenkins said, adding that maybe there should be.

Tricks of the trade

After the dinner incident, Jenkins looked for ways to cut his own token burn — much of it, by his account, learned from AI-optimization tips circulating on TikTok rather than from his own engineering team. He began routing different tasks to different models based on complexity: lighter models like Claude's Haiku, Anthropic's smallest and cheapest tier, for basic math, mid-tier models such as Sonnet for routine coding, and the most expensive, highest-reasoning models reserved for genuine strategic planning.

He also adopted what he called orchestration layers — third-party tools, often distributed as free GitHub repositories, designed to compress AI output and cut wasted tokens. One, which he called “Caveman mode,” forces an AI assistant to reply in short, blunt sentences instead of long, elaborated answers, which Jenkins estimated cuts token use by 70%. Another mode he described, “grunt mode,” compresses replies to a word or two: “It's very trite.” He named other tools, including “Superpowers” and “Ponytail,” as part of the same underground ecosystem of cost-saving hacks.

The catch, Jenkins said, is that none of it is accessible to a typical employee. “These are nerdy things,” he said. “Do we need sales leaders and service leaders and marketers finding this stuff?” That gap — between what power users like himself know and what the rest of the workforce can access — is where he says the real organizational risk begins.

Why he warns insecurity, not cost, is the bigger threat

Asked to rank the problems he has encountered rolling out AI across his several hundred employees, Jenkins did not lead with cost. He named three forces he says he has to actively manage: inefficiency, inequality, and — the one he returned to repeatedly — insecurity.

Jenkins said he is proud that everyone at his company is becoming a builder through vibe-coding permissions, but the shift is disruptive in a very human sense.

“[Token overspend] is really not the problem, but it could easily be the excuse,” Jenkins said. He described building tools and automations inside his own leadership team's departments, unprompted, simply because he had learned how — and watching the reaction turn uneasy. “One of them came to me and said, ‘This put me on edge. I should be coming to you with these things. I've got to catch up. I feel so behind.”

The dynamic, he said, repeated itself down through multiple layers of management: “Am I doing my job? Am I keeping up? Will this replace my job? Will this replace my team members?”

He also pointed to a version of the same anxiety playing out at the departmental level, rooted in unequal access to tools. His company initially rolled out ChatGPT company-wide, then began licensing the pricier Claude for a smaller group of roughly 50 employees concentrated in sales and marketing — prompting pushback from teams left out. “People were like, wait a minute, why don't I have Claude? Why do they get that and we don't get that?” he said. “That's an inequality.”

Jenkins argues that the anxiety is ultimately more corrosive to a company's culture than any single runaway invoice, because it shapes whether employees engage with the tools at all — or quietly resist them out of fear.

Building an org chart for humans and agents

Maxio's response, Jenkins said, has been structural. His entire executive leadership team went through a formal org-design exercise in which each executive mapped out their department not just in terms of the people who report to them, but the AI agents those people now manage directly. The result is a literal hybrid org chart — human names and agent functions layered together — that the company treats as a living management document.

He has also expanded the company's DevOps organization specifically to govern the growing number of employee-built, informally coded internal tools — what he and others in the industry call “vibe-coded” software — that have become load-bearing parts of the business despite originating as side projects. “It can't just be with Tim that vibe-coded it on the weekend,” Jenkins said, citing continuity risk if the employee who built a critical internal tool leaves or gets sick, along with unresolved questions about security and scalability. It is a governance question a growing number of companies now face as AI-assisted building spreads beyond engineering teams.

The bigger financial story, in his telling, is not the occasional four-figure token overrun but a shift in the company's underlying labor math. Annual recurring revenue per employee is a standard yardstick of operating efficiency in the software industry, measuring how much subscription revenue each worker supports. Jenkins said Maxio's headcount has stopped scaling with revenue the way it once did, driving up its ratio of annual recurring revenue per employee — a metric he calls the clearest signal of what AI is actually doing to his cost structure, as opposed to the more visible but comparatively minor token bill.

“I'm not arguing that we want to reduce a whole bunch of headcount because of AI, but we should not be growing the headcount at the same rate that we were before,” Jenkins said. “That's a big change in our business, and it's all attributed to AI.” He also agreed that the occasional ping of $1,000 in token burn at dinner is like a tax you pay — an AI agent colleague simply got the wrong idea of what the job was.

Jenkins said he does not see any of that as a reason to pull back. “Right now there's so much good that outweighs a lot of this,” he said, arguing that the efficiency gains from AI adoption are large enough that occasional waste, hallucination-driven detours, and even four-figure surprise bills are simply the cost of getting there. His head of engineering, when first told about Jenkins's token-saving tricks, waved them off, telling him the company was “not at the level of we're spending more on A.I. than [on] people, because there are examples of that out there. We're nowhere close to that.” Jenkins said he didn't disagree — but he is convinced that day is coming. “It probably will be a thing.”

For this story, Fortune journalists used generative AI as a research tool. An editor verified the accuracy of the information before publishing.

This story was originally featured on Fortune.com.