We’re building something for the people now staring at the invoice. We’re not ready to say what yet. But we’re ready for you to be on the list.
The bill went up. The story didn’t. We’ve been collecting the pieces.
After his CTO disclosed Uber had exhausted its entire 2026 Claude Code budget by month four, COO Andrew Macdonald called it a "head-exploding moment" and drew the only honest conclusion available: the link between token consumption and shipped features is not there yet.
Microsoft is winding down most internal Claude Code licences across its Experiences and Devices division. The pilot launched only in December — six months in, the company is reportedly standardising developers on GitHub Copilot CLI, with cost control cited as the underlying driver.
The fact that tokens went from something no one even put in a budget line a year ago to an absolute requirement for coding now is the cause of handwringing, not that AI is not turning out to be useful (& increasingly necessary). No one knows who should get tokens, how much & how to control it all.
This is what we’ve been seeing with every company we work with.
Try justifying spending $100k on token spend when only $18k even makes it to a stable prod feature.
In the rush to maximize AI token spend, companies are wasting over 44% on bug fixes.
Incredible to me that the companies that publicly prided themselves in being data-driven now have token leaderboards. No related metrics for value created or if what’s being shipped isn’t broken, just number of tokens used.
Of course this isn’t working. You’re being willfully dumb.
Average monthly spend per Uber engineer was reported between $150 and $250, with heavy users said to reach $2,000. Naga reportedly spent $22k in a single two-hour demo. Uber did not stumble into this passively — it ran internal leaderboards ranking engineers by Claude Code activity.
“When a measure becomes a target, it ceases to be a good measure.”
In 2026, the measure is tokens. Tokenmaxxing is just Goodhart wearing a hoodie. We made it the metric, so the agents made it the goal. Now four C-suites in one month have said the ROI isn’t closing. The thing in the way is the unit. We’re building a different one.
Tokenmaxxing is just electricity-maxxing.
Picture a factory CEO walking around saying “look how productive we are!” while every single light is on, all the machines are idling, and he’s bragging about the power bill. That’s what a lot of AI companies sound like right now.
"Amazon employees are doing random unnecessary task automations to consume tokens and to show their bosses that they’re using AI more," per FT.
The token furnace.
Engineers have stopped their primary roles and simply research the latest tool and give classes every week because of a blind post saying the only people to be promoted are working in AI. No one is creating anything of value. Middle managers are desperate to show they are team players, so everyone is full on tokenmaxxing.
Nvidia exec Bryan Catanzaro recently commented that, within his team, AI compute power is more expensive than actual workers. Almost 80,000 layoffs in Q1 were pinned on AI — even as the math powering that decision quietly stopped working.
Do you want your best engineers to use the most tokens? Your worst? Can you even figure that out? Will your other engineers be resentful that they can’t burn as many tokens? How do you deal with competing projects and constraints?
The AI subsidy era is ending in real time. The same company that put $13B into OpenAI and built the Azure infrastructure powering most of Anthropic’s compute just looked at the bill from a competitor’s coding tool and decided it was not worth paying.