The Signal #012 — Dakota’s read on the AI news that actually matters to people running a business.

Most people read a headline about Uber’s AI spending and think, “That’s a big company problem.” It isn’t. The number they landed on tells you something useful about where AI tool costs are heading, and what a reasonable budget looks like when someone actually does the math.

What happened

Uber blew through its entire 2026 AI budget in four months. After that, they put a policy in place: every employee is capped at $1,500 per month in token spending (a token is a small chunk of words or code the AI reads or writes) per AI coding tool. As Simon Willison reported on June 3, 2026, citing Bloomberg, the cap applies specifically to agentic coding software (AI tools that write and run code on your behalf, like Cursor or Anthropic’s Claude Code). Each tool gets its own $1,500 bucket. Using two tools doesn’t mean you split $1,500. It means you get $1,500 per tool.

Willison ran the numbers from there. If an engineer actively uses two tools, that’s $3,000 a month, or $36,000 a year per engineer in AI tool spending. The median yearly compensation package for an Uber software engineer, per Levels.fyi, is $330,000. That puts the AI spending cap at roughly 11% of median compensation. Willison also noted that his own usage runs about $1,000 per month against each of Anthropic and OpenAI, which currently costs him around $100 per provider on subsidized individual plans. Those subsidized plans are not available to larger companies like Uber.

Why it matters for operators

You are not Uber. You are running an HVAC company or a roofing crew or a plumbing operation. But the pricing signal here still applies.

AI tools for your business are not free, and they are not going to stay cheap forever. The subsidized, nearly-free pricing that individual users get right now exists because these companies are buying market share. When you are a business, the pricing is different. When the subsidies shrink, the pricing gets closer to what Uber is paying.

The $1,500 per month figure is also a useful benchmark for thinking about value. If a tool saves a software engineer at Uber enough time to justify $1,500 a month, ask yourself what the equivalent looks like in your business. An AI answering service that handles missed calls. A scheduling assistant that stops double-books. A follow-up tool that texts estimates without you touching your phone. None of those cost $1,500 a month right now. Most cost a fraction of that. The value math is actually easier for a home services operator than it is for an enterprise engineer, if you are paying attention to it.

The other thing worth noting: Uber burned its annual budget in four months because they set that budget in 2025, before anyone understood how fast people would actually use these tools. That is a planning mistake, not a technology mistake. Costs can scale faster than you expect when AI tools become part of daily workflows.

What most people get wrong

The reflex reaction to a story like this is to see it as a warning. AI is expensive. Big companies are cracking down. Better wait and see.

That is the wrong read. Uber did not cap usage because the tools were not working. They capped usage because people were using them so much that costs outran a budget set before anyone knew how popular the tools would become. That is a sign the tools are delivering something real. Nobody burns budget on something useless.

The other mistake is thinking that enterprise pricing decisions have nothing to do with small operators. They do. Enterprise contracts fund the model development that makes the tools you use better. Enterprise overspending is what pushes providers to build better pricing controls, which eventually become the tiered plans small businesses actually buy. Uber’s situation is an early signal of a maturing market, not a collapse.

And the last mistake is treating AI tool costs as a line item to minimize. The right question is not “how do I spend less on this.” It is “am I getting more back than I am putting in.” Willison’s math works out to 11 cents of AI spending for every dollar of engineer compensation. If your AI answering service costs you $150 a month and converts two jobs you would have missed, the math is not complicated.

The short version

A $1,500 monthly cap from a company that just overspent its AI budget tells you two things. First, these tools have real costs that scale faster than most people plan for. Second, someone inside Uber decided the tools were worth capping rather than cutting, which means they are working. For a small operator, the lesson is the same as it always is. Know what you are spending. Know what you are getting back. Adjust accordingly.

If you want help thinking through what AI tools actually make sense for a home services business, and what to expect on cost and payback, start at xovionlabs.com.