AI & Machine Learning

Your team is spending $5K a month on AI tools. Where’s the ROI?

Engineering teams are spending more on AI coding tools than ever. Most have no idea what it’s producing.

Billy Boozer | CTOJul 9, 20263 min read

$3K

Routine AI spend per developer, every month

$1.3M

Run up by a ~3-person team in a single month

603B

Tokens consumed across 7.6M API requests

$500M

One enterprise’s Claude bill in a single month

AI coding tools are now a standard line item in every engineering budget. The problem isn’t whether they work — they do. Developers ship faster with Claude Code, Cursor, and Copilot than without them. The problem is that most engineering leaders have no real visibility into AI tool costs, no attribution by project or developer, and no way to answer the CFO’s basic question: what are we spending, and what are we getting for it?

That’s not a hypothetical. It’s what’s happening right now at some of the largest companies in the world.

The AI spending problem is bigger than you think

In early 2026, Uber’s engineering teams adopted Claude Code aggressively. By April, they had consumed their entire annual AI tools budget. Sundeep Gupta, Uber’s CTO, was direct about it.

“We saw limited correlation between
AI spend and measurable ROI.”
Sundeep GuptaCTO, Uber

That’s not a small company without processes. That’s Uber. A ~3-person team called OpenClaw ran up $1.3M in AI tool costs in a single month: 603 billion tokens, 7.6 million API requests, no usage guardrails in place. One enterprise hit $500M in Claude API charges in a single month with no caps and no governance. Microsoft quietly revoked thousands of Claude Code licenses over unsustainable token costs. Routine per-developer AI spending is now $3K/month, with single sessions topping $1,400.

These aren’t edge cases. They’re the new normal for teams scaling AI coding tools without a visibility layer.

Three reasons AI tool governance breaks down

Costs are unbounded.

Spend on AI coding tools scales directly with usage, and usage has no natural ceiling. Bills arrive after the fact, often with no breakdown by developer, project, or model. By the time the number lands, the damage is done.

Compliance exposure is real.

Every prompt sent to an AI provider is a potential leak of secrets, PII, or PHI. Most engineering teams have no audit trail, no redaction policy, and no way to prove what was or wasn’t sent last quarter. When the compliance conversation happens — and it will — most teams aren’t ready.

AI productivity metrics don’t exist.

Developers are clearly shipping faster with AI. But “faster” isn’t a number you can put in a board deck. Without attribution by project and developer, there’s no way to tie AI tool spend to actual deliverables or measure ROI in any meaningful way.

What real AI spending visibility requires

Solving this isn’t about restricting tools or cutting budgets. It’s about building the infrastructure to see what’s happening. Specifically:

Cost attribution

By developer, project, and model — not just a monthly invoice total from five different portals.

Real-time alerts

Immediate signal when spend exceeds thresholds or usage spikes unexpectedly.

Compliance scanning

On every AI interaction, with an immutable audit log that auditors can verify independently.

The question worth asking now

If your CFO asked Monday morning what you spent on AI tools last month and what it produced, could you answer? If not, you’re not alone. But the engineering organizations that build this visibility layer now will have a measurable advantage over those still assembling answers from disconnected billing portals.

If you want your organization to have full visibility into AI tool spend, reach out to us. We’ll show you exactly how Aixle works in your environment.

Webinar

See the numbers behind the spend

Where AI budgets actually go

Dualboot CTO Billy Boozer walks through the data, the pattern behind runaway AI tool costs, and what engineering leaders are doing about it — with live examples from real deployments.

Watch the recording