The Audit Nobody Wanted

One of our consultants texted me last week. Not with a win. With a number. His client had just gotten the Anthropic bill, and nobody in the room wanted to be the one to explain it.

We weren't surprised. We've been watching this build for months.

The WSJ piece that's been circulating confirms what we're seeing from the inside: companies that said yes to everything AI-related are now getting the invoices, and the invoices don't match the business cases. Uber burned through its annual AI coding budget in four months. Microsoft canceled most of its internal Claude Code licenses and told engineers to find a cheaper alternative.

This is not a story about AI failing. It's a story about adoption outrunning accountability. Companies maxed out usage before they figured out what problem they were actually solving (and some of them laid people off in the same breath, citing AI efficiency gains). That's a hard room to walk back into when you led the charge on both decisions.

The use cases that hold up under scrutiny are those where the output is measurable. Code is the clearest example. You can defend that line item. A lot of the others? Much harder.

What we keep telling our clients: find the places where AI actually delivers, build the business case before you add the seat, and don't be embarrassed when the audit comes. The audit was always going to come. The companies that built discipline into the rollout are fine. The ones that tokenmaxxed first and asked questions later are having a different kind of meeting right now.

What's your AI spend actually tied to? Not the vision. The specific outcome that justifies the line item. If you can't answer that today, someone else will ask it for you soon.