Anthropic launches tools to measure the return on enterprise AI spending

Anthropic announced on Wednesday (28th) a new set of reporting tools aimed at enterprise customers, with the goal of showing in granular detail what each team is getting in return for its token spending with Claude. The information comes from TechCrunch.

The package combines internal usage data with surveys conducted directly with employees. Based on this, managers can view, for a chosen period, what a team produced, how much it cost in tokens, and at what point in the workflow bottlenecks occurred. The reports also include recommendations on how to improve these results, and Anthropic states that it will not use this data to train models by default.

Why Anthropic is doing this now

The launch comes alongside financial figures shared with investors this same week: the company posted operating profit in the second quarter and a net revenue retention of 900% among recurring customers, meaning spending by the same customers in the third quarter of 2026 grew ninefold compared to the third quarter of 2025.

This growth in spending is exactly what's driving the new tool. Large enterprise customers have been demanding a straightforward answer from Anthropic to a simple question: what is the money invested in Claude Code actually bringing back? The company is also creating pooled budgets by team, so that teams can manage token consumption collectively instead of relying solely on individual limits.

What market numbers already show

An independent snapshot of this dynamic came from a report by Jellyfish, a company that makes tools for engineering teams, released on Tuesday (27th) based on the use of agentic coding tools by 276,000 customers. Claude Code and other tools built on Anthropic's models remain in the lead, accounting for nearly a third of all code changes proposed and submitted for review.

More revealing is the pace of spending: according to Jellyfish, investment in these tools has grown 10-fold since the beginning of the year, and now accounts for about 2% of total engineering costs at the companies analyzed.

Even if you're saying 2% is still small, that 10x growth is growth that most CFOs didn't plan for. We're certainly getting to the point where people are realizing, okay, it's not enough to just spend a bunch. You now have to optimize it, make it work.

Nicholas Arcolano, chief scientist at Jellyfish

The dispute with OpenAI hasn't shaken Anthropic

The launch of OpenAI's Astra model this past North American summer seemed like a direct threat to Anthropic's position among enterprise developers. So far, according to Jellyfish, that hasn't slowed adoption of Anthropic's tools among this audience.

Nikolas Albarran, head of AI product at Jellyfish, described a disconnect between public perception and the actual data on enterprise usage.

It took a long time for us to see Codex really pop in our data at all, and it has come up with Astra.

Nikolas Albarran, head of AI product at Jellyfish

Looking at X, he continued, you get a clear impression that competition has heated up. "Everyone in the Valley is like, 'oh, Codex is better,'" said Albarran. In reality, he continued:

Anthropic was just there with great enterprise motions, and they had an incumbent weight of their own, this stuff just worked and worked well, and a lot of people just put a lot of effort into standardizing workflows around it. And there's not a big enough reason to switch.

Nikolas Albarran, head of AI product at Jellyfish

The cost of switching tools isn't zero

Albarran points to a structural factor behind this loyalty: the investment AI labs have been making in forward-deployed engineering (FDE) teams embedded with the customer, who help the organization get agentic workflows up and running. Anthropic's new usage reports function as a leaner version of this, automated tracking in place of in-person support.

Once a team has standardized its workflows around a specific tool, switching becomes expensive. Engineering projects develop their own conventions (from the structure of agent.md files to automation scripts) that don't migrate for free to another model or another vendor.

We're at a point now where it's mature enough, where once you're in, it's going to be harder and harder to switch because you've spent the effort. Even if they're all mostly the same, they're all going to be slightly different, have their own slight quirks of how they handle agent.md files, etc., and so that cost to switch is not zero.

Nikolas Albarran, head of AI product at Jellyfish

What changes for developers and engineering leaders in Brazil

For Brazilian teams, spending on frontier models tends to hurt more: it's billed in dollars, while the engineering budget is planned in reais. A tool that shows, by team and by period, how much each squad spent on tokens and what that produced is exactly the input that's missing today to justify (or cut) investment in Claude Code in front of a CFO who watches the exchange rate closely.

It's also worth noting the data governance detail: Anthropic states it will not use the information from these reports to train models by default. This matters for teams operating under LGPD (Brazil's data protection law) that already hesitate to expose internal productivity metrics or code snippets to a third-party vendor.

And there's a more strategic point for those deciding on architecture: Albarran's argument about switching costs isn't just a market observation, it's a warning for architects who are standardizing agentic workflows right now. Conventions like agent.md files, orchestration scripts, and model-specific integrations carry real migration costs. It's worth designing these workflows with portability in mind from the start, rather than discovering the price of lock-in only when it's too late to switch vendors.