AivexaNewsSearch
AI news for builders and product teamsChecked every hour

Anthropic just launched tools to show companies what their AI spending is buying

Collected Oct 28, 2026

Anthropic on Wednesday announced a suite of reporting tools aimed at enterprise customers who want a clearer picture of what their token spending is actually producing. The tools let managers use internal analytics and employee surveys to view a team's work over a chosen date range, what the team produced, how much it cost, and where its workflows broke down. The reports also offer recommendations on how to improve results. Anthropic says it will not use this data for model training by default.

The company is also introducing pooled group budgets, which let teams manage token spending collectively rather than account by account. Together the two features address a practical gap in AI procurement: usage dashboards typically show consumption, not output, so managers can see a rising bill without an easy way to tie it to shipped work. Anthropic's framing puts cost and output in the same view, with a suggested next step attached.

The launch lands alongside financial numbers Anthropic shared with investors this week, which the company says show an operating profit in the second quarter and 900% net revenue retention — meaning spending from the same customers in the third quarter of 2025 grew nine times by the third quarter of 2026. Claude Code, the company's AI coding tool, is presented as the driver behind that performance. Context for the timing: the reporting tools arrive ahead of a planned IPO, so demonstrating that customers extract measurable value from Anthropic's models serves both retention and the investor story.

Third-party data points the same direction on adoption. Jellyfish, which makes tools for software developers, published a report based on usage of agentic coding tools by 276,000 of its customers. Claude Code and other tools built on Anthropic models remained the most widely used in that survey, accounting for nearly a third of all proposed code changes submitted for review. Spending on those tools has grown tenfold since the start of the year and now equals about 2% of total engineering cost. "Even if you're saying 2% is still small, that 10x growth is growth that most CFOs didn't plan for," Jellyfish chief scientist Nicholas Arcolano told TechCrunch, adding that the question has shifted from how much to spend to how to make that spend work.

Anthropic leads among enterprise customers, and Jellyfish's data suggests that lead has been durable. OpenAI's Astra model, released this summer, looked like a potential threat, but it has not slowed Anthropic's adoption numbers so far. Nikolas Albarran, Jellyfish's AI product lead, said Codex took a long time to register meaningfully in the company's data and has only picked up with Astra, even as online discussion portrays a more heated contest. His read is that Anthropic had strong enterprise sales motions and an incumbent position: workflows were standardized around its tools and they worked well enough that switching lacked a compelling reason.

The stickiness goes beyond model quality. Once a frontier lab is embedded in a large organization, switching carries real cost — accumulated effort, plus small behavioral differences in how each vendor's agents handle files and similar mechanics. Albarran noted that all the major labs are building forward-deployed engineering organizations, teams of embedded consultants who help customers get running. The new reporting tools can be read as a lighter version of that hands-on help, delivered through software rather than staff.

For developers and product teams, the practical change is that token consumption becomes a line item someone can interrogate, with output attached. This suggests internal pressure to justify agent usage with produced work rather than seat counts will intensify, and that budget pooling shifts spend control from individual accounts to team-level allocations — a structure that can simplify chargebacks but also make one team's overrun another team's constraint. The default opt-out from training is a meaningful procurement detail; in practice, enterprises weighing vendors will want to confirm how survey data and work-product metadata are stored and retained, since the reports combine subjective employee input with usage analytics. A likely trade-off is that cost-per-output reporting rewards easily counted artifacts — merged pull requests, tickets closed — which may not capture exploratory or maintenance work.

Why it matters: teams already spending heavily on agentic coding tools now get vendor-supplied numbers tying spend to output, which changes renewal and expansion conversations from usage volume to measured return. Expect finance and engineering leadership to ask for these reports by name, and competing labs to answer with comparable dashboards. The bigger risk, by inference, is that the metrics chosen become the target, nudging teams toward work that reports cleanly rather than work that matters.

Read at TechCrunch · AI

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

The frontier lab is lifting the curtain on how users spend money on its models ahead of a planned IPO.