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

Anthropic launched reporting tools that let managers see what a team produced over a date range, what it cost, and where workflows broke down, drawing on internal analytics and employee surveys. The reports come with recommendations, and Anthropic says it won't use the data for training by default. It's also adding pooled group budgets for token spending.

The demand is real. A Jellyfish report cited by TechCrunch found spending on agentic coding tools has grown 10x since the start of the year, now roughly 2% of total engineering cost, while most of the Fortune 500 still aren't seeing major gains from AI.

We like this. The single biggest question in enterprise AI right now isn't which model is smartest; it's whether the money is buying anything. Tools that show a team's output next to its cost, and point at where work stalled, push the conversation from vibes to evidence. That's healthy for buyers, and frankly healthy for the industry's credibility.

The numbers explain the urgency. Anthropic told investors it recorded an operating profit in the second quarter and 900% net revenue retention, meaning the same customers spent nine times more year over year. Jellyfish's chief scientist put it plainly: 10x growth is growth most CFOs didn't plan for. When a line item grows that fast, someone in finance starts asking for proof. Anthropic would rather supply the answer than have a CFO cut the budget looking for one.

Now the eyes-open part. These reports are also retention machinery. Jellyfish's AI product lead described Anthropic's position as incumbent weight: teams standardized workflows around Claude, and switching costs aren't zero even when tools are similar. A dashboard that lives inside the vendor's own product, measuring value by the vendor's own definitions, deepens that lock-in. TechCrunch calls it a kind of forward-deployed-engineering lite, and that's about right.

One more caution: employee surveys plus usage analytics is a lot of visibility into how individual people work. Anthropic's no-training-by-default pledge is welcome, but the bigger question is how managers use these reports. A tool meant to find broken workflows can just as easily become a productivity scoreboard. We'd want to see companies set those ground rules before they turn the reports on.

Jellyfish's data, drawn from 276,000 of its customers, found Claude Code and other tools built on Anthropic models were behind nearly a third of all proposed code changes submitted for review. OpenAI's Codex has gained ground since its Astra model arrived this summer, but Jellyfish says that hasn't slowed Anthropic's adoption numbers so far.