The key facts
- Anthropic launched reports for enterprise customers: what a team did, how much it cost, and where processes broke down; the data is not used for model training by default
- The company recorded an operating profit in the second quarter and 900% net revenue retention
- According to a Jellyfish report, spending on Claude Code and similar tools grew tenfold since the beginning of the year and accounts for about 2% of engineering costs
What Anthropic introduced
On Wednesday, Anthropic announced the launch of a set of reporting tools for enterprise customers, designed to show what value they are getting for their token spending. Using internal analytics and employee surveys, managers will be able to see their team's work over a selected period: what it produced, how much it cost, and at which stages processes broke down. The reports will also include recommendations for improving results. Anthropic says it will not use this data to train models by default. Additionally, the company is introducing pooled budgets so teams can jointly manage token spending — as stated in its own announcement.
Financial context
The launch follows figures Anthropic presented to investors this week: the company recorded an operating profit in the second quarter, and net revenue retention reached 900% — the same customers' spending in the third quarter of 2026 grew ninefold compared with the third quarter of 2025. As TechCrunch notes, the company has deliberately worked on retaining enterprise users, and the new reports are an attempt to clearly demonstrate the return on investment from its tools and suggest how to increase it. This addresses the main question of large customers: according to the outlet, most Fortune 500 companies do not yet see significant returns from AI adoption, although it is venture companies that drive the bulk of token spending.
Market data: spending grew tenfold
Additional context comes from a report by Jellyfish, released the day before based on the use of agentic coding tools by 276,000 of its customers. Claude Code and other tools built on Anthropic models remain the most widespread: they account for nearly a third of all proposed code changes sent for review. Spending on such tools has grown tenfold since the beginning of the year and now makes up about 2% of total engineering costs. «Even if you say that 2% is still small, this tenfold growth is growth that most CFOs did not plan for,» Jellyfish's chief scientist Nicolas Arcolano told TechCrunch. According to him, the market has reached the point where simply spending a lot is no longer enough — spending needs to be optimized.
Why customers are not switching to competitors
The new reports can also be seen as part of a retention strategy. Jellyfish representative Nicolas Albaran noted that, despite the buzz around Codex and OpenAI's summer release of the Astra model, the data show no slowdown in Anthropic's growth: the company «was simply there with strong enterprise processes,» customers standardized their workflows around its tools, and there are no compelling reasons to switch. He also pointed to the investments all major AI labs are making in forward-deployed engineering organizations — teams of consultants embedded in client companies; in this sense, Anthropic's new reports resemble a lightweight version of such support. In his assessment, the market has matured to the point where, after the effort invested, switching to another provider becomes increasingly costly: the tools are «mostly the same,» but each has its own quirks, and switching costs are not zero.
