Anthropic on October 28, 2026 launched a set of reporting tools that show enterprise customers what their token spending actually produces, including what a team built, what it cost and where its workflows broke down. The company also introduced pooled group budgets so managers can control spending across teams. The launch targets the question large buyers ask most often about AI tools: whether the money is worth it.
What the tools do
According to TechCrunch, the reports combine internal analytics with employee surveys. A manager can select a team and a date range and see its work: the workstreams, what each produced, the cost, the most common points of friction, how complex autonomous tasks turned out and what to change. The reports include recommendations for improving results. Anthropic says it will not use this data for model training by default.
The pooled group budgets answer a related problem. When individual developers each hold their own limits, spending is hard to plan; shared budgets let a team lead allocate and cap tokens for a whole group.
The spending context
The launch follows a Jellyfish report released on October 27, based on agentic coding tools used across 276,000 of the firm’s customers. It found that Claude Code and other tools built on Anthropic models were the most widely used and accounted for nearly a third of all proposed code changes submitted for review. Spending on such tools has grown tenfold since the start of 2026 and now equals about 2% of total engineering cost. Jellyfish chief scientist Nicholas Arcolano called that “growth that most CFOs didn’t plan for.”
TechCrunch also reports that Anthropic told investors it recorded an operating profit in its second quarter and that net revenue retention was 900%, meaning the same customers spent nine times more in Q3 2026 than in Q3 2025. These figures come from the company’s investor communications and fit the wider picture in our coverage of its IPO filing.
Who is buying, and what keeps them
Venture-backed firms drive most token spending, TechCrunch notes, while most Fortune 500 companies are not yet seeing major gains from AI. That gap is the opening for tools that prove or improve return on investment. Nikolas Albarran, Jellyfish’s AI product lead, described the new reports as a lighter version of the forward-deployed engineering teams AI labs embed with customers. He also argued that switching costs are substantial, because customers have standardized workflows around a tool and products differ in small ways, such as how they handle agent.md files.
OpenAI released its Astra model this summer and has courted Anthropic customers with discounts, as we reported in OpenAI’s up to 90 percent discounts. Albarran said Astra has not slowed Anthropic’s adoption numbers so far, crediting what he called Anthropic’s enterprise sales motions and customer investment in its workflows.
What buyers should ask
For companies weighing the reports, the practical questions are about method: how outcomes are measured, whether survey answers are voluntary, who within the company can see individual-level data and how long it is kept. Anthropic’s statement on training covers model training by default, not internal access or retention, so those points need answers in contracts.
What to watch next
- Whether other model providers ship comparable cost-and-outcome reporting for enterprise accounts.
- Further Jellyfish or similar data on how spending shifts between vendors.
- Anthropic’s IPO process and the disclosures that come with it.
- Customer feedback on how reliable the recommendations are.
Key takeaways
- Anthropic’s new reports show what teams produced, what it cost and where work stalled.
- Pooled group budgets let managers cap and allocate tokens across teams.
- Jellyfish data shows agentic coding spend up 10x since January and about 2% of engineering cost.
- Anthropic told investors of a Q2 operating profit and 900% net revenue retention.
- Vendors are competing on proof of ROI, not only on model quality.
Frequently asked questions
What did Anthropic launch on October 28?
A suite of enterprise reporting tools covering team output, cost, friction points and recommendations, plus pooled group budgets for token spending.
Will Anthropic train on this reporting data?
Anthropic says it will not use the data for model training by default. Companies should check contract terms for retention and access.
How fast is spending on AI coding tools growing?
Jellyfish reports tenfold growth since the beginning of 2026, reaching about 2% of total engineering cost across its customers.
