Anthropic adds reports that show companies what their Claude spending produces

Anthropic announced a set of reporting tools on Wednesday, October 28. They are meant to show business customers what they get for the money they spend on its models. Managers can pick a date range and see what a team produced, how much it cost and where its workflows broke down. The reports also suggest ways to improve the results. Anthropic is also adding pooled group budgets so teams can manage token spending together, TechCrunch reports.

The timing matters. Companies are spending far more on AI coding tools than they did at the start of the year, and many cannot yet say what that money buys. Anthropic is also preparing for a planned IPO, and its pitch to investors depends on enterprise customers staying and spending more.

What the reports contain

The reports draw on two kinds of input: Anthropic's internal usage analytics and surveys of employees. From these, a manager gets a view of one team's work over a chosen period. It covers output, cost and the points where work stalled. Anthropic says it will not use this data to train its models by default.

The pooled budgets are a separate feature. They let a group share one allowance of token spending instead of managing it person by person or key by key.

TechCrunch describes the reports as a clear attempt to show a return on investment, which large customers care about a great deal. They can also be read as a lighter version of forward-deployed engineering. That is the practice, now common at the major AI labs, of placing consultants inside customer companies to help them adopt the tools. The reports offer some of that guidance without sending people.

The numbers behind it

According to TechCrunch, the financials Anthropic shared with investors this week show an operating profit in the second quarter. They also show 900% net revenue retention: the same customers spent nine times as much in the third quarter of 2026 as in the third quarter of 2025. TechCrunch credits much of this to Claude Code, Anthropic's coding agent.

A separate view comes from Jellyfish, which sells analytics tools for software teams. Its October AI Engineering Trends report came out on October 27. R&D World reports that it covers more than 1,300 companies, more than 276,000 developers and more than 100 million pull requests, with data through September. According to TechCrunch, tools built on Anthropic models, Claude Code among them, were the most widely used in the data. They produced nearly a third of all proposed code changes submitted for review.

TechCrunch reports that spending on these tools has grown tenfold since the start of the year and now equals about 2% of total engineering cost. R&D World puts the median Claude Code cost at $63 per active developer per week by late September, or $252 per month. At the 90th percentile the monthly figure was $1,666. At the median company, AI agents opened or committed code in 27.7% of pull requests in September, up from about 2% in January.

"Even if you're saying 2% is still small, that 10x growth is growth that most CFOs didn't plan for," Nicholas Arcolano, chief scientist at Jellyfish, told TechCrunch. He added that companies are realising it is not enough to spend: "You now have to optimize it, make it work."

TechCrunch also notes a gap. Venture-backed firms are driving much of the token spending, but most of the Fortune 500 are not yet seeing major gains from AI.

Why customers stay

OpenAI released a model called Astra this summer. TechCrunch reports that so far it has not slowed Anthropic's adoption. Nikolas Albarran, AI product lead at Jellyfish, said OpenAI's Codex took a long time to show up in Jellyfish's data and has become more visible since Astra. He described a gap between online opinion and what companies actually do. Anthropic, he said, had strong enterprise sales and an incumbent position. Its tools worked, and many teams had built standard workflows around them, so there was "not a big enough reason to switch."

Albarran argued that switching costs grow as adoption matures. Even if the tools are "mostly the same," he said, each has its own quirks, such as how it handles agent.md files, "and so that cost to switch is not zero." Reports that teach a team how to get more out of one vendor's tools add to that cost.

What it means for work

For engineering managers and finance teams, AI coding spend is becoming a line item that has to be justified like any other. These reports take over part of that work by tying token costs to team output and pointing to where workflows fail. For developers, the effect is more visibility. Their AI-assisted work will be measured by date range, cost and breakdowns, and surveys will ask how it went. The figures come from the vendor whose spending they measure, so companies will still need to decide how far to trust them as evidence of value. The harder question is also still open: whether the gains that venture-backed firms report will show up at the large companies that have not seen them yet.