Financial reports Anthropic shared with investors this week suggest that its AI coding tool, Claude Code, is delivering every bit of business success Frontier Labs had hoped for. The company posted an operating profit in the second quarter and a 900% net revenue retention rate. In other words, spending from the same customers in Q3 2025 increased nine times in Q3 2026.
Anthropic is focused on retaining its enterprise users, and on Wednesday announced a new suite of reporting tools designed to help those customers better understand the value they’re getting from their token spend.
Managers can use internal analytics and employee surveys to see their team’s work over a specific date range, what they created, what it cost, and where workflows broke down. Anthropic says it doesn’t use this data to train models by default. The report also provides recommendations on how to improve your results.
This is an obvious attempt to demonstrate the return on investment of Anthropic tools and show customers how to improve it. This is a major concern among large customers. Anthropic is also creating pooled group budgets to help teams better manage token spending.
Businesses have been searching for that kind of information when figuring out how best to deploy AI tools. While usage among venture-backed companies is driving token spending, most Fortune 500 companies are not realizing significant benefits from AI adoption.
Jellyfish, a company that develops tools for software developers, released a new report yesterday based on the usage of its agent coding tools by 276,000 customers. Claude codes and other tools built on human models remain the most widely used in research, accounting for nearly one-third of all code change proposals submitted for review. More importantly, spending on these tools has increased tenfold since the beginning of the year, and their costs now represent approximately 2% of total engineering costs.
“Even though 2% is still small, 10x growth is growth that most CFOs weren’t planning for,” Nicolas Arcolano, chief scientist at Jellyfish, told TechCrunch. We’re certainly at a point where people are starting to realize that just spending a lot of money isn’t enough. Then you need to optimize it and make it work. ”
A new report from Anthropic talks about that challenge. While there is a humanitarian lead among enterprise customers, it’s hard to pinpoint what makes an agent coding tool truly sticky. The release of OpenAI’s Astra model this summer seemed like a potential threat to Anthropic, but so far Anthropic’s adoption numbers have not slowed down.
“It took a long time for Codex to actually make it into data, but now Astra (OpenAI’s latest model) is here,” Nikolas Albarran, head of AI products at Jellyfish, told TechCrunch. When you look at X, you get the distinct impression that the competition is more heated, he continued. “Everyone in the Valley is saying, ‘Oh, Codex is better,'” Albarran continued. “Anthropic was just there with a great enterprise motion, and they had their own existing weight. It just worked and worked well. And a lot of people just put a lot of effort into standardizing the workflows around it. And there’s no big reason to switch.”
This is not just a model or workflow issue. Business realities mean that once Frontier Labs acquires a customer, it’s harder for them to switch than for an individual programmer or small organization. Albarrán points out that all major AI labs are investing in forward deployment engineering (FDE) organizations, teams that embed AI consultants within customer organizations to help them get off the ground. These new reporting tools can be considered a type of FDE-lite, a lightweight version of Anthropic’s hands-on help for customers.
“We’re at a stage now where it’s mature enough. Once you get in, it’s going to be harder and harder to switch because you’ve put in the effort,” he said. “Even though everything is pretty much the same, everything is slightly different and has some quirks in things like how agents handle ND files, so the cost of switching is non-zero.”
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
