Anthropic Launches ROI Reports, Jellyfish Sees 10x Spending

For large enterprises, switching AI tools becomes challenging once workflows are standardized, according to Jellyfish. Anthropic is launching new reports detailing usage and costs to help its clients manage return on investment, while spending on agentic coding tools has increased tenfold since the beginning of the year in Jellyfish's panel.

Change Costs Recognized by Existing Clients

Nikolas Albarran, AI product manager at Jellyfish, believes that once an AI lab has established itself with a major client, changing tools becomes more difficult than for an individual or a small organization. He notes that large labs invest in forward-deployed engineering teams that integrate consultants within client organizations. According to him, even similar systems have their specificities regarding agent management and ND files, so the cost of switching is not negligible. He argues that once workflows are standardized, there is often no sufficient reason to switch to another tool. He also describes Anthropic as benefiting from strong engagement with businesses and the weight of incumbency, with tools perceived as effective and around which many have standardized their processes.

ROI Management: Usage Reports, Costs, and Recommendations

Anthropic plans to roll out a suite of reporting tools aimed at helping its clients understand the value derived from their token expenditures. The analyses are intended to allow managers to visualize, over a given period, team activity, production achieved, costs incurred, and points of breakdown in workflows. According to the company, this data will not be used by default to train its models. The reports will include recommendations to optimize results. Anthropic also announces the establishment of pooled group budgets to facilitate token spending management by teams.

Corporate Adoption: Spending x10 and 2% Share According to Jellyfish

Jellyfish reports, based on the usage of 276,000 of its clients, that spending on agentic coding tools has increased tenfold since the beginning of the year. These tools now account for about 2% of the total engineering cost in this panel. Claude Code and other tools based on Anthropic's models account for nearly one-third of the code change proposals submitted for review. For Nicholas Arcolano, chief scientist at Jellyfish, a tenfold increase was not anticipated by most CFOs, even though 2% remains a relatively low level. He adds that companies are now looking to optimize their spending to make these tools truly effective.

Competition: Astra's Arrival Has Not Stopped Anthropic's Adoption

The release this summer of OpenAI's Astra model seemed to pose a potential threat to Anthropic, but the adoption figures associated with Anthropic have not shown any slowdown so far. Nikolas Albarran indicates that it took time to see Codex appear in Jellyfish's data and that this signal coincided with Astra. He observes that online discourse suggests an intensification of competition, with some claiming that Codex is superior. However, it remains difficult to pinpoint exactly what makes an agentic coding tool truly attractive.

Value Objective: Companies Seeking Proof and Anthropic's Anchoring

Companies are asking for more elements to decide how to best deploy AI tools, while usage in venture-backed companies fuels token spending. Most Fortune 500 groups do not report significant gains related to AI. Anthropic's new reports can be seen as a lighter form of support for its clients. Meanwhile, Anthropic has argued to investors that Claude Code is commercially viable, that the company achieved an operating profit in the second quarter, and that it boasts a net revenue retention rate of 900%, described as a ninefold increase in spending from the same clients between the third quarter of 2025 and the third quarter of 2026.