Anthropic is moving to prove the return on investment of its AI coding products with a new set of reporting and budgeting features for enterprise customers, according to TechCrunch. The announcement arrives just after financial figures shared with investors portrayed Claude Code as a clear commercial success: the company posted an operating profit in the second quarter and reported net revenue retention of 900%, meaning customers who were spending in Q3 2025 were spending roughly nine times as much by Q3 2026.

What the new tools do

The reporting suite combines internal analytics with employee surveys. Managers can select a date range and see what a team worked on, what it delivered, how much token spend it consumed, and where workflows ran into friction. The reports also include recommendations for improving results. Anthropic says that, by default, it will not use this data to train its models. Alongside the reports, the company is adding pooled group budgets so teams can manage token consumption collectively instead of in isolated allotments.

The features respond to a growing concern among large buyers: it is hard to tell whether AI spending is paying off. Usage among venture-backed companies has pushed token spending sharply higher, but most Fortune 500 firms have yet to see major productivity gains from deploying AI. Giving managers visibility into outcomes and costs is an attempt to defend budgets as scrutiny rises.

Market data from Jellyfish

Jellyfish, which sells tooling to software engineering organizations, published a report based on agentic coding usage across 276,000 of its customers. Claude Code and other tools built on Anthropic models remain the most widely used, responsible for close to a third of all code changes proposed for review. Spending on those tools has grown tenfold since the start of the year and now represents roughly 2% of total engineering cost.

Jellyfish chief scientist Nicholas Arcolano noted that while 2% may sound modest, the pace of growth was something most CFOs had not budgeted for. He argued that organizations have reached a stage where simply spending more is insufficient and they must optimize usage to get value.

Competition from OpenAI

OpenAI's Astra model, released this summer, looked like a potential threat to Anthropic's position. Jellyfish's Nikolas Albarran said Codex took a long time to register meaningfully in the company's data and has risen since Astra arrived. Social media chatter in the tech industry suggests Codex has overtaken rivals, but he said the reality in enterprise accounts is different. Anthropic had strong enterprise sales motions, got in early, and built a reputation for tools that simply work. Many customers invested effort in standardizing workflows around Claude, and there is not yet a compelling reason to move.

Switching costs and the forward-deployed model

Albarran also pointed to the way big companies differ from individual developers or small teams: once a frontier lab is embedded, leaving is harder. All of the major AI labs are building forward-deployed engineering groups that place consultants inside customer organizations to speed adoption. Anthropic's new reports can be viewed as a lightweight version of that support delivered through software. Even if the leading agentic tools are broadly similar, each has quirks, such as how it handles agent configuration files, so migrating carries a real cost that grows as teams invest in setup.

Analysis

The move reflects a broader shift in the AI coding market from raw capability competition to retention, governance and cost justification. For Anthropic, tying usage analytics to concrete recommendations deepens lock-in while addressing the budget-holder's central question. It also suggests that model quality alone may not decide enterprise share; workflow standardization, observability and support are becoming differentiators. OpenAI's Codex is gaining but faces entrenched incumbency, and the rapid cost growth flagged by Jellyfish implies that cost-management features will become table stakes across vendors.