Claude Enterprise now shows spend per workstream and pools budgets by team. Your client automations need the same view

AI spend is becoming a line item that finance wants explained, workflow by workflow.

What happened

Anthropic made three features generally available for Claude Enterprise: smart reports, in-app surveys and pooled group budgets.

Smart reports let an admin pick a team, a date range and products. Claude analyzes a sample of the team's transcripts, groups the work into workstreams, and shows the spend for each one, the deliverables produced and common friction points, such as a missing connector or a failed tool call. Each report ends with recommendations, including recurring workflows to turn into skills. Anthropic says the reports are not for evaluating individual performance, and TechCrunch reports it will not use this data for training by default.

In-app surveys ask members questions such as what would break or slow down first if the team lost access tomorrow. Pooled group budgets give a team one shared monthly budget, while individual limits still act as a guardrail. Admins get email alerts at 50%, 75%, 95% and 100%.

TechCrunch cites a Jellyfish report that spending on agentic coding tools has grown 10x since the start of the year and now equals about 2% of total engineering cost. Jellyfish's chief scientist said that growth is something most CFOs did not plan for.

My take

The feature I care about is spend per workstream. That is the question every client eventually asks about their automations too: what does the lead enrichment flow cost us a month, and what did it produce?

Most automation stacks cannot answer that. Tokens get billed to one API key, runs are logged in three tools, and nobody joins the two.

What I now build into client systems from day one:

• One API key or project per workflow, so the bill splits itself.
• A log row per run with tokens, cost and the outcome it produced, such as a qualified lead or a closed ticket.
• A weekly summary that ranks workflows by cost per useful outcome.
• Friction tracking: failed tool calls and manual fallbacks counted, not just errors.

A shared budget with per-run guardrails is also a good pattern for agent workflows. It stops one runaway loop without starving everything else.