Agent Upstart

Cool things people are building with agents — and practical help getting the most out of them.

Playbooks · Teams

Claude Code pools budgets and reports costs via API

Written November 13, 2026

Pooled group budgets, layered managed policies, a cost-report admin API, and deny-instead-of-prompt permissions: the admin batch.

What changed

On October 28, Anthropic made three Claude Enterprise admin features generally available: smart reports, in-app surveys, and pooled group budgets. For a platform lead, the budget piece is the one that changes day-to-day governing. Until now, spend control meant a limit per person. A pooled budget gives a whole group one shared monthly amount, so a heavy user in a busy week draws from the same pool as a light user instead of stopping at a personal ceiling while a colleague's allowance sits unused.

The reporting side moved with it. Admins now have a spend report in the organization analytics, a programmatic route to per-user cost numbers on Enterprise plans, and smart reports that group sampled work into workstreams with the spend behind each one. Taken together, the batch answers the two questions a platform team gets asked: how do we cap this, and how do we show what it bought.

How a pooled budget works

Anthropic's pooled budget guide describes a pool as one shared monthly amount for a group, added on top of each member's own monthly spend limit. The two limits run together rather than replacing each other:

Pools are set in Organization settings, on the By group/tier tab of the spend limits table. Anthropic's starting-point advice is plain: base the first amounts on last month's real spend, with a pool near three times the group's monthly spend and a member limit near three times the top spender's. If a pool is 75% used before the 20th of the month, the guide suggests raising it by half rather than letting the team stall.

Reporting the cost

Three layers of reporting sit behind the budgets. The Claude Code cost documentation maps them by plan. The spend report in org analytics shows estimated spend per user and per model, exports to CSV, and refreshes daily; it covers usage-credit spend and appears once usage credits are on. On the Enterprise plan, the Enterprise Analytics API returns per-user usage and cost reports across Claude surfaces, Claude Code included, and a Primary Owner creates the key with the read:analytics scope. Teams-plan admins use the CSV export instead, which lists token usage and estimated spend per user and per model.

Smart reports are the narrative layer on top of those numbers. An admin picks a team, a date range, and the products to include, and Claude groups a sample of the team's transcripts into workstreams, showing the spend for each one, the deliverables produced, and the common points of friction. Anthropic is explicit about the boundary: the reports are for understanding adoption and planning investment, not for judging individual performance or making employment decisions.

The policy layer underneath

Budgets cap money; managed settings cap behaviour. Claude Code's cost page shows the pattern in miniature: an administrator can set a modelPricing table in the organization's managed settings so the cost figures developers see in /usage use contracted rates instead of list prices. Spend limits themselves layer at three levels, organization, group, and individual member, and the seat allowance remains the default ceiling underneath them all.

The permission side of the same layer is the deny rule. A permission setting that denies a tool outright, instead of prompting whoever is present, is what keeps an unattended run from stalling or wandering. The client behaviour for that setting is covered in Claude Code 2.1.319: unattended runs stop stalling; this page stays on the admin side of the line.

Limits

Related playbooks

Sources

Back to: Teams playbooks hub · Also: Claude Code 2.1.319: unattended runs stop stalling · Claude Code 2.1.323: Code Mode writes the tool calls