Spending on agentic coding tools has grown roughly 10x since the start of the year and now equals about 2% of total engineering cost, according to figures from engineering analytics firm Jellyfish reported by TechCrunch. Once a cost reaches that size, someone in finance asks what it bought. On 28 October 2026, Anthropic released three tools for Claude Enterprise that are meant to answer exactly that, and it says all three are now generally available.
Smart reports read the work
Smart reports start with three choices. An admin picks a team, a date range and the Claude products to include. Anthropic says Claude then analyzes a sample of that team's transcripts and groups the work into workstreams. Each report shows the spend for each workstream, what was delivered, and the friction points where people got stuck. It finishes with one consolidated list of recommendations.
Admins who also have billing permission can raise a spend limit straight from the report, which is handy when a report shows a team is running out of budget on work that is clearly paying off.
In-app surveys ask the people
The second tool asks the people doing the work. In-app surveys come with default questions about whether the work was worth it, and admins can customize them. Admins choose who receives a survey, when it runs and where it appears inside Claude. Responses show up on the survey page as they arrive and can be exported to a spreadsheet. Anthropic says survey responses are treated as the customer's data and accessed only as needed to run the service.
The useful part is the combination. A smart report tells you what the transcripts suggest, and a survey tells you what the team thinks. When the two disagree, that gap is usually the most interesting thing in the room.
Pooled budgets cap the bill
The third tool is about control. Pooled group budgets give a group one shared monthly budget on top of any individual limits. Every request counts against both, and a person stops at whichever runs out first. Admins with billing permissions get email alerts at 50%, 75%, 95% and 100% of a budget, so nobody is surprised at the end of the month.
The privacy line worth holding on to
Anthropic's announcement says smart reports are not meant for evaluating individual performance or making employment decisions. TechCrunch also reports that Anthropic will not use this data for model training by default. If your company turns these features on, it is reasonable to ask for that first rule to be written into your internal AI policy, because tools that read transcripts make people nervous for understandable reasons.
Availability depends on configuration. Anthropic notes that organizations with a HIPAA configuration or customer-managed encryption keys may see differences. The announcement names Claude Enterprise only, so Team plan customers should not expect these features yet.
What to do this week
If you run Claude Enterprise, generate one smart report for a single team over the last month and read the friction points before you look at the spend. Then bring these five questions to your next AI budget conversation:
- Which workstreams use the most spend?
- What did each one deliver this month?
- Where do people get stuck most often?
- Do the survey answers match what the report says?
- Which one recommendation do we try first?
Our honest view is that this is the right direction, because AI budgets will only survive if teams can show what the money produced. Measuring workstreams instead of people is the part to protect.
For a wider look at how teams combine AI tools, see Stacks, and the Claude page in Learn covers how the plans fit together. Our earlier piece on Claude's monthly API credits explains the credit side of the bill.