Claude finally shows where the AI money goes (and how to cap it)

Plenty of companies adopted artificial intelligence without ever really knowing what it costs them. You turn on a subscription, a few employees use it, then the invoice climbs with no clue why or who is consuming what. On July 2, 2026, Anthropic, the company behind the Claude assistant, published an update that goes straight at that black hole.

The news is not about a faster or smarter model. It covers something far more down-to-earth for a manager: seeing where the money goes, setting limits and getting a warning before the bill runs away. For a Quebec business starting to fold AI into daily operations, that kind of tool matters as much as performance.

1. What did Anthropic announce on July 2?

The update targets the Claude Enterprise plan, the one organizations use to give their teams access to AI. Anthropic adds three main capabilities: a detailed analytics dashboard, spending alerts, and the ability to choose which model starts by default. Together they answer a recurring complaint from managers, who received a lump-sum invoice with no way to break it down.

In practice, an administrator can now see usage and cost broken out by group and by person. The dashboard also shows what was produced: documents created, files edited, connectors and skills used, all sitting next to their cost. According to cost analysis firm Finout, the accompanying programmable interface even returns detail by named person, with their consumption and their spend in dollars. You go from a vague estimate to a sharp picture.

2. Seeing where the AI money actually goes

Visibility is the heart of this update. Before, an invoice increase looked like a black box. Now a manager can answer specific questions: which teams use AI most, on what kinds of tasks, and at what cost. The dashboard filters against the real structure of the company, so the numbers follow your existing teams rather than an abstract split.

Anthropic also added a section for developers using Claude Code, its programming tool. It shows the number of active people, the most frequent commands and an estimate of the productivity gain. A natural language analysis feature even lets you ask a question about usage and get an exportable chart, useful for presenting numbers to management without building a report by hand.

• Usage and cost by team and by person, filtered against your existing groups.
• Detail of what was produced (documents, files, connectors) next to the cost.
• A programmable interface that plugs into spend-tracking tools such as Datadog or CloudZero.

This visibility logic matches what we have preached in IT management for a long time: you only control what you measure. It is the same principle behind our managed IT services, where tracking real consumption across a fleet prevents month-end surprises.

3. Alerts before you hit the wall

The second addition speaks loudest to a small business manager: spending alerts. When an organization approaches its cap, the administrator receives a warning at 75 percent and again at 90 percent of the limit. That leaves time to react, adjust the budget or raise the limit before someone gets blocked in the middle of a task.

On the employee side, a notification appears in the application at 75 percent and 95 percent, with the option to request a higher limit directly, without leaving the tool. The manager keeps the final word, but the request arrives cleanly instead of through a panicked email. For larger structures, an administration interface can automate the handling of those requests and flag usage that is climbing fast.

This kind of guardrail changes the relationship with AI. You move from a budget you discover after the fact to a budget you steer in real time. That is exactly the type of control we put in place when helping a company get its new digital tools under management. If you are wondering how to frame AI at your shop, we can talk it through on our contact page.

4. Choosing the right model at the right price

The third piece is subtler and very profitable. Anthropic now gives administrators control over which models are available. You can set which Claude model starts by default for a new conversation, in chat and in the work tools. You can also decide which models are accessible to which roles.

Why does that matter? Because models do not all cost the same. Running every routine task on the most powerful model is like taking a truck to pick up milk. By steering ordinary work to a lighter model and reserving the top-end model for complex tasks, a company can cut its bill without losing quality where it counts. It is a setting you configure once that pays every month.

• An adjustable default model so nothing automatically runs on the priciest option.
• Model access adjustable by role or for the whole organization.
• Individual analytics on by default starting July 11, 2026, a detail to know on the privacy side.

5. What should your business take from this?

You do not need an Enterprise plan to learn something from this announcement. The underlying message applies to any digital tool: adoption without measurement always ends up costing, in money and in risk. AI follows the same rule as cloud services or software licences. If nobody watches consumption, it drifts.

For a company in the Mauricie or Trois-Rivières starting to bring AI in, three habits are worth their weight in gold. First, know who uses what. Second, set a clear limit and a warning before it is reached. Third, reserve the costliest tools for the cases that justify them. These are the same principles we apply in fleet management, and they now apply to AI.

Frequently asked questions

Are these new Claude features free?

They are part of Anthropic’s Claude Enterprise plan, aimed at organizations. The new dashboards, spending alerts and model controls arrive at no extra cost for customers on that plan. Individual plans do not offer the same level of administrative control.

How do you find out which employee uses the most AI?

The administration dashboard shows usage and cost per person and per team, filtered against your existing groups. Note that individual analytics become active by default on July 11, 2026, which is worth checking against your privacy policy before you rely on it.

Can you really cap AI spending?

Yes. An administrator gets an alert at 75 percent and 90 percent of the set cap, and employees are warned in the application before being blocked. You can also choose which models are accessible, which keeps everyday work off the most expensive option.

Governing AI at your company without losing control

AI is becoming a line item like any other, and it deserves the same seriousness as a fleet of workstations or a suite of licences. The new Claude tools show that even the large vendors take the cost question seriously. At OKTO Solutions, we help businesses in Trois-Rivières and the Mauricie adopt these technologies with clear guardrails. See how we can help on our IT services page, or write to us directly through our contact page to frame your AI usage.