This piece is by Katie Barnett, Director of Cyber Security at Toro Solutions
For most business technology, the cost is relatively easy to understand. You buy a licence, agree a contract or pay for a certain level of usage, and you have a reasonable idea of what you will spend over the year. AI is making that much harder.
As businesses move beyond individual AI subscriptions and start using AI across more of their operations, costs can vary considerably depending on which models are being used, how often they’re being used and what they’re being asked to do.
AI agents make this harder again. Instead of someone making a single request and receiving an answer, an agent can work through a task independently, making multiple requests, analysing information and interacting with other systems along the way. Each action consumes resources and can add to the cost.
The individual costs can look fairly insignificant, but across hundreds of employees, different platforms and automated processes running throughout the day, they can add up quickly.
Uber is a good example of just how quickly this can happen. The company reportedly used its entire annual AI budget within the first four months of 2026 as the use of AI coding tools grew across its engineering teams. It subsequently introduced monthly spending limits for some agentic coding tools, alongside dashboards allowing employees to track their own usage.
The issue wasn’t that the technology had failed, but that usage had grown much faster than the original budget allowed for.
From fixed licences to variable consumption
Businesses have dealt with variable technology costs before, particularly with cloud computing. AI takes this further because the eventual cost can be heavily influenced by the way people actually use it.
Two employees with access to the same platform could generate very different costs. One might occasionally use it to summarise a document, while another uses it throughout the day to analyse large amounts of information or carry out more complex tasks.
The model being used matters too. The most powerful model might make sense for a difficult piece of analysis, but not for a straightforward task. If the most expensive option becomes the default, the difference can quickly add up.
Then there is the number of tools businesses are starting to accumulate. Different teams may be paying for different platforms, while AI functionality is increasingly included in software the organisation already uses. The overall cost can grow without anyone making a deliberate decision to spend more.
Buying the technology is only part of the budgeting decision. How it’s used afterwards can have just as much influence over what the business eventually spends.
More use doesn’t always mean more value
Businesses understandably want employees to use the technology they have invested in, but higher AI usage does not automatically mean the organisation is getting more value from it.
If someone can complete an hour-long task in ten minutes, the benefit is fairly easy to see. If an agent removes a repetitive process that previously took several people to complete, its running cost may be easy to justify.
The calculation becomes harder when businesses don’t know what all that consumption is producing. The cost of the AI itself is only one part of that. If an employee saves an hour using AI but then spends much of that time checking or correcting the output, the productivity gain starts to look different.
Are expensive models being used for simple tasks? Are departments paying for tools that do broadly the same thing? Are automated processes consuming resources without producing enough useful output?
AI is still developing quickly, so businesses need to give people room to experiment. But as usage grows, they also need to understand where the money is going and what they’re getting back.
Cost can expose a wider problem
Unexpected AI spending can also tell you something about how much visibility an organisation has over its AI use.
If nobody can easily explain why consumption has increased, there are some fairly basic questions to answer. Which tools are being used? Who is using them? What have they been connected to? Have tools introduced for one purpose gradually started being used for something else?
The same visibility needed to understand AI spending is needed to understand what information AI can access, what employees are sharing with it and what permissions have been given to agents and automated systems.
An agent making hundreds of unnecessary requests may be wasting money, but you also need to understand why it’s doing it, what it has access to and whether its behaviour is being monitored.
Start with what you have
Controlling AI costs doesn’t have to mean restricting every tool or putting barriers in the way of employees who are finding useful ways to work with AI. A better starting point is understanding what you already have.
Which AI tools are being used across the organisation? What are they costing? Who is using them and for what? Where are different teams paying for similar capabilities? Which applications are consuming the most resources and are they delivering enough value to justify it?
Once you have that picture, it’s much easier to decide where spending needs to be controlled and where investment makes sense. It also gives security teams a clearer understanding of where AI is being used and where closer attention may be needed.
As AI use grows, businesses are going to need a much clearer view of what they’re spending and what they’re getting in return. That means looking beyond licences and subscriptions to understand how different tools and models are actually being used across the organisation.
Without that visibility, controlling the cost becomes difficult and so does understanding the wider risk.
Be the first to comment