AI Spending Is Forcing CIOs to Put Usage Rules Around Everyday Tools
Technology leaders who once pushed broad AI adoption are now adding budgets, model guidance, and usage caps. The shift is less about slowing AI down than making sure trained teams use the right tools for the right work.
For employers, the next phase of workplace AI is becoming less about access and more about management. After several years of encouraging employees to try chatbots, coding assistants, and AI agents, technology chiefs are now confronting a more prosaic question: who pays when usage scales across the company?
At Samsara, chief information officer Stephen Franchetti has approved a broad set of tools for the company’s 4,100 employees, including Anthropic’s Claude, Google’s Gemini, OpenAI’s ChatGPT, and the AI coding agent Cursor. The technology company has also built an internal system that tracks AI costs every day where spending can be measured.
That openness now comes with boundaries. Samsara has introduced usage caps for some non-technical employees, while teams such as research and development get more flexibility because their work involves heavier coding and data analysis.
“It took us a while to settle on the right caps, to make sure everyone was well served,” Franchetti told Fortune. “But it puts people in the position where they’re kind of in control and they can make choices as to which models they use.”
AI adoption now has a meter attached
The wider market explains why executives are paying closer attention. Global AI spending is projected to reach $2.5 trillion this year, up 44% from the previous year, according to figures cited by Fortune. For CIOs and CTOs, that growth is turning AI from an innovation budget item into an operating discipline.
Some companies have found that 2026 AI budgets have exceeded expectations without delivering matching business value. Large AI providers have heard the concern and responded with lower-cost models or price reductions.
Gartner analyst Will Sommer described the moment bluntly. “2026 is the year of everyone finding out that AI is actually really hard,” he said. “It’s not a free lunch. It requires a lot of thought and effort to get right.”
Sommer also warned that businesses can spend thousands of dollars per employee on AI tools that produce poor work and fail to improve productivity. In June, Gartner issued a separate warning that AI coding costs could exceed the average developer’s salary by 2028, driven by rising token consumption and pricing models based on use.
For non-specialists, tokens are the small units of text an AI system processes when it reads prompts, reviews documents, writes code, or generates answers. The more context a model is asked to consume, the more expensive the task can become.
Smaller models, tighter context
Docusign chief technology officer Sagnik Nandy said the company has taken these lessons seriously internally and is also applying them externally. Within Docusign, every engineer has adopted AI tools, and 75% of the code they develop is initiated by AI, he said.
But the company found a cost problem in how coding agents gathered information. The tools were initially set up to draw from Docusign’s full code base before completing a task. “That’s a lot of tokens, because you’re trying to read everything,” Nandy said.
Docusign changed the default behavior so the agents pull only the context relevant to a specific developer task. Nandy said that reduced token use by almost 50%.
Yum Brands, the operator of KFC and Taco Bell, is watching the same trend. Chief digital and technology officer Jim Dausch said token usage is not yet “a material number,” but the direction of travel has caught the company’s attention. Earlier this year, Yum saw both AI token usage and related expenses rise.
Dausch said most AI work may not require the most powerful model available. In his view, perhaps as much as 95% of tasks can be handled by simpler, cheaper models. Yum is therefore training employees on model selection and encouraging business leaders to manage digital spending with the same care they apply to headcount budgets.
“We’re trying to kind of democratize where the costs live and how they’re managed, so it isn’t just an IT line item,” Dausch said.
Cigna Group is taking a portfolio approach. The healthcare company has approved more than 70 AI models for internal use, including small language models and older, cheaper versions that can handle tasks with less demanding reasoning needs. Katya Andresen, Cigna’s chief data, digital, and AI officer, said the expensive mistake is using the priciest systems without limits.
“The way you really run up costs is you use the most expensive models with no guardrails around them,” Andresen said. Although Cigna’s compute and token usage has grown, total spending has not risen at the same pace because the company routes work across different model types.
What managers should take from the shift
The reported facts point to a practical management lesson: AI governance is moving closer to the team level. The issue is not whether employees should use AI, but whether they know which tool fits the task, what the cost implications are, and when human review is essential.
At Compass, chief technology officer Shay Artzi has focused AI investment on three groups: engineers, the company’s AI Assistant for real estate professionals, and general corporate employees. Engineers were early adopters, but Artzi said the company was careful about token spending from the start and did not require all code to be written with AI. Compass tested several AI coding tools and chose partnerships with Anthropic and Google, with financial controls in place. “We also put budgets for every engineer, so they are aware of how they’re spending,” Artzi said.
For business leaders building human-led AI workflows, the message is clear. Usage caps, model menus, training, and budget ownership are not signs that AI programs are failing. They are signs that AI is becoming part of normal operations, where good judgment, clear accountability, and well-designed workflows matter as much as the software itself.
Reported by Hybrion Insights with reference to Fortune AI.
hybrion
Better together. Built for what's next.
Stay updated
Get insights on AI, teams, and the future of work.
