TheChek News Flow

The Takeaway

Executives at Microsoft and other enterprise software firms heralded 2025 as the year artificial intelligence would be capable of automating tasks that involve multiple steps, such as generating dashboards based on company sales data. But as the year comes to a close, Microsoft has lowered expectations for how quickly it can get customers to spend money on these newer products, known as agents.

Multiple Microsoft divisions, for instance, have lowered how much salespeople are supposed to grow their sales of certain AI products after many of them missed sales-growth goals in the fiscal year that ended in June, according to two salespeople in Microsoft’s Azure cloud unit. It’s rare for Microsoft to lower such quotas for specific products, the people said.

The change shows how Microsoft is adjusting to resistance from companies to pay more for AI. Corporate customers have complained over the past year that it’s hard to measure the savings from using the technology for tasks like drafting reports on customer spending and sales leads, and that it can be difficult to get the AI to work perfectly in cases where small mistakes can be costly, such as automating finance and cybersecurity tasks.

That said, AI has been a major boon to Microsoft’s business. That’s largely thanks to new spending by AI firms such as OpenAI, which has projected it would rent about $15 billion worth of cloud servers from Microsoft this year, as well as Microsoft’s sales of AI software such as its 365 Copilot workplace tools and GitHub Copilot coding agent. Microsoft itself and other large tech companies also have gotten productivity boosts from using AI tools internally.

But getting traditional businesses to boost their spending on advanced forms of AI hasn’t been as easy.

For instance, private equity fund Carlyle last year started using Copilot Studio, a Microsoft product that lets companies develop AI to automate tasks like summarizing meetings or drafting financial models based on Excel spreadsheets, without needing to write any code.

But in the months after Carlyle started using the tools, its representatives told Microsoft they were having trouble getting the AI to reliably tap data from other applications such as Salesforce’s customer relationship management app, which was necessary for some of Carlyle’s automations, according to someone with direct knowledge of the situation and a second person briefed on it. This fall, Carlyle reduced the amount it was spending on the tools, the people said.

The change was part of a broader company effort to reduce AI spending and consolidate the AI tools it uses, one of the people said.

A Microsoft spokesperson declined to comment on the changes in sales quotas but pointed to growth in the company’s overall cloud business, which has been lifted by rentals of AI servers by OpenAI and other AI developers.

Adjusting Expectations

Microsoft isn’t the only firm adjusting expectations for revenue from AI agents that automate complex tasks. OpenAI, for instance, recently lowered its projections for AI agent revenue by $26 billion over the next five years compared to earlier numbers.

OpenAI expects to make up for that by growing subscription revenue from ChatGPT, which includes some agents, such as deep research for creating research reports, as well as from new products, which could include selling ads on ChatGPT, according to the projections. CEO Sam Altman said Monday he was putting off work on ads as well as agents related to health and shopping as the company focuses on fixing problems involving ChatGPT.

The challenge of increasing revenue from AI agents has been particularly acute at enterprise software firms such as Salesforce that, unlike Microsoft, don’t have the benefit of owning a large cloud server business.

Salesforce and other firms such as ServiceNow have been offering steep discounts to customers that try new agent products to automate workplace tasks like closing IT tickets or onboarding new employees. Other firms including Amazon Web Services and Anthropic have been pouring resources into helping customers set up AI applications to run properly, similar to the way consulting firms help their clients.

It’s not the first time a large cloud and software providers adjusted expectations regarding how much enterprises will spend on AI, even as sales of chatbots and coding models have surged. Last year, for instance, Google and Amazon tamped down expectations for enterprise AI sales after companies didn’t start paying for new AI tools as quickly as they had hoped.

“Talk to me in three months and I’ll show you ROI”—a return on the investment, said Brian Spanswick, chief information officer at cybersecurity firm Cohesity.

Microsoft is also planning to overhaul AI agent features that can automatically take actions on a customer’s desktop computer, such as moving data between different files, which it plans to sell to buyers of its Windows PC software, according to someone involved in the effort. A previous version of the features, dubbed Recall, was widely panned over privacy concerns, and the company subsequently rolled it back last year.

Microsoft developed a new open-source model, Fara-7B, to handle such tasks in a more privacy-friendly way than the previous version. Although the feature requires Microsoft to take up to three screenshots of the customer’s computer while handling the tasks, it deletes them when the task is completed. The model is also small enough to run entirely on the computer.

Ambitious Quotas

At Microsoft’s Azure cloud unit, multiple sales teams set ambitious goals to sell AI products in the 12 months ending in June this year, but they tempered those goals for the current fiscal year after many salespeople failed to meet their quotas.

For instance, one U.S. Azure sales unit set quotas for salespeople to boost customers’ spending on a product called Foundry, which helps customers develop AI applications, by 50% in the last fiscal year, ending in June.

Less than a fifth of salespeople in that unit met their Foundry sales-growth targets, and in July Microsoft lowered their targets to roughly 25% growth for the current fiscal year compared to the last one, according to a Microsoft salesperson with direct knowledge of the business. The size of the quotas per salesperson for Foundry can range from several tens of thousands of dollars annually to hundreds of thousands, the people said.

In another U.S.-based Azure unit, most salespeople didn’t meet an earlier quota of doubling Foundry sales in the fiscal year that ended in June. Their quotas fell to 50% for the current fiscal year, according to another Microsoft salesperson with direct knowledge.

But there may be sunshine at the end of the AI agent tunnel.

Brian Spanswick, chief information officer at cybersecurity firm Cohesity, said his company has been testing AI agent features from Copilot Studio and Foundry that can write reports about its customers for salespeople, based on internal data Cohesity has on them as well as publicly available information.

While agents have failed to deliver on their promise so far, Cohesity is writing code to better connect them to data sources from other applications with hopes of making them work over the next few months.

“That’s what I’m telling my board,” he said. “Talk to me in three months and I’ll show you ROI”—a return on the investment.