In 2026, AI and agents have transformed from being "novelities" in the eyes of businesses to becoming priorities. A survey on agentic and transformation released by Deloitte ( Deloitte ) in August showed that 74% of business leaders expect that nearly half of their business processes will be redesigned around AI and agents over the next four years; however, only 16% of the respondents indicated that they are already prepared for this.

As part of the 2026 AIQ Top 75 list, Fortune is looking for companies that have deeply deployed AI agents and are engaged in large-scale automated workflows. This year's survey includes several new questions that focus specifically on the internal use of agentic AI within enterprises.

Among the companies that rank high in agentic AI, Adobe, Kyndryl, and Salesforce have shared their practical details to date. They have widely applied agentic AI in various scenarios ranging from invoicing, product research, software engineering to employee and customer service. These companies have all experienced failures, adjustments, and significant progress, and they have all come to the same conclusion: it is essential to focus on a specific business problem; to involve both technology and business stakeholders; and to always maintain a focus on ROI.

Less, but more focused agents

The Chief Information Officer Kim Basile stated that she is not aware of how many workstreams within the company currently involve agentic AI.

To be honest, in the first year we were constantly counting agents, but that didn’t give me any real insight into the situation. So we realized that we needed to change our approach and truly start thinking about the flow of business processes, she said.

Salesforce has also made similar adjustments. The company sells the agentic and AI systems it uses internally as products as well. Andy White, the senior vice president responsible for the internal implementation of Salesforce products, stated that the company has shifted from initially using hundreds of agents to focusing on a few that it is convinced can bring tangible benefits. He doesn't know exactly how many agentic workflows have been implemented yet, but he knows that this is happening in every department.

"At first, we chased after too many things, and as a result, we couldn't concentrate enough," he said.

International mining companies like Freeport-McMoRan don't have much room for trial and error. The industry in which these companies operate is highly capital-intensive, and the back-end workflows are often related to the operation of equipment worth millions of dollars. Therefore, from day one, results are taken very seriously. In order to narrow down the scope and identify the processes that can generate the greatest benefits, management focuses on those workflows that are both high-frequency, labor-intensive, and rely on agentic and AI for data processing.

Freeport-McMoRan is currently being deployed in about 10 fields. agents. The Senior Vice President and Chief Information Officer, Bert Odinet, stated that the most prominent use case is in sales invoice processing. The company's invoice pricing varies depending on the by-products contained in each batch of materials, making the process particularly complex and data-intensive. Odinet mentioned that after using agentic to process approximately $500 million in invoices over a period of six months for just one product line, the company has already achieved significant cost savings and plans to expand this to other product lines as well.

What exactly does measuring ROI mean?

Odinet indicates that Freeport-McMoRan sets specific KPI for each agentic process to ensure that they indeed achieve the desired results, and to prevent us from deluding ourselves into thinking that these agents have been successful.

Adobe, Vice President of Digital Employee Experience, and AI, Co-Chairperson, also emphasized that it is necessary to measure whether the expected results have actually occurred. Moreover, this measurement should not be limited to just productivity or efficiency.

“Does it help you close deals faster? Does it help you win customers because you understand them better?” she asked, adding that the real question is, “So what after saving time?”

For example, Adobe redesigned the internal help desk using agentic and AI, and this help desk serves over 30,000 employees. The company is not only concerned with whether agentic and AI can close more tickets more quickly, but also focuses on the quality of responses—whether the answers are accurate, whether employees are satisfied—and whether this process frees up capacity for the support team.

This is also an important point of growth for Kyndryl. Basile mentioned that initially, the company understood agentic and ROI as cost-cutting measures. In the first year, the focus was on employee productivity, while in the second year, attention was turned to departmental efficiency.

"That's not the correct answer," she said. "What I mean is, of course, this is definitely an important part of ROI, but what we really need to do is to shift our mindset and think: 'How can I bring more value to the company? How can I help the finance department shorten the closing time from five days to three days, and then use the extra two days to prepare for investors?' "

In the third year, Kyndryl now pays more attention to the impact of these workflows themselves, especially how they permeate the entire organization and connect with each other.

To reshape, collaboration between both parties is essential.

When Salesforce noticed that some agents achieved the expected results, while others did not, the company's management began to reflect on the reasons. Apart from those agents that were not linked to solving specific problems, they found that workflows that gained more support from business stakeholders were significantly more likely to succeed.

To solve this issue, the company is now pairing the agentic team members from the IT team with business stakeholders to form small "rapid innovation teams," and each business unit has one such team set up.

White said, "You must cooperate with a stakeholder who is independent-minded and willing to disrupt the status quo of their own organization, because that will inevitably bring about changes."

For example, a fast innovation team focused on the legal department attempted to make the company's legal case allocation process more efficient—that is, to assign new cases to the most suitable and experienced lawyers for handling. White indicates that the new agentic process increased the proportion of cases that were correctly assigned and resolved from the first attempt from 12% to 34%, saving the company over $350,000 in one year.

Among these companies, all the leaders interviewed by Fortune emphasized that more than ever is it necessary to bridge the gap between IT and the business departments. White said that it's not as simple as deploying a agent and then calling it a day; it requires a lot of continuous feedback, supervision, and constant improvement, and a large part of this responsibility falls on the business teams.

Vanwinkle also signifies: 'This is what we expect from leaders. 'This is an area where we need your full commitment; you are the ones who will help us to find the right direction.'

To a large extent, this is because, as these leaders unanimously agree, implementing agentic and AI requires reimagining the workflow from scratch. Simply copying the existing processes and automating them may actually have the opposite effect.

Basile says that one of the biggest mistakes Kyndryl made in advancing the transformation of agentic AI was to devote more effort to automating a process that had not been fully verified in the first place.

“It was already bulky to begin with, but we just wanted to automate it, and as a result, we didn’t solve any problems; instead, we made it even more complicated,” she said.

This is also why Odinet emphasizes the need to obtain new perspectives from people who are not familiar with these processes, such as those from other business areas. Functional stakeholders are accustomed to doing things in a set way, and consciously taking a step back and asking "why" requires deliberate effort.

"You need smart thinkers," he said, "but you also hope that they can be open-minded and rethink the entire process."

This report is part of the Fortune series titled “Takeaways from the Fortune AIQ 75”, which consists of five parts. It explores how companies, as revealed by the Fortune rankings for 2026, utilize AI – ranging from agentic workflows and industry-specific applications to the operational models behind their large-scale adoption.