CIOs’ top 10 takeaways from the year AI got practical

Artificial intelligence dominated the CIO agenda in 2025, so it’s no surprise that the technology, its application, and how to get the best business value out of it gave IT leaders much to think about this year.

What might be surprising, however, is that CIOs, when asked to look back on 2025, don’t cite many technical lessons-learned. Rather, they talk about how AI is teaching them about the technology’s potential and its impact on leadership, the organization, staff, and society itself.

Here, eight representative CIOs share their biggest takeaways from the past year.

1. ‘2025 is the year AI became useful’

CIOs have had years of AI experiments and pilots, which studies have found often failed to take flight. But Matthew Richard, CIO of the Laborers’ International Union of North America (LIUNA), sees 2025 as the year that AI — and generative AI more specifically — hit its stride.

“2025 is the year AI became useful,” he declares.

Richard says this past year his IT team found that the enterprise AI tools offered by vendors “have been super interesting and exciting and are helping our end users work in more productive ways.”

He also found that AI tools advanced so quickly in 2025 that his team had been able to launch initiatives that were unthinkable at the start of the year. “Based on how these tools have evolved, we have new ways of getting information and putting it at the fingertips of users without being overly complex,” he says.

Case in point is an AI-based product that his team developed to capture information from new union contracts and then analyze that information against historical and current data to ensure contract terms meet standards and expectations. The product is now in its pilot phase, says Richard, noting that the product uses AI to access a “massive trove of knowledge” that would be practically impossible to fully leverage otherwise.

“It’s a bit surprising how quickly these tools became useful, compared to what I thought of them a little more than a year ago. I thought these tools were too clunky. But now there is a lot of exciting possibilities on the horizon with LLMs and agentic AI and integrating them into the massive data sets that organizations own,” Richard adds.

2. ‘It’s important to keep a fast pace and experimentation mindset’

David Beitel, who as CTO of Zillow Group also oversees internal IT, had a similar take.

“AI models and tools, and in particular LLM-based models, have improved dramatically in the past 12 to 24 months and continue to improve at an accelerated pace, enabling innovative customer-facing products and internal tooling and automation in ways we didn’t think were possible [entering] 2025,” he says.

At Zillow, Beitel and team are exploring ways AI can help make customer and employee experiences more personalized, efficient, and rewarding.

“The internal productivity gains are allowing our employees to level up their work and insights and focus more time on strategic planning and execution. AI models and technologies and tools, coupled with strong professionals, operators, and employees make a powerful combination. That human in the loop is important as we look to deliver the best products and provide the right guardrails and oversight,” he says.

He adds, “I would also say that things are changing quickly, so it’s important to keep a fast pace and experimentation mindset and continue to dream of what is possible with these new technologies.”

3. ‘Thinking in terms of work charts not org charts’ is a must

Sanjeev Satturu, senior vice president and CIO of convenience store chain Casey’s, says the past year has taught him the value of “thinking in terms of work charts not org charts.”

“The nature of work is fundamentally shifting in the age of AI,” he says. “Projects and workflows often cut across traditional silos, requiring rapid reconfiguration of teams. We have to optimize for tasks and outcomes by mapping work and not roles. The future of work is skills based and not position based. Work charts foster agile networks enabling faster launches, cross-functional collaboration, and real-time adaptation to market.”

He notes that “work charts represent how work actually gets done — dynamic networks of tasks, skills, and collaborations. Work charts make workflows visible, reducing ambiguity and improving engagement.”

Satturu further makes his case for this approach by explaining that “work is no longer static; it’s fluid, cross-functional, and project-based” and that “skills matter more than positions.”

Given that “AI optimizes tasks and outcomes, not titles or hierarchies,” Satturu has learned that it’s imperative “to design for work, not hierarchy, as work charts tell you how value flows.”

4. Problematic data is the new ‘legacy tech’

Like many IT leaders, Savio Lobo, CIO of Ensono, which provides technology advising and managed services, has had to contend with problematic data. He has seen “mismatched data,” where the same type of data is treated differently from one business department to the next, as well as other inconsistencies and gaps in the organization’s data.

Such problematic data has become the new “legacy tech,” he says, calling it a “millstone” that is holding organizations back the same way that outdated IT systems do.

“That has been a lesson-learned for me,” he says, noting that the past year has shown that IT leaders need to put data governance and data stewardship higher up the priority list — and start thinking about data as an organizational asset.

“Yes, the business owns the data, but data goes across units,” Lobo says, “so getting the organization rallied around data definitions, data quality, meta data, etc., is critical.”

To do that, Lobo has established a solid centralized enterprise data catalog and a cross-functional data team that prioritizes and enforces data quality standards.

Lobo acknowledges the monumental effort required to get enterprise data in pristine shape — a task that’s just as challenging for most organizations as modernizing legacy technology.

So he’s approaching modernizing his company’s data like many CIOs approach modernizing legacy tech: by doing it incrementally. “You can’t boil the ocean, but you can tailor it, focus it, focus on data cleanup in support of the highest value AI use cases,” he says.

5. IT needs to use leading indicators ‘to look at projects in flight’

Lobo cites another key takeaway from 2025: the need to identify and track leading indicators instead of lagging ones to measure projects, particularly large transformation projects.

Use of leading indicators gives stakeholders a clearer picture on whether the initiative is on track to deliver expected outcomes, he says, adding that leading indicators deliver that insight in a more timely manner than lagging indicators do.

“We have a metrics dashboard in place to look at projects in flight, and we make sure they’re reported in the right way with the right tools in the right place,” Lobo says, noting that this allows for “intervention midstream, not after the fact,” allowing adjustments that get any faltering projects back on track so they deliver returns sooner than later.

This is particularly important now as organizations tighten budgets and stress the need for projects — including those with AI — to show value.

“We need the budget to do AI and other transformation work, and in order to make discretionary funds available for that work, we need to ensure we’re spending the money we get to get the outcomes we want,” Lobo adds.

6. CIOs can’t ‘wait for that perfect moment’ to move forward

There was a point when Raul Pena, CIO of VLS Environmental Solutions, felt his company wasn’t ready for AI.

“When I joined the company a few years ago, my boss started asking me about AI, and I pushed back because I thought we had to be at a 10 from a sophistication standpoint, and we were toward the lower end of the scale. We had disparate systems, and we had challenges with data,” Pena says.

While the company still isn’t at 10 (Pena puts it at a 7 or 8), he has come to see that he couldn’t “wait for that perfect moment to say, ‘We’re ready.’ If we wait for that moment, we’ll fall behind.”

As a result of that recognition, Pena and his company are now well on their AI journey, embedding the technology in systems and processes.

“At this point anything we do has an AI component to it,” he says. “We want to position ourselves as a future-ready tech-enabled company that fosters innovation and that empowers our people. People want to work for companies that are tech enabled and are willing to invest in technology to innovate.”

7. ‘AI is as much a social and cultural phenomenon as it is a technical one’

Pena also has come to see that as a CIO implementing AI in his company he also has to contend with the societal issues and implications that come with it. “The realization is that AI is as much a social and cultural phenomenon as it is a technical one,” he says.

To start, not everyone is a fan. “There’s a big chunk of society that doesn’t understand it, doesn’t use it, doesn’t like it,” he says.

And many fear what AI means to them and their livelihood. “There are fears that AI will steal jobs. And it will take some jobs. Like other technology improvements in the past, there is disruption and destruction of some people’s jobs, but there will also be new jobs because of it,” he says.

“We want our employees to know that we know that and see that and we’re making investments to help them do their jobs using AI,” he adds. “That’s the message we want to give to employees — that if they learn to use it, they’ll create value for themselves and the company.”

8. ‘There’s a need to treat AI projects like real programs’

To harness value from AI, which multiple studies have shown to be a challenge for most, organizations must bring the same business and fiscal discipline to AI initiatives as it does any other project.

That’s the conclusion that Mark Sherwood, executive vice president and CIO at Wolters Kluwer, reached this year.

“There’s a need to treat AI projects like real programs,” he says, noting that many organizations have gotten away from that to varying degrees because of the pressure they’re getting to quickly adopt AI.

“What I’ve seen and what we’ve struggled with a bit with my team is there is so much innovation with AI in the company, which is great and I want to encourage, but I also want to understand the business value of what we’re doing. It comes down not to whether we could do it, but whether we should do it,” he says.

To answer that, Sherwood requires AI innovators at Wolters Kluwer to articulate the “qualitative value and quantitative value” of what they’re pursuing.

“It’s almost like project management 101. It’s the basics that we had been doing in IT for a long time, making sure we have the fundamentals in place before we jump in. As boring as it might sound, it’s making sure [with AI innovation] we focus on what are the business problems we’re trying to solve,” he explains.

With that lesson in mind, Sherwood has been applying project management principals to proposed AI initiatives, “understanding what the business needs are, being clear on goals and timelines, and putting together clear success metrics.”

He has a vetting process to review AI-related proposals and prioritize them based on business value. He also has guidelines in place for business teams experimenting with AI so they, too, are focusing on value. And he has hackathons and innovation forums to surface promising ideas, whether they feature AI or other technologies, thereby ensuring that the guardrails he has in place don’t eliminate opportunities that show promise.

“AI can be a solution in search of a problem, a hammer that’s always searching for a nail,” Sherwood adds. “I always prefer to start with the problem rather than coming in with a solution. AI is not a magic bullet. It’s not a bit where you throw stuff in and magical qualities come out the other side. It’s much like any other tool.”

9. ‘Transformation hinges on people far more than on technology’

For Beth Clark, CIO of Harvard Business School, the past year has taught her that the CIO job is more about helping people than it is about leading technology.

“Transformation hinges on people far more than on technology, and change is hard for people,” she says.

Clark says to help people transform, she to bring clarity to staffers on what her organization wants to achieve and how AI and other technologies will help them get where they want to go.

“It’s helping them brainstorm about their vision for the future,” she says, “asking them what their [work] looks like today, asking what it could be in the future, and crafting that vision with them. It’s presenting the ‘What if?’ and ‘What if you could do x, y, and z?’ questions. That requires being a visionary and having enough depth of knowledge about how the business works day to day and what the technology possibilities are so I can craft the vision side by side with them.”

At that point, Clark says, it’s just as vital to build excitement for the transformation change. “It’s letting them see the art of the possible and telling a story about what life could be like in two, three, five, 10 years if they’re not just invested in change but excited by it,” she says.

That, she adds, requires sympathy and empathy, understanding that people are being asked to give up “tried and true ways of doing things” and recognizing that that’s a big ask for most people.

“It’s helping people adapt to a world that is changing very, very fast,” Clark says. “It’s letting them know that it is a continuous partnership to make sure we’re keeping that vision really clear and we’re doing all the work to get us there.”

It’s a long road, Clark acknowledges, and it’s understandable that some will question whether it’s worth it for them to go along.

“You have to hold people’s hands at the same time you’re kicking them in the butt,” she adds. “You need a really good influence toolkit, to influence them in different ways based on what’s needed, whether that’s leading difficult conversations when they’re stuck or being a shoulder to cry on.”

10. ‘Only multi-agent architectures can deliver the agility and resilience required to thrive in today’s environment.’

Neal Ramasamy, global CIO at Cognizant, says 2025 taught him it’s time to create architecture that can support widespread agentic AI.

“Single AI solutions are giving way to coordinated agent ecosystems,” Ramasamy says. “As business processes become more interconnected and dynamic, only multi-agent architectures can deliver the agility and resilience required to thrive in today’s environment. These systems enable specialized AI agents to collaborate seamlessly, breaking down operational silos and driving end-to-end automation across functions like finance, HR, supply chain, and customer service.”

Ramasamy stresses the need to act now.

“Enterprises without multi-agent strategies will face significant productivity gaps within 24 months,” he adds. “But organizations that invest now in orchestrating diverse agents — while embedding robust governance and interoperability — will set the standard for enterprise performance and innovation.”