Artificial intelligence promised to transform IT operations. Instead, it is exposing the shaky foundations that were there all along.
The struggle to push AI from pilot to production has become a defining story in enterprise IT. Leaders who "let AI loose" in messy environments are discovering that AI in IT operations magnifies whatever is broken underneath, says Kevin Carroll, director of next-generation data services for the global tech solutions integrator Insight.
"From what I see in the industry and what I see with our clients, what AI has brought forward is the fact that they weren't doing good IT," Carroll told The Forecast at Splunk's .conf26 conference in Denver. "If the telemetry that AI is receiving is suspect, if there's no quality in that data, then the output of AI is not quality."
The evidence is mounting. Ever since MIT's Project NANDA claimed that 95% of AI initiatives produce essentially zero value in a widely circulated July 2025 report, a new statistic seems to emerge each week illustrating the challenge of turning AI investments into lasting enterprise value. In an August 2026 report, McKinsey & Company found that AI is improving individual productivity for 8 in 10 respondents, but only 37% say AI is contributing to their organizations' EBIT. In early September, Gartner reported that just 22% of organizations have successfully scaled AI across multiple business units or adopted an AI-first approach. And only 12% of executives report both cost reductions and revenue gains from AI, according to PwC.
AI in IT operations is the use of artificial intelligence to monitor, manage and automate enterprise IT environments, from patch management to infrastructure upgrades. Analysts have blamed the slow return on investment on factors ranging from GPU shortages to talent gaps to unchecked costs, with scattered, siloed data environments perhaps the most commonly cited culprit.
Carroll describes the problem as more holistic. Most organizations, he said, have not made their existing IT operations observable, predictable and repeatable enough to fully take advantage of AI.
"If you were doing good IT, if you had solid processes and solid policies, you can now use AI to take advantage of that and accelerate it," Carroll said. "But AI is exposing bad IT."
Why Is AI in IT Operations Exposing Bad IT?
The answer is speed. As companies move from generative AI assistants toward agentic systems capable of taking autonomous actions, IT fundamentals matter even more.
Not only are leaders determined to make their own processes more efficient by offloading repeatable tasks to AI agents, Carroll said, but attackers have access to AI agents as well. While enterprise IT teams must deploy agents carefully, attackers can unleash malicious agents without worrying about enabling improper permissions or breaking business systems.
"How do we do security when the attacks are coming at machine speed?" he said. "By the time we see it, it may have proliferated through the entire environment."
Currently, Carroll said, some IT teams are doing a high-tech version of unplugging their networks, throttling firewall traffic to the point that business-critical workflows like order processing grind to a halt. Organizations with sufficiently rich, contextualized telemetry can respond more surgically, he said, quarantining affected systems or tightening individual firewall rules without shutting down access.
This gets at what Carroll means by "good IT." Before organizations can safely and effectively scale AI, he said, they need to understand the state of their environments, including what they can monitor, how systems normally behave and what effects changes are likely to have.
How Fast Can Good IT Move With AI?
When that understanding is in place, AI can deliver transformative results. Carroll offered the example of one Insight customer whose Splunk infrastructure upgrades previously took around 14 months. By the time one upgrade was finished, the organization needed to begin planning for the next one.
Insight took about eight weeks to map the customer's telemetry and establish observability, figuring out what could be monitored, what could not and what "normal" operating conditions looked like.
"It took them 14 months to upgrade their infrastructure, and now it takes 14 hours," Carroll said. "That's massive business value. Then, you go after other use cases and more value based on that."
Where Will AI Deliver Value First?
Outside the IT shop, Carroll said, many organizations are likely to see the greatest near-term value in the form of "citizen development," with line-of-business users amplifying their own productivity without waiting for the organization to provide them an app that meets their unique needs. He pointed to the ability of executives and other non-IT employees to use natural-language interfaces to interrogate their data.
"AI is going to make you better," Carroll said. "But it's not always going to make the entire business better right away. We're enabling the tooling that all of us geeky guys in IT use, and offering automation to fill the needs of somebody who needs to perform an action."
The near-term opportunity for businesses is less about using AI to replace employees than allowing existing teams to multiply their efforts, Carroll said. "I argue it's not a cost-control thing," he said. "It's a scale discussion."
Already, Carroll said, the industry conversation has moved away from the idea that AI will "do it all" and toward a model in which people remain responsible for supplying important context, exercising judgment and approving critical actions. In IT, professionals may begin to take on more comprehensive roles covering security, application development and infrastructure, while relying on AI to help them cover these broader portfolios. AI is likely to introduce more than enough new challenges and opportunities to keep existing IT teams busy.
"I see the industry changing," he said. "I don't see it getting smaller."
Related:
Calvin Hennick is a contributing writer. His work appears in BizTech, Engineering Inc., The Boston Globe Magazine and elsewhere. He is also the author of Once More to the Rodeo: A Memoir. Find him on LinkedIn.
© 2026 Nutanix, Inc. All rights reserved. For additional information and important legal disclaimers, please go here.
Related Articles
