When the IT team at Seagate decided to replace the ITSM platform that had run their global IT operations for more than a decade, they had three months to do it.
That was the deadline imposed by a hard contract expiration. Three months to move 30,000 employees across Seagate’s global storage and infrastructure operations onto an entirely new system. Most organizations, in that situation, do the obvious thing: lift the existing configurations, drop them into the new environment, and reconcile the mess later. It’s the safer path. It’s also the one that almost guarantees the AI capabilities the team was counting on will never fully work.
The team chose the harder path. They rebuilt from the ground up — restructured the service catalog, established consistent SLAs across regions, rewrote the category hierarchies so tickets could route themselves without an agent guessing where they belonged. They did so because they intentionally did not want to bring forward their legacy processes. A year in, the AI agent the team deployed on top of that foundation now deflects roughly a third of incoming tickets. First-contact resolution is now 27% above the industry standard.
That decision — to rebuild rather than replicate — is the real story of what separates the companies pulling ahead with AI from the ones that aren’t. And it has almost nothing to do with which model they’re running.
The Complexity Tax
A growing share of enterprise AI investment is being consumed before any value reaches the business. MIT found that 95% of generative AI pilots fail to scale into production. Boston Consulting Group’s September 2025 research found that 60% of companies generate no material value from AI — a figure that worsened from the year prior, despite better tools and more experience. Freshworks’ upcoming Cost of Complexity research puts a finer point on why: one quarter of AI budgets get eaten by integration work, data cleanup, and the labor of forcing systems that were never designed to talk to each other into some kind of coherent conversation.
The pattern is consistent across industries. Programs stall, reset, or quietly get cut. Not because the models don’t work. Because the operating environment underneath them wasn’t ready to support them.
This falls disproportionately on a specific kind of company, the kind I’ve come to call the agile enterprise. These are businesses with five hundred to twenty thousand employees, running lean IT teams, with far less margin for a failed technology bet than a company with a half-billion-dollar transformation budget. When a company in that position loses a quarter of its AI spend to integration overhead, that’s not a rounding error. That’s a canceled initiative.
What the Companies Pulling Ahead Have in Common
But a smaller group of agile enterprises is producing a very different result. They’re not spending more. They’re starting in a different place.
Seagate is one version of this. New Balance is another. Nike runs on 80,000 employees. New Balance runs on 9,000. And New Balance is taking share, not by getting bigger, but by getting faster and sharper. The company didn’t win that ground by doing anything glamorous. It won it by consolidating a fragmented IT stack onto one platform with a single source of truth, freeing teams from maintenance work and rewiring how the business operates.
That’s the kind of foundation work that pays off well before AI enters the picture, and it’s exactly the foundation that lets AI work when it arrives. Companies like Nucor and Steel Dynamics, two of the top four U.S. steel manufacturers, show the same pattern at industrial scale: decades of operational discipline produced operating environments that AI could actually optimize.
Across all of them, AI is working where the operating model was ready for it. Not perfect. Ready. Meaning the data was consolidated, the workflows were defined, the systems could pass information without manual intervention, and there was a clear, measurable outcome the AI was being asked to improve.
How to Start When You’re Starting From Messy
Most companies aren’t where Seagate is now. Most are somewhere in the middle — a legacy platform that’s been in place too long, data scattered across systems that don’t quite line up, an IT team that’s spent more of the last five years keeping things running than rebuilding them. The question isn’t whether AI will work on top of that environment, but rather where to start.
Robert Lyons, the CTO of Katz Media Group, has one of the cleanest answers I’ve heard. Katz is an eight-hundred-person business unit inside a ten-thousand-person parent company, exactly the kind of agile enterprise that can’t afford to chase every AI initiative that sounds compelling. Lyons maps every potential AI project onto what he calls a value/effort matrix: ease of implementation on one axis, business value on the other. He starts in the high-value, low-effort quadrant and works outward from there. “Don’t start with the worst problem first,” he said recently. “You’re not going to deliver the value. Focus on ease of implementation with immediate payback.”
Before Lyons’ team deployed any AI tool, they did two things most organizations skip. They cleaned and labeled their data – because feeding messy data to AI and wondering why the results disappoint is the most common failure mode in the enterprise right now. And they ran an AI primer webinar for every employee in the company, presented not by IT but by a neutral third-party research firm. “It’s not IT barking at you,” Lyons said. “A neutral party socializing this makes it land differently.”
That sequenced, disciplined, outcomes-grounded approach separates the companies that are getting AI to work from the ones that are still talking about it.
Where the Advantage Actually Lives
Across every agile enterprise I’ve seen succeed with AI, three operational traits show up consistently. None of them are about which model the company chose.
They reduced fragmentation before they added intelligence. Not by consolidating everything into a single super-platform — that’s a different and usually a more expensive conversation — but by making sure the systems that mattered could exchange information without manual handoffs. This isn’t glamorous work. It doesn’t make for exciting board presentations. But it’s the single highest-leverage thing a mid-market company can do before writing a check for any AI tool.
They applied AI where it improves execution, not where it creates more complexity. The best use cases in the agile enterprise aren’t moonshots. They’re workflow acceleration: faster ticket resolution, smarter demand planning, automated quality inspection, predictive maintenance scheduling. Use cases where the inputs are structured, the outputs are measurable, and a human stays in the loop.
They treated AI adoption as an operating discipline, not a technology project. The companies pulling ahead didn’t hand AI to an innovation team and wait for a report. They embedded it into the daily work of the teams closest to the customer, the production line, or the revenue cycle — and they measured it the same way they measure any other operational investment: by whether it moved a number that matters.
The Agile Enterprise Moment
AI is often discussed as though it’s a capability only the largest and best-resourced companies can deploy at scale. That framing is wrong, and it risks becoming a self-fulfilling prophecy for the agile enterprises that believe it.
Agile enterprises represent the vast majority of businesses globally. If AI’s productivity promise is real, it will be proven or disproven in these organizations, not in the handful of trillion-dollar enterprises running bespoke foundation models.
Part of a CEO’s job right now is to live in the future and understand where technology is going. But the other part – the harder part – is to bring that vision back to the present and ship something that solves a real business problem today. The companies I watch doing both aren’t the ones with the biggest budgets. They’re the ones that made a deliberate choice, somewhere along the way, to stop dragging their past forward and start building for what comes next.
That’s a choice any agile enterprise can make, starting now.
The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.
-1024x768.jpg "Red Snapper Used as Cudgel by Fed Judge")
