A Fortune 500 CTO Confesses: $20 AI Subscriptions Became $200 Chaos

A Fortune 500 solutions integrator watched modest $20-a-month AI subscriptions quietly balloon to $100 or $200 per employee before anyone put a single owner on the bill. Juan Orlandini, chief technology officer of North America for Insight Enterprises, made that confession in a first-person account of corporate AI sprawl — and argued the cure already exists, because the industry has lived through this mess once before.

The numbers he cites are small individually but telling in aggregate. Well-meaning employees swiped corporate credit cards for AI tools at $20 a month. When finance finally looked closely, some of those same subscriptions had become five to ten times more expensive. The fix, Orlandini said, was structural: one global agreement with the frontier model providers turned countless expense reports into a single line item that someone owns. Whether that someone has actually tamed unit costs, he does not say.

What went wrong beyond the subscriptions?

Token costs were the second trap. Orlandini's developers built tools and workflows without thinking past the point of innovation to the running cost. He summarized the episode in three words: "Mistakes were made." No figure is attached to that admission, which is worth noting; the reader is asked to accept the damage on trust.

The third problem was organizational. Every function — partner management, finance, and the rest — got its own AI team, each picking its own frontier model and its own tooling. Orlandini called it "same tune, different song." Five teams meant five frontier models, five sets of tools, and, in his words, "five angry emails from the CFO." Five separate stacks compound technical debt faster than capability is added, and somebody has to become the expert in all five.

Why does the author say this is DevOps all over again?

The precedent dates to 2009, when software development shifted toward the small-team model that became DevOps — echoing Jeff Bezos's axiom that a team should be no bigger than what two pizzas would feed. Companies staffed DevOps projects with five to seven of their best people and shipped features faster. Then the warts appeared: each team chose its own languages, frameworks, and persistence layers, tangling into what Orlandini calls a pile of clothes hangers.

The industry responded with platform engineering and internal developer platforms — a "golden path" where supported tooling is standardized. The approach worked well enough that Gartner predicted in November 2023 that 80% of software engineering organizations would have platform teams by 2026. Gartner's own press release framed that as mainstream adoption within two to five years, a timeline worth watching given how many such predictions slip.

Cloud economics produced a parallel discipline, FinOps, because spinning up servers and consuming APIs looked cheap until the bill arrived. Orlandini's diagnosis of developer behavior is blunt: developers take the shortest path, and the fastest, most expensive button gets pushed over and over. FinOps put a meter on the button and made someone answerable for the number. That cloud-cost problem, he conceded, is still being solved — a caveat that sits awkwardly beside his confidence about AI.

The stakes have grown. AI appeared on 337 S&P 500 earnings calls in a single quarter, per FactSet — the highest count in a decade.

What is the recommended fix?

Orlandini's prescription is sequential, not simultaneous. Customers arrive with a pile of independent AI plans; that independence is the sprawl. His advice: build the single most important plan first, use it as the foundation for the next, and let a small set of capabilities become a flywheel that accelerates projects three, four, and five. The explicit goal is minimizing the blast radius of what gets implemented — platform engineering, rebuilt for the AI side of the house.

He framed the cultural shift in C-suites as already underway, with CIOs and CFOs now saying, "Perhaps we should start consolidating the number of tools we're using." He opened the piece with Mark Twain — "A man who carries a cat by the tail learns something he can learn in no other way" — and closed with the practical corollary: with scar tissue accumulated early, the most valuable lesson is to let someone else carry the cat.

Notably absent from the confession: any hard dollar figure for total AI spend, any named model providers in the global contract, and any metric showing the consolidation actually cut costs. The lesson is offered as experience; the audit trail is not. For companies still issuing corporate cards, the takeaway is plain — the $20 subscription is the smallest number in this story.