The Zylo 2025 SaaS Management Index, which tracks real subscription spend across thousands of companies, arrived at a number that sums up the state of corporate technology adoption well: companies with more than 10,000 employees run 660 different applications on average. Smaller companies, under 500 employees, run 152. And the same report measured that these organizations waste an average of $21 million a year on licenses nobody uses — 14.2% more than the year before.
That number was already high before generative AI became routine corporate spending. It is about to gain a new, bigger chapter. Gartner forecasts that a Fortune 500-scale enterprise will have more than 150,000 AI agents in use by 2028 — up from fewer than 15 in 2025. That is not gradual growth: it is the same tool-sprawl logic already choking software budgets, applied to a kind of system that acts on its own, makes decisions, and needs permission to do so.
Faced with each new problem, the temptation is always the same: buy one more tool. But ten scattered subscriptions do not add up to one AI capability — they add up to ten islands, each with its own login, its own access policy, and no shared budget between them.
In Brazil and Latin America, this pattern usually shows up department by department: marketing signs one tool, support signs another, legal tries a third — each contract approved in isolation, with nobody adding up the total or asking whether one of the three could already cover the other two.
The problem is not the tool, it's the sum of them
Flexera's "2026 State of the Cloud Report," based on more than 750 cloud decision-makers, found that cloud waste grew 29% — the first increase in five years — driven specifically by AI workloads. The same survey found that 81% of companies already use generative AI, up from 72% the year before and 47% in 2024. Adoption is accelerating faster than the ability to manage what has been adopted.
The pattern repeats across all three studies: fast growth in tools, slower growth in control over them. And when control does not keep up with purchasing, the cost shows up two ways. The first is direct — a license paid for and unused, like the $21 million Zylo measured. The second is indirect and more expensive: every new tool needs its own access setup, another place to audit, one more surface where a permission mistake can expose sensitive data.
It is worth noting that Flexera's own 81% growth figure for generative AI use did not come from a centralized plan — it came from dozens of one-off decisions, each made by a team trying to solve its own problem as fast as possible. It is exactly this pattern of uncoordinated decisions, added up, that produces the sprawl all three reports are measuring.
None of these tools is the problem on its own. The problem is that each one solves a small slice of the company's need and arrives with no conversation with the other nine already in place.
Why buying one more tool does not solve it
Every new tool added to this landscape arrives with its own account, its own login, and its own audit trail — or, in most cases, no audit trail at all. Nobody in legal or security sees the full picture, because there is no full picture: there are ten separate dashboards, each showing one slice.
The budget suffers the same effect. When every team subscribes to its own AI tool, the company pays N times for the same basic capability — text generation, document summarization, answering a simple question — just embedded in N different products, with N contracts, N renewal dates, and N support teams to call when something breaks.
There is a third cost, the hardest to put on a spreadsheet: what one team discovers about using AI well never reaches the others. If the sales team figures out the right way to configure an agent to qualify leads, that setup stays locked inside the tool only sales uses. The support team, facing a similar problem, starts from zero in a different tool, under a different contract.
What has to be in place
Consolidating into one governed AI environment solves all three costs at once — not by removing capability, but by organizing it.
A single identity for every connected tool. People sign in once, with the company login, and the same access covers whichever agents and integrations they are authorized to use — instead of one password per subscription.
Access by role, not the whole tool. An integration might have dozens of functions; the support team gets the two it needs, not the entire tool "because that was the only option available."
Shared budget, visible consumption by team. Instead of one subscription per person — with a license sitting idle while someone is on leave or changes roles — AI credits shared across the team, with spend traceable by department.
One audit trail. A single place to answer who used what, when, and under what permission — instead of reconstructing the story across ten different systems' histories.
Reuse across teams. An agent configured by sales can be adapted and reused by support, with access defined by role, instead of every team paying again for the same solution.
This is what Skyller delivers as the default: a single corporate identity, AI credits shared across the team, and more than 100 pre-configured agents any department can adapt instead of buying from scratch.
From fragmented cost to shared capability
The most visible gain from consolidation is financial: a shared budget with visible consumption by team replaces the scenario of an idle license on one side and people hitting a limit on the other. But the bigger gain is operational. When there is a single environment, what one team learns stops being locked inside the tool it happened to pick on its own.
This also changes the conversation around technology budgets. Instead of approving one more subscription because "this tool solves exactly this problem," the natural question becomes whether the environment the company already has can solve the same problem with a new agent, configured inside it — without another contract, another login, and another audit trail to maintain.
With the number of AI agents per company about to jump from dozens to thousands, according to Gartner itself, that decision only gets more urgent. A company deciding today where and how each new agent will live is, in practice, deciding whether it will manage hundreds of separate systems three years from now, or a single environment.
That urgency also changes the role of whoever approves the technology budget. The question stops being only "does this tool solve the problem?" and starts including "where will this capability live two years from now, when the company has ten times more agents in use than today?" Without that second question, every individual approval keeps making sense on its own — and the sum of them keeps producing the same sprawl.
Questions for the next budget meeting
Before approving another AI tool, it is worth bringing these questions to whoever decides budget and technology:
- How many AI tools does the company already pay for today, and does anyone know the exact total? If the answer requires pulling spreadsheets from different departments, the problem has already started.
- Could a tool the company already has solve this new request with an agent configured inside it? Often the answer is yes, and nobody asked before signing another contract.
- If someone leaves the company tomorrow, in how many of these tools does their access stay active? Every uncertain answer is a license — and a risk — nobody is monitoring.
- Does what one team learns about using AI well reach another team, or does it stay locked in the tool only that team uses? If it stays locked, the company is paying to relearn the same thing in every department.
