Fewer Than 15 Agents. That Was the Average in 2025.

Ask a CIO how many AI agents run inside their company today. Watch the pause before the answer. Gartner says that same average enterprise is headed for more than 150,000 agents by 2028. Not tools. Not licenses. Agents, each one making decisions inside a workflow. Only 13% of organizations believe they have real governance over them.

I think about the finance director who signs the invoices. A year ago she approved three or four AI line items: a writing assistant, a forecasting model, a chatbot for the help desk. This year the requests arrive weekly, from teams she doesn't always recognize, for tools that do the same job under different names. She isn't approving software anymore. She's approving a population.

Here is what's actually happening beneath the AI-transformation headlines. Enterprises adopted first and counted second. The bill for that order came due in 2026.

Zylo's 2026 SaaS Management Index draws on more than 40 million licenses and $75 billion in tracked spend. It puts the average enterprise portfolio at 305 applications. Large companies add 21 more every month. AI-native app spend at companies with over 10,000 employees jumped 393% year over year. That growth doesn't stay inside the IT budget. Business units now control 81% of SaaS spend; IT directly manages just 15%. The tools spread faster than anyone assigned to track them.

Salesforce's 2026 Connectivity Benchmark surveyed 1,050 IT leaders across nine countries. It found something smaller but just as telling. Enterprises run 12 AI agents on average today, and half of them sit in silos, unable to see or coordinate with the others. Twelve looks small next to Gartner's 150,000. That's the point. The sprawl already started, at a scale nobody named yet.

The cost shows up in ways a budget line can't capture. Zylo found 78% of IT leaders hit unexpected charges tied to AI or consumption pricing this year. 61% had to cut a planned project because costs ran past what they'd approved. Multiply that across a few thousand companies and you get the real story behind the AI capex numbers. Money isn't only funding strategy. A meaningful share of it is funding confusion.

I watched a version of this before, in a different decade, with the same root cause. In the 1990s, companies raced to add enterprise software the same way: a system for finance, a system for sales, a system for HR, each bought by a different department, each sold as the fix. It took most of a decade and a wave of ERP consolidation projects before anyone admitted the sprawl itself had become the cost center. The individual tools weren't the problem. Nobody owned the whole picture until someone was forced to build one.

Every one of these AI purchases got approved by a person solving a real problem in front of them. None of them checked with the twenty other people solving their own version of it the same week. Vendor sprawl is a visibility failure, not a judgment failure. It compounds the way debt does. Small and easy to ignore for a year. Then all at once, in a budget review nobody enjoys.

Sprawl costs more than money. It shortens the runway on the skills your people are building. Every tool carries its own half-life, and when that tool gets replaced or folded into something else 18 months later, the muscle memory goes with it. A company running hundreds of overlapping AI systems pays twice for the same capability. Then it asks its people to relearn the same job again, on a schedule the company set and never said out loud.

If you're the one learning the fourth new tool this year, not the one buying it, that schedule is not your fault. Ask your manager plainly which of your current tools survive the next consolidation round, and which training you're doing is likely to outlast the vendor. You deserve a straight answer before you invest another weekend in a tool nobody committed to keeping.

Real consolidation gives people fewer tools to master, so they can go deeper on the ones that matter. It builds the human-plus-agents stack around a short list instead of a shifting one. A Relearning Organization treats every new AI purchase as a training obligation, not just a line item. It counts the tools the way it counts the people.

Multiply that discipline, or the lack of it, across a few thousand companies and the effect reaches past any single budget. Households plan around a paycheck that assumes a job stays roughly the same shape for a few years. A workforce retrained every 18 months by a vendor's roadmap, instead of by its own choices, doesn't get that stability. The sprawl in the data center becomes the churn at the kitchen table.

The finance director signing this year's invoices doesn't need a bigger budget. She needs a map. Which of these tools does the same job as three others already running? Who in this building actually knows the full list, and when did anyone last ask them?

If that answer took longer than a breath, you already know where next quarter's real work sits. I'm easy to find.

Sources: Gartner, "Gartner Identifies Six Steps to Manage Artificial Intelligence Agent Sprawl," April 28, 2026 · Salesforce, 2026 Connectivity Benchmark, February 5, 2026 · Zylo, 2026 SaaS Management Index, January 29, 2026.