Every company that has adopted AI over the last two years now has a hidden inventory problem. Nobody built these tools with a plan to track them, and most were never meant to be permanent. Yet many of them quietly became part of how the business runs.
This did not happen because of one big decision. It happened through thousands of small ones, each one reasonable on its own. Someone in sales built a workflow to draft follow up emails, someone in operations connected a chatbot to a spreadsheet, someone in finance automated a report.
Each one worked, so it stayed. It got copied, connected to other systems, and leaned on by more people over time. Nobody ever made the call to turn it into permanent infrastructure, it just became that anyway.
Gartner has put a number on where this is heading. The research firm predicts that a typical large global company will be running more than 150,000 AI agents by 2028, up from fewer than 15 of these tools in 2025. That is the difference between a handful of tools a manager can name from memory and a number no one can count by hand.
The governance side has not kept pace. Gartner also found that only a small share of companies believe they actually have the right controls in place to manage this growth. Separately, multiple surveys this year found that a large majority of employees at big companies already use AI tools their employer never approved, often connecting those tools to email, shared files, or customer records without informing IT.
This matters because unmanaged AI tools are not free even when they look free. A widely circulated MIT study tracked company spending on generative AI and found that ninety five percent of projects delivered no measurable return, despite the industry spending tens of billions of dollars chasing results. The problem was rarely the technology itself, it was that nobody had a plan for what happens to a tool after it launches.
This is a repeat of something that happened before with company software, just faster and less visible. Years ago, business teams tired of waiting for IT started buying their own software directly, which came to be known as shadow IT. Companies eventually fixed that by building simple rules: know what exists, know who owns it, and retire tools that are no longer needed.
AI needs the same fix, and it needs it sooner. A chatbot or workflow does not look like software to the person who built it, so it never gets added to any list. It just runs, quietly, until it touches something important like customer data or a financial decision.
The fix is not to slow down, it is to keep a running list of what AI tools exist across the company and who is responsible for each one. Companies that build that habit now will spend the next few years scaling AI with confidence. Companies that skip it will spend those same years finding out what they built only after something goes wrong.