Get ready for the agent store

Most businesses don’t have a clear inventory of the agents already running inside them, because every department is quietly hiring. Sales has an agent chasing leads, finance has one reconciling invoices, and operations is watching stock levels. None of it went through much of an approval process, which is unusual for most technology but not for this one.

Agents are being adopted faster than most IT and risk teams can keep track of, and the pace has probably caught your business out too.

None of that is a bad thing on its own. Move fast with the right foundation in place and you can have new agents live in weeks while competitors are still running pilots. Move fast without it and that same speed leaves you with agents nobody can properly account for. The industry already has a name for that problem: agent sprawl.

The numbers tell the story

You are not imagining the pace. Gartner expects 40% of enterprise applications to ship with task specific AI agents by the end of this year, up from under 5% in 2025, and nearly four in five companies now report some form of agent adoption.

But adoption is not the same as scale. McKinsey finds fewer than a quarter of organisations have taken an agent past pilot into something the business actually runs on. Most agents never get much further than a limited deployment, and that is the real problem.

The average business already runs dozens of agents, a number that climbs every quarter, built by different teams on different frameworks with no shared oversight. Fewer than one in five organisations has a mature way of governing them.

Agent sprawl is what happens when a business ends up with more AI agents than it can see, track or govern, usually because different teams built them independently. Moderna is a good illustration of how fast this can happen.

It has reported that its employees built more than 3,000 custom AI assistants once they were given the tools to build their own, with no central engineering team involved. Not all of those are full agents in the sense this piece means, but the pattern holds regardless: give people the tools and they will use them, often well ahead of anyone tracking what has been built.

The businesses that make it past the pilot stage usually did the unglamorous work first: systems integrated, data connected, clear permissions and governance, one place to see what every agent was doing, before the agents arrived rather than scrambling together afterwards. Without that, nobody can say who approved a given agent, what data it can reach, or what happens if it fails.

What businesses need before the agent store arrives

Before app stores existed, installing software on a phone was chaos. No review, no permissions model, no way to know what an app could access until it was too late. App stores fixed that with curation, permissions and oversight, and agentic AI is heading the same way. Eventually, every business will have its own agent store, a single place where every agent is catalogued, approved, permissioned and watched.

That is the future state. Before businesses get there, they need the foundation in place: systems that are properly integrated, data unified into one view, and the insights to know what is actually happening across the business. On top of that, they need an AI Control Tower, a single place to deploy, govern and scale every agent they run.

Not being agentic ready will cost you

Fewer than one in five organisations can say they have proper oversight of their agents, and that gap does not stay hidden for long. Regulators, customers and risk committees are already asking harder questions about how AI is used inside a business, and executives who cannot answer basic questions about their own agents will be the ones caught out.

Businesses that put the basics in place now will still be scaling agents with confidence in two years. The ones that do not will be explaining, after the fact, why nobody was watching.

You do not need to start from scratch

Xiatech will soon formally launch its AI Control Tower, built on the Xfuze platform. The foundation and the control come from the same place, so adding an agent does not mean adding a blind spot. More on what agents actually need to work: AI Agents explained: What they are, and what they need to work.

This October we are also announcing a new wave of Xfuze Agents, alongside our existing Xfuze Stock Replenishment Agent, including for the Google Cloud Marketplace.

Quick answers

What is agent sprawl? More AI agents than a business can see, track or govern, usually because different teams built them with no shared oversight.

What is an agent store? A single place where every AI agent a business uses is catalogued, approved, permissioned and watched, the same way an app store governs apps on a phone.

Is Xfuze available on the Google Cloud Marketplace? Soon. A new wave of Xfuze Agents for the Google Cloud Marketplace lands this October, alongside the Xfuze Stock Replenishment Agent already live with customers.

Full details land in October. Get in touch with your Xfuze contact for a preview, or see what AI can already do for your business: Xfuze AI Lab: See what AI can do for your business.

Read more about the upcoming launch: Major Xiatech Product Launch.