A new survey of nearly 1,900 IT leaders has confirmed what many business owners already sense: AI agents have moved from experiment to infrastructure almost overnight. The problem is that most organisations did it without a plan.

The research, published April 7 by OutSystems, found that 96% of enterprises are already using AI agents in some capacity, and 97% are actively exploring system-wide agentic AI strategies. That is close to universal adoption. But the same study found that 94% of organisations are raising concerns that AI sprawl is increasing complexity, technical debt, and security risk.

In other words, the agents are in production. The governance is not.

What the Data Actually Shows

The OutSystems survey — conducted across global IT leaders between December 2025 and January 2026 — paints a picture of an industry that moved fast and is now dealing with the consequences.

38% of organisations are mixing custom-built and pre-built agents, creating fragmented AI stacks where different teams use different tools, different vendors, and different data sources. Nobody has the full picture.

Perhaps most telling: only 12% of organisations have implemented a centralised platform to manage agent sprawl. That means 88% of companies with active AI agent deployments are operating without a unified way to monitor, govern, or audit what those agents are doing.

This is not a theoretical problem. When agents are scattered across business units — each built differently, connecting to different systems, with different levels of oversight — you get compounding risk. A fragmented agent stack creates security gaps, makes troubleshooting nearly impossible, and generates the kind of technical debt that takes years to unwind.

Why This Moment Matters

The shift from single-agent deployments to multi-agent systems has happened faster than most IT teams anticipated. A year ago, enterprises were running small pilots. Today they are running workflows where agents hand off work to other agents, where customer interactions are managed end-to-end without human involvement, and where business decisions are shaped by AI outputs that nobody quite understands.

The 94% sprawl concern number is striking because it is not coming from sceptics. These are IT leaders who are already deployed. They are not worried about whether AI agents work. They are worried about managing what they have already built.

The risk profile of agent sprawl is different from traditional software sprawl. When a company accumulates too many SaaS subscriptions, they waste money. When they accumulate too many AI agents operating without oversight, they expose themselves to compliance failures, data breaches, inconsistent customer experiences, and decisions made by systems nobody can audit.

The Governance Gap Is Real

Most of the AI conversations in 2025 centred on capability. What can agents do? How smart are the models? That conversation has largely been won. Agents can do a lot.

The conversation that needs to happen now is about control. Who owns each agent? What data does it access? What decisions can it make autonomously? How do you trace an error back to its source when three agents were involved in the process?

Only 12% of enterprises have a centralised answer to those questions. The other 88% are figuring it out case by case.

What This Means for Business

If you are running AI agents in your business today, this research should prompt a practical audit. Not a theoretical one.

Start by listing every AI agent or automated AI workflow your team uses — across every department, not just IT. Include the tools individuals are using on their own, the vendor integrations that include AI components, and anything your team has built internally. Most business owners are surprised by how long that list is.

Then ask: what does each one access? Who reviews its outputs? What happens if it makes a wrong decision?

You do not need a formal AI governance framework to start. You need someone accountable for the question.

The organisations that will perform best in the next 18 months are not the ones with the most agents. They are the ones that know what their agents are doing and can manage them as a coherent system rather than a collection of individual experiments. The distinction between an AI workforce and a set of disconnected AI tools is exactly what separates companies that scale from companies that accumulate debt.

---

If you’re deciding where to start with agents, start here. The free Working With Claude field guide walks through the ecosystem, Claude Code, and a real rollout plan. Get your copy.