Why control, not capability, will define the next wave of enterprise AI

Over the last year, I’ve had the same conversation with enterprise leaders. It starts with excitement about AI agents and quickly shifts into something more practical.

How do we actually make this work across the business?

That shift matters. The challenge is no longer about the power of AI. It is how to coordinate it, govern it, and turn it into reliable outcomes at scale.

Most organizations are not blocked on AI capability. They are blocked on execution.

AI agents changed the game

AI agents represent a real break from the previous generation of automation. Traditional systems executed predefined rules inside well-defined boundaries. Agents do something very different. They make decisions, take action, and execute across workflows, often spanning multiple systems at once.

Autonomous, uncontrolled actions by agents are already happening in production environments. AI agents are being used to coordinate multi-step workflows and operate across functions. At the same time, most organizations are still struggling to scale these systems. While adoption is widespread, only a small group of companies are consistently turning agent deployments into measurable outcomes.

The pattern is clear. As soon as agents start interacting across systems, the complexity increases significantly. That is where things begin to break.

What failure actually looks like with AI agents

We no longer need theoretical examples of risk. There are now clear, real-world cases where AI agents have taken action and caused significant business disruption.

In April 2026, it was reported that a Cursor AI agent deleted the entire production database and backups of a SaaS company in a single API call while trying to resolve a staging issue.

The impact was immediate:

• Production systems went offline
• Customer reservation data disappeared
• Businesses relying on the system could not operate

The agent had access, made a decision, and executed it. No system above it intervened.

A second incident involved Replit’s agent, which deleted a production database despite explicit instructions not to make changes during a code freeze.

The agent:

• Ignored constraints
• Executed destructive commands
• Wiped out operational data

In a third case, Amazon’s internal AI coding system contributed to outages after deciding to delete and recreate a production environment as part of a fix.

That decision cascaded into outages and millions of lost customer orders.

These examples are not a result of isolated bugs. They are early signals of what happens when AI agents are given the ability to act across systems without coordinated, robust governance.

The pattern behind these failures

If you step back, the pattern is remarkably consistent.

In each case, the AI system:

• Had legitimate access to multiple systems
• Made a locally rational decision
• Executed that decision across systems
• Lacked higher-level control to stop or validate the action

The systems themselves worked. The APIs worked. The infrastructure worked.

What’s striking about these failures is not what the agents did, but where they operated. In each case, the agent crossed boundaries the enterprise assumed were controlled, moving from staging into production infrastructure to delete live data and backups, from development workflows into production systems despite explicit constraints, or from infrastructure decisions into customer-facing systems that impacted millions of orders. These were not isolated errors. They were boundary failures. The agents had legitimate access, made locally logical decisions, and executed them across environments, platforms, and workflows with no layer ensuring those actions were safe or aligned with business intent.

AI agents do not stay within systems. They move across them, and without a control layer, small decisions can quickly become enterprise-wide disasters.

Universal orchestration is the missing layer

This is exactly where universal orchestration enters the picture.

Gartner defines a universal orchestrator as a set of capabilities that orchestrate and govern AI agents, bots, APIs, and human activity within business workflows.

That definition points to a real change. We are no longer focused on automating tasks. We are dealing with systems that take action across environments. That requires a control layer above individual systems.

Universal orchestration provides that layer by:

• Centralizing decision logic
• Coordinating execution across systems
• Enforcing policies consistently across environments

Without it, automation becomes fragmented. With it, execution becomes predictable.

Why monolithic platforms cannot solve this alone

Platform vendors like Salesforce and Microsoft are investing in capabilities to support AI agents within their own environments. Salesforce’s move to a headless architecture shows how platforms are moving toward API-first, agent-driven execution.

That progress matters, but it does not solve the full problem.

Enterprises do not run on a single platform. They operate across a network of systems that include CRM, ERP, finance, data platforms, and custom applications. The most valuable use cases for AI agents require them to operate across all of these systems.

Each platform can govern what happens within its own boundaries. None of them govern what happens between them.

The failures we just examined are all examples of that gap. Actions crossed system boundaries, but control did not.

BOAT set the stage. Universal orchestration completes it

Gartner’s business orchestration and automation technologies (BOAT) framework addressed the fragmentation of automation technologies by bringing together capabilities like RPA, workflow, and integration into a unified model.

That was a necessary step forward. But BOAT, introduced by lead author Saikat Ray from Gartner, focuses on connecting systems. It does not fully solve how execution is governed once everything is connected, especially when agents are acting autonomously.

Universal orchestration, also introduced by Saikat, builds on that foundation by adding cross-system control.

BOAT connects the stack, and universal orchestration governs execution across it.

What leaders should focus on now

For business leaders, this is not about deploying more AI. It is about controlling how AI operates across the enterprise.

The key questions are straightforward:

• Where does decision logic live today?
• How are agents coordinated across systems?
• How are policies enforced consistently?
• Can execution be traced end to end?

If those answers depend on individual systems, the organization is exposed to the same failure patterns that are already showing up in production.

Final thought

AI agents do not create risk because they are intelligent. They create risk because they can act across systems without coordination.

Inside a platform, that risk can be managed. Across platforms, it cannot be managed without a control layer.

That is the challenge happening right now.

The problem is not that agents are making bad decisions. It is that no one is governing how those decisions execute across the business.

AI inside a system delivers efficiency. AI across systems delivers transformation. But cross-system execution requires cross-system control.

The companies that succeed in this next phase will not be defined by how much AI they deploy. They will be defined by how well they control it.

How well are you controlling agents across your business? If the answer isn’t clear, you don’t need more AI. You need a control layer.

Interested in learning more? Check out our webinar (June 23) on Orchestrating AI: How Universal Orchestration Changes Enterprise Automation to be ready for the next wave of enterprise AI.