Autonomy without accountability: Why enterprise AIOps stalls at scale

During a major sales event, a global retailer experiences a surge in online demand. The cloud infrastructure scales automatically, security systems tighten controls, monitoring tools trigger remediation workflows. On paper, everything works.

In reality, performance degrades. Customer experience suffers. Costs spike. Engineers from infrastructure, security and application teams end up in the same control room, trying to explain how each system did what it was designed to do, yet still produced the wrong outcome.

This issue isn’t a one-off failure. It’s a pattern many enterprises now recognize.

As organizations introduce autonomous capabilities across their business, systems are increasingly able to act on their own. But decisions are being made inside individual tools and domains, without a shared understanding of intent or tradeoffs across the broader environment.

McKinsey describes the rise of agentic AI as a “transfer of decision rights,” forcing enterprises to rethink accountability when systems act on their behalf. That shift is already underway in IT operations. The question is no longer whether systems can act autonomously; it’s how those actions stay aligned and governed when outcomes span teams, platforms and business priorities.

At the heart of stalled AIOps initiatives isn’t lack of automation. It’s a lack of coordination.

Over the past decade, enterprises have steadily increased operational autonomy through layers introduced across IT domains. These capabilities are now deeply embedded in everyday operations, enabling systems to act faster and with less human intervention than ever before.

Individually, these systems perform exactly as designed, but only within their own domains. A cost-optimization action in cloud infrastructure might conflict with a resilience requirement in customer-facing applications. A security remediation might introduce latency that affects performance objectives.

As environments grow more complex, the coordination gap becomes increasingly visible. We see this play out in three structural scenarios:

Mergers and acquisitions can reshape enterprise architectures overnight. New applications, duplicate platforms and inherited workflows introduce layers of dependency that are rarely harmonized immediately. What once operated within a single stack now spans multiple clouds, vendors and governance models.

Many organizations begin their AI journey in a narrow domain like IT service management or infrastructure automation. Over time, they might expand into security, cloud optimization, data pipelines and application observability.

Each additional domain increases the number of possible interactions between systems, and the complexity multiplies.

Not all systems modernize at the same pace. Some core applications might remain in place for years due to regulatory, cost or operational constraints. These legacy systems often resist integration and introduce brittle dependencies that complicate coordination.

AIOps has advanced quickly, but its evolution has largely emphasized speed and efficiency over alignment.
Broadly, the AIOps evolution has unfolded in three stages:

·      Alert-driven operations focused on noise reduction and event correlation.

·      Insight-driven operations used analytics and, later, generative AI to improve root cause analysis and speed investigation.

·      Agentic operations introduced autonomous action within defined domains.

Each stage brought meaningful improvements, but left a more fundamental question unanswered: how should decisions be coordinated when actions span multiple systems, teams and business priorities?

Enterprise-scale agency is the ability to coordinate intelligent decisions across the enterprise in alignment with business priorities.

To do this, organizations must clearly define priorities that their system can act on, ensure that they share data and context across domains, establish clear boundaries for where autonomy is permitted and maintain full accountability through decisions that can be traced and explained.

This is the role of governance.

As autonomous systems move from insight to action, they begin to influence decisions that carry financial, operational and regulatory consequences. Enterprise leaders remain accountable for those outcomes, even when actions are taken at machine speed across multiple domains.

Industry data reflects this tension. In its State of AI in the Enterprise: 2026 report, Deloitte found that while most enterprises plan to deploy agentic AI within the next two years, only a small fraction report having mature governance models in place. Leaders cite accountability, transparency and risk management as the primary barriers to scaling autonomous systems beyond isolated use cases.

Most leaders aren’t asking whether AI works, leaders are asking who answers for it.

Trust can’t be assumed.  It must be deliberately engineered through transparency and controls that allow confidence to be built and strengthened over time. Without governance, teams are left explaining outcomes they didn’t directly control to regulators, executives and customers.

One of the clearest ways to close this gap is through guardrails.

Autonomy should not be treated as a binary switch between manual and fully automated operations. Instead, mature organizations establish graduated zones that clarify where AI systems can observe, recommend, execute with approval or act independently within clearly defined constraints.

These guardrails encode business intent: controlling risk exposure, defining acceptable tradeoffs and setting escalation paths when conditions fall outside expected thresholds.

Sometimes, limited cross‑domain coordination is allowed, but only within boundaries that control the blast radius and ensure human intervention remains possible when needed. That’s what makes autonomy predictable and what allows organizations to expand it safely over time.

Governance also requires that every autonomous action be explainable.

Trust erodes quickly when decisions cannot be reconstructed. For agentic systems operating in complex environments, leaders need visibility into what action was taken, why it was taken and what data informed the decision.

This requires traceable decision paths, clear data lineage, defined confidence thresholds and explicit escalation triggers. It also requires the ability to measure downstream impact across systems. When decisions are observable and auditable, autonomy becomes accountable in a way that is demonstrated, not assumed.

Technology platforms can provide agents, orchestration engines and integration across complex environments, but software can’t define business priorities or reconcile competing objectives across teams.

Software cannot determine when resilience should take precedence over cost, or when compliance constraints should override speed. And it cannot realign workflows across security, cloud or application teams that operate with different incentives and accountability models.

As a result, organizations that approach agentic AI as a plug-in capability often automate faster without improving coordination. This amplifies fragmentation rather than reducing it.

Agentic AI has moved past being considered just the next feature; it shifts how operations are designed and governed. Real value emerges when technical capabilities are paired with operating model changes, including clearly defined intent, governance frameworks, escalation paths and cross-domain collaboration.

A mature agentic operating model changes how work is divided between humans and machines. Machines execute decisions at speed, scaling resources, resolving incidents and coordinating actions across domains. They are guided by declared business priorities and constrained by clear guardrails.

Humans step out of constant reactive execution and into supervisory roles: setting intent, monitoring outcomes and intervening when conditions fall outside expected bounds.

In practice, this doesn’t happen all at once. Autonomy expands gradually as systems demonstrate reliability and teams gain confidence in how decisions are made and explained. Tradeoffs are evaluated continuously rather than during high‑pressure incident calls.

Accountability is clear, even when actions happen automatically.

In the case of the global retailer, every system acted correctly but the outcome still failed. Accountable autonomy is what would have closed the gap between action and alignment.

IBM’s perspective is that enterprise‑scale agency works only when orchestration, openness and governance are designed together. Drawing on experience in regulated and mission‑critical environments, IBM works with organizations to define operational intent, establish boundaries for autonomy and ensure that decisions can be observed, evaluated and improved over time.

Through integrated AIOps capabilities and advisory‑led transformation, IBM helps enterprises move from isolated automation to coordinated operations, where resilience, efficiency and speed aren’t just optimized locally. They are owned end to end.

Explore IBM Consulting® AIOps