Industry executives and experts share their predictions for 2026.  Read them in this 18th annual VMblog.com series exclusive. 

By Rick Clark, Global Head of Cloud Advisory for UST

The pressure on IT leaders to adopt AI is no longer about experimentation or innovation. It is about survival. Boards are asking where AI fits into operations. Vendors are promising autonomous remediation. Consultancies are pitching AI-native transformation.

Yet most enterprises are running headlong into a constraint they cannot brute-force their way around.

AI cannot operate what it cannot understand, and most cloud environments are architected in ways that make understanding impossible for machines.

This is not a maturity gap. It is not a tooling gap. It is not a model gap.
It is the accumulated result of architectural decisions made over the last decade.

The “Shift Left” era, from DevOps to full-stack development, delivered speed by redistributing responsibility. Teams were empowered to choose their own tools, define their own architectures, and operate their own services. In isolation, these moves made sense. At scale, they hollowed out architectural cohesion.

We built environments that humans can navigate through experience, intuition, and institutional memory, but that are fundamentally incoherent to machines. To move from AI hype to operational reality, enterprises must first repair three foundational architectural failures.

1. High-Fidelity Observability: The AI Nervous System

AI cannot reason about systems through fragmented logs, aggregated metrics, and partial traces. Yet this is precisely how most enterprise cloud environments are instrumented.

Dashboards summarize behavior for human consumption. They hide variance, smooth over causality, and collapse complexity into something an operator can glance at. AI does not need summaries. It needs truth.

If an organization cannot describe normal behavior through high-fidelity, high-cardinality telemetry, AI cannot reliably detect anomalies, predict failures, or understand the impact of change. Without correlated signals across performance, reliability, and cost, AI is reduced to pattern matching without context.

This is why observability is not a tooling discussion. It is architectural.

In systems that are too complex for humans to reason about manually, observability becomes the governing substrate. It is the only objective source of truth shared across teams, platforms, and environments. Without it, both humans and machines are forced to operate on approximations.

Most enterprises believe they have observability because they have dashboards. They do not. They have visualizations layered on top of incomplete data. AI cannot compensate for that gap.

2. Restoring Structural Integrity

The Shift Left era replaced centralized bottlenecks with distributed autonomy, but it did so without restoring architectural authority at the platform layer. Teams optimized locally. Platforms became optional. Guardrails turned into guidelines.

Humans learned to compensate.

They memorize which services are fragile. They know which alerts to ignore. They understand which failures are benign and which are dangerous. They bridge gaps with tribal knowledge and heroics.

AI cannot do this.

AI requires deterministic patterns, stable boundaries, and enforceable contracts. It cannot infer intent from chaos. It cannot safely intervene in systems whose behavior is shaped by undocumented exceptions and one-off decisions.

This is why the prerequisite for AI success is not more intelligence, but more structure.

Enterprises must move away from the full-stack myth, where developers are expected to be accidental operators of infrastructure, security, reliability, and cost. That model scales human burnout, not system integrity.

The alternative is Declarative Intent.

Developers should declare what the business needs. The platform must own how that intent is implemented, governed, secured, and operated. This is not a return to centralized control. It is the re-establishment of architectural responsibility at the layer where it belongs.

Platforms that enforce patterns do not slow teams down. They remove entire classes of decision-making from the critical path. They create environments that are predictable, legible, and safe for both humans and machines.

AI depends on that predictability.

3. Declarative Economics: Ending the Cost Guessing

Cloud economics did not become chaotic by accident. They became chaotic because economics was treated as an operational afterthought instead of an architectural constraint.

Costs are discovered after the invoice arrives. Optimization happens reactively. In response, many organizations attempted to shift financial responsibility onto developers.

This was a fundamental error.

Asking engineers to manage cloud costs is equivalent to asking them to price the business without access to margins, revenue models, or strategic intent. Developers do not know the value of a transaction. They cannot make informed tradeoffs between cost, risk, and outcome. Pushing economics onto them did not create discipline. It created noise.

This was not empowerment. It was abdication.

AI makes this failure impossible to ignore. Optimization without declared unit economics is dangerous. AI will optimize locally and destroy value globally, faster and more confidently than any human system ever could.

For AI to operate safely, cost must become a design constraint, not a reporting artifact.

Enterprises need an Economic Contract where the business defines the value envelope and the platform enforces it automatically. When economics is declarative, cost stops being guessed at and starts being governed. The platform becomes an active governor, not a passive observer.

This is architectural work. AI cannot retrofit it after the fact.

Conclusion: Readiness Before Adoption

Organizations that rush to adopt AI before repairing these foundations will not become more efficient. They will become more fragile. Unstable automation, hallucinated remediation, and unpredictable outcomes are not edge cases. They are the expected result of applying AI to incoherent systems.

AI readiness is not an experimental initiative. It is an architectural mandate.

By restoring structural integrity, enforcing declarative intent, and making economics a first-class design constraint, enterprises create the only environment in which AI can safely participate in production operations.

Adoption can be accelerated.
Readiness cannot be skipped.

##