Enterprise AI readiness is getting more complex.

Model capability, data quality, and use-case fit still matter. But as AI moves from answering questions to taking actions, organizations also need to understand the systems surrounding the model.

Who can act?
What can they access?
Where do workloads run?
What happens when a critical provider fails?

That leads to a more important readiness question:

Can the organization maintain control as AI becomes more autonomous and infrastructure-dependent?

Three developments this week show why that question matters.

1. Compute governance is becoming AI governance

NVIDIA confirmed that Jensen Huang resigned as president and CEO, with CFO Colette Kress appointed interim CEO.

Separately, semiconductor research firm SemiAnalysis estimates that more than 1.5 gigawatts of export-restricted NVIDIA AI hardware reached China despite U.S. controls. Tom’s Hardware reported SemiAnalysis’ view that the issue may help explain the timing of Huang’s departure, but that remains speculation. Neither NVIDIA’s announcement nor its SEC filing attributes his resignation to unauthorized chip exports.

For enterprise leaders, the more important issue is not the speculation. It is visibility into the infrastructure powering AI.

Organizations increasingly need to know:

  • where AI workloads run

  • who owns and operates the infrastructure

  • who can access it

  • which jurisdictions and regulations apply

  • how compute capacity is sourced

  • which third parties are involved

This is sometimes referred to as compute provenance: the ability to trace where computing resources come from, where they operate, and who controls them.

That visibility is especially important in regulated, sovereign, and sensitive environments.

AI governance can no longer stop at models and data. It must also extend to the infrastructure that powers them.

2. Agents require more than authentication

United Airlines announced that it is restricting unauthorized third-party AI agents from interacting with its booking and travel systems.

United cited risks involving customer credentials, sensitive information, unintended purchases, account changes, MileagePlus miles, and regulatory requirements.

The important issue is not whether United supports AI.

It is who controls the transaction boundary.

Traditional authentication answers one question:

Who has access?

Agentic systems introduce another:

What is software allowed to do after access is granted?

That requires a stronger control model.

Organizations deploying agents should define:

  • a distinct identity for the agent

  • explicit permissions

  • task and transaction boundaries

  • human approval points

  • complete auditability

  • revocation procedures

  • recovery or rollback paths

An employee authorizing an agent should not automatically translate into unlimited authority for everything that agent can technically reach.

That is why agent readiness requires a higher threshold than chatbot readiness.

3. Cloud dependence needs measurable resilience

Microsoft’s global service disruption affected Exchange Online, Outlook, Teams, Copilot, and other Microsoft 365 services.

Microsoft ultimately restored service and reported no evidence of malicious activity.

The incident reinforces a familiar but increasingly important cloud principle:

High availability does not eliminate dependency risk.

As AI becomes embedded in business workflows, outages at a major platform can affect more than email or meetings. They can interrupt AI assistants, integrations, automated workflows, data access, and operational decision-making.

That exposure is growing quickly.

Gartner forecasts that worldwide AI infrastructure spending will reach approximately $2.4 trillion in 2027, representing about 59% of total global AI spending.

As more capital and workloads move into AI infrastructure, resilience needs to be evaluated alongside capability.

The question cannot stop at:

Can this platform run our AI workload?

Leaders also need to ask:

What happens when it cannot?

Four readiness tests for leaders

1. Identity

Can you identify every human and non-human actor operating across your AI environment?

Agents should have defined owners, permissions, and audit trails.

2. Infrastructure

Can you trace where workloads run, where sensitive data moves, and which providers and jurisdictions are involved?

You cannot govern infrastructure you cannot see.

3. Resilience

Can critical workflows continue when a major platform or integration becomes unavailable?

Fallback paths, dependency maps, and recovery procedures should be tested, not assumed.

4. Control

Can you limit or revoke agent authority quickly?

Autonomous systems need boundaries, intervention points, and a reliable way to stop execution when conditions change.

The takeaway

AI readiness is no longer just about the model.

It now depends on the broader operating environment, including strategy, data, workflows, cloud infrastructure, integration, security, vendor risk, workforce, and governance.

That is why scaling AI before understanding organizational readiness can produce a dangerous result:

More capability and more risk at the same time.

The organizations best positioned for agentic AI will not necessarily be the ones moving fastest.

They will be the ones that know where autonomy begins and ends, understand their critical dependencies, and can regain control quickly when conditions change.

Know your readiness before you scale

CloudBait Navigator evaluates AI readiness across seven connected domains, helping leaders identify strengths, gaps, risks, and sequencing priorities before funding pilots or scaling AI.

For organizations specifically exploring agents, the AI Agent Readiness path examines whether workflows, controls, data, and governance are prepared for systems that can take action.