Most enterprises do not have an AI problem. They have a transformation problem.

By 2026, the enterprise AI story is no longer mainly about access. Stanford’s 2026 AI Index reports that 88% of surveyed organizations used AI in at least one business function in 2025. Yet McKinsey’s 2025 survey found only 7% of respondents describing AI as fully scaled across their organizations. The gap between those numbers is the transformation problem.

1. Adoption is broad; organizational depth is not

The most important current distinction is between using AI and changing the enterprise with AI. A team can use a copilot, model API or agent without changing a workflow. A function can run pilots without changing decision rights. An organization can license AI broadly without changing data architecture, performance measures, role design or operating routines.

Those states are not failures. Experimentation is necessary. But they should not be confused with transformation. Stanford’s AI Index shows how quickly use has spread; McKinsey’s 2025 evidence shows how little of that use has become organization-wide scale. Deloitte’s January 2025 enterprise survey adds another signal: more than two-thirds of respondents expected 30% or fewer of their generative-AI experiments to be fully scaled over the following three to six months.

2. The bottleneck moves from capability to configuration

Frontier-model capability continues to improve, but enterprise outcomes depend on a wider system. The model needs usable context. The workflow needs a legitimate place for machine output. Someone must own the decision. Exceptions need escalation. Data quality and access boundaries need to be adequate. Users need to understand when to rely on the system and when not to. Measures need to distinguish activity from value.

This is why a stronger model does not automatically rescue a weak intervention. If the underlying work is poorly defined, the information environment fragmented, or the decision rights ambiguous, greater model capability may simply make the organization generate low-quality work faster.

3. Workflow redesign is becoming the center of gravity

McKinsey’s 2025 research on organizations capturing value from generative AI found workflow redesign to have the strongest relationship with reported EBIT impact among the organizational attributes it examined. The same research showed that fewer than one-third of respondents reported their organizations following most of the adoption and scaling practices tested, and fewer than one in five said they tracked KPIs for generative-AI solutions.

This matters because workflow redesign forces concrete questions that “AI strategy” can avoid. Where does the work begin? What evidence enters? Which steps add judgement and which add delay? What can be delegated? What must be reviewed? Who is accountable for an AI-assisted decision? What changes when the system is wrong? What metric would tell us the intervention is worth keeping?

Those are transformation questions. They connect technology to operating reality.

4. Data and risk are not side constraints

Deloitte’s 2024 enterprise survey found data-related issues caused 55% of surveyed organizations to avoid some generative-AI use cases. Risk-related concerns were also among the leading barriers to deployment. These findings are easy to misread as arguments for more governance. The more useful interpretation is architectural: the enterprise has to make data, controls and accountability part of the intervention rather than bolt them on after a pilot succeeds.

That does not mean every use case needs a heavy control regime. It means control should be proportionate to consequence. A drafting assistant, a maintenance recommendation, a customer-facing agent and an automated financial decision do not carry the same failure cost. Transformation architecture should make those differences explicit.

5. Agents intensify the organizational question

Agents expand the issue because they move AI from producing outputs toward taking sequences of actions. Stanford’s 2026 AI Index reports that agent deployment remained in the single digits across nearly all business functions in 2025. That early state is important. The possibility is expanding faster than mature enterprise operating evidence.

An agentic workflow therefore needs more than a technically successful demonstration. It needs a bounded objective, tool permissions, observable state, escalation logic, failure recovery, human decision rights where consequence warrants them, and evidence that the new configuration is better than the old one.

The unit of design is no longer just the model. It is the Human–AI work system.

6. A transformation problem has a different sequence

If the problem is framed as technology adoption, the sequence often begins with platform selection and then searches for use cases. If it is framed as enterprise transformation, the sequence can begin elsewhere:

1. Locate material friction or opportunity. Start with a consequential workflow, decision, constraint or unmet capability.
2. Establish the baseline. Understand current work, economics, failure modes, information and decision rights.
3. Choose the smallest meaningful intervention. Decide whether the answer is augmentation, automation, redesign—or no AI intervention yet.
4. Architect the work system. Define Human–AI roles, data, integrations, controls, escalation and ownership.
5. Implement and instrument it. Build enough of the real system to observe operational behavior.
6. Use evidence to decide the next commitment. Expand, refine, constrain or stop.

This sequence does not reject AI strategy. It makes strategy testable.

7. Read the evidence at the right resolution

The surveys cited here cover different respondent populations and methodologies. Stanford’s adoption figures are directional; Deloitte’s respondents were already involved in AI adoption. Read together, they do not produce a single measure of enterprise AI maturity.

They do show a consistent pattern: AI use is widespread; enterprise-scale transformation and measurable value are much less widespread; organizational redesign, data, adoption practices and evidence remain material constraints.

8. The enterprise implication

Most enterprises now have access to capable AI. Many have experiments. Increasing numbers have production use cases. The scarce capability is becoming the ability to turn a moving technology frontier into coherent enterprise change without confusing motion with value.

The requirement is dual: execution capability to redesign and implement real work systems, and living research intelligence to keep updating those systems as models, agents, enterprise evidence and organizational understanding change.

The next AI decision should therefore not automatically be another tool decision. It should be a transformation decision: what should change in the enterprise, why now, and what evidence would justify going further?

Open questions

Three questions deserve continued observation: whether agentic systems produce repeatable value outside bounded technical domains; which Human–AI control patterns preserve accountability without recreating all the manual work AI was meant to reduce; and whether organizations that redesign workflows around AI sustain value after the novelty and initial productivity effects fade.