Why the greatest constraint on artificial intelligence is no longer technology, but the organization itself
In May 2026, Gartner published a forecast that, at first glance, seems almost paradoxical. According to its analysts, by 2027 up to 40% of enterprise AI agents will be significantly restricted, placed under human supervision, or removed from operation altogether.
For the past two years, the conversation has focused on the opposite scenario. We have become accustomed to discussing how artificial intelligence will automate professions, reshape the labor market, and gradually replace people in specific functions. For the first time, however, a different possibility is emerging: organizations may soon find themselves “dismissing” their own intelligent agents.
The reason behind Gartner’s forecast is more significant than the number itself. The firm does not attribute this potential reversal to the quality of AI models or technological limitations. The real challenge lies much closer to the Board than to the engineering team. Organizations are beginning to realize that autonomous systems require a fundamentally different management architecture from the one developed over decades for managing people.
Had this conclusion appeared in a single report, it could easily have been dismissed as an isolated hypothesis. Yet remarkably similar signals have emerged almost simultaneously from McKinsey, Deloitte, Teradata, and several other research organizations. Although they examine different industries and issues, they all point to the same underlying reality: companies have learned to deploy artificial intelligence far more quickly than they have learned to transform their own operating models.
From Models to Organizations
The first two years of generative AI were defined by technological competition. Companies competed over model quality, context windows, computing costs, and the speed of product releases. The prevailing assumption was that access to the most advanced models would become the primary source of competitive advantage.
That logic is now beginning to lose its dominance. Most large organizations can already choose between several leading AI platforms, and the differences between models are becoming less important than an organization’s ability to integrate AI into its operating environment.
As intelligent systems evolve from performing isolated tasks to managing entire business processes, executives are being forced to address an entirely different set of questions. Which decisions can autonomous agents make independently? Where should the boundaries of their authority be drawn? Who remains accountable for their actions? And how does an organization maintain control?
This is why the AI conversation is gradually moving beyond technology. It is becoming a conversation about how modern organizations themselves should be designed.
The Illusion of Adoption
McKinsey’s latest research illustrates this shift particularly well. More than 75% of companies now use generative AI in at least one business function, yet only around 1% believe they have successfully integrated AI into their operating model at a strategic level.
This gap is revealing. Most initiatives remain focused on automating individual activities such as document preparation, analytics, marketing, or customer service, while the organization itself changes very little. Business processes remain unchanged, decision rights stay the same, accountability structures are untouched, and corporate data continues to exist in fragmented silos.
The result is an illusion of transformation. AI has entered the company, but the organization has not yet evolved enough to make it part of its business model. We increasingly encounter similar situations in our own practice: executive interest in agentic systems is growing far faster than the readiness of business processes, data, and governance mechanisms required to support them.
The Real Constraint Is Not Intelligence
Research from Deloitte, Gartner, Teradata, and Box converges on another important conclusion. The greatest obstacle to scaling AI is no longer the capability of the models themselves, but the maturity of the organization deploying them.
This is why AI is gradually moving beyond the responsibility of IT departments. It is becoming a leadership issue for CEOs, Boards of Directors, and Chief Risk Officers.
For some companies, the bottleneck is the absence of robust AI Governance. For others, it is fragmented data or the inability to integrate intelligent agents into existing workflows. Regardless of the specific challenge, the conclusion is the same: the next stage of AI adoption will be determined less by technological capability than by the quality of corporate governance.
At the same time, an entirely new discipline is beginning to emerge: Machine Identity Governance. For the first time, organizations must manage not only people but also digital entities that access corporate data, initiate actions, and participate directly in business processes. The governance architecture that was built over decades around human employees is gradually becoming insufficient for this new operational reality.
What Does This Mean for Business?
For years, digital transformation typically began with selecting a technology.
For executives, the central question is gradually changing. It is no longer about choosing the right AI model or technology provider. The more important question is whether the organization itself is ready to work alongside intelligent agents. Are business processes transparent enough? Is data ownership clearly defined? Where should the boundaries of autonomous authority be drawn? And how are decisions governed in a new hybrid operating environment?
This is likely to define the next phase of AI. Competitive advantage will depend less on access to the latest language model and increasingly on an organization’s ability to redesign itself while preserving control, accountability, and operational resilience.
If the past two years were shaped by one central question—how intelligent can AI models become?—the next stage will likely be defined by another: How should an organization be designed when part of its operations is carried out not by people, but by intelligent agents?
