Despite rapid progress in artificial intelligence, many organizations are still struggling to turn that progress into meaningful transformation. The gap between what AI can do in demonstrations and what it changes inside enterprises is often misunderstood as a technical problem. But increasingly, the limitation is not the capability of the models. It is the structure of the organizations trying to use them.
As Sean Iannuzzi, Global AI Center of Excellence Leader at NewRocket, explains: “AI has not stalled because the technology is limited; it has stalled because most enterprises are still operating on models built around constant human intervention.”
That observation points to a deeper issue. Most enterprises were built around workflows that assume people will interpret information, make decisions, and move work forward step by step. These systems rely on approvals, handoffs between teams, and human review at multiple stages. This structure was designed to ensure control, accountability, and risk management.
AI is being introduced into this environment, but often without changing how the environment itself works.
In many organizations, AI is used to accelerate specific tasks. It can draft documents, summarize reports, assist with coding, or help analyze data. These are real efficiency gains. However, the broader process around those tasks often remains unchanged. Work is still routed through the same approval chains, and decisions are still finalized by people at each stage.
As Iannuzzi notes, “Organizations have deployed AI to accelerate tasks, but they have not redesigned how work actually moves, so people still interpret signals, sequence actions, and resolve every exception.”
This creates a predictable outcome. AI improves parts of the workflow, but the workflow itself stays intact. The result is faster execution of individual steps without a corresponding change in how the system operates end to end. Many companies experience this as productivity gains in pockets of the organization, rather than transformation across the business.
When AI is layered onto existing processes without structural change, it tends to produce incremental optimization. It helps people do the same work faster, but it does not change what the work is or how it flows. Iannuzzi describes this clearly: “That operating model turns powerful intelligence into incremental optimization instead of transformation.”
The reason this pattern is so common is not only technical. It is organizational. Enterprise systems are built around human control points. These control points exist for valid reasons, including compliance, quality assurance, and operational risk management. But they also create a design where human intervention is required by default, even when automation could handle parts of the process safely.
As a result, AI is often positioned as an assistant rather than an active participant in workflows. It generates output, but humans remain responsible for moving work forward at every stage. This limits the degree to which AI can reshape how organizations operate.
A more transformative approach requires changing how work itself is designed. Instead of inserting AI into existing steps, organizations would need to rethink those steps entirely. That includes deciding which decisions require human involvement and which can be handled within governed AI systems.
In this model, AI would not just support work. It would participate in it. Systems would be designed so that AI can act within defined boundaries, learn from outcomes, and coordinate with other tools and processes. This is what Iannuzzi refers to when he says: “Real value emerges when work is architected for autonomy, where digital teammates can act, learn, and collaborate as governed contributors within the system.”
This shift is not simply about technology adoption. It is about redesigning how work moves through an organization. That includes rethinking roles, accountability structures, and the points where human judgment is truly necessary.
Without that redesign, most organizations will continue to see the same pattern. AI will improve efficiency at the task level, but the overall system will remain largely unchanged. As Iannuzzi concludes, “Until the architecture of work changes, AI will continue to make existing problems faster rather than fundamentally different.”
The opportunity, then, is not just to adopt AI tools but to reconsider how work is structured in the first place.
For leaders and teams exploring AI today, the next step is clear. Move beyond task automation and ask a harder question: which parts of your workflow still depend on human intervention by design, and which parts could be re-architected for autonomy?
Start there, and AI stops being a productivity tool. It starts becoming a redesign of how work gets done.
