«Avoiding »Pilot Purgatory": Activate AI Agents in a Controlled Manner
Many companies are debating which AI model is best—but not focusing enough on the right technical foundation. Sophie Riedel, Partner and Managing Director for Switzerland at Zoi, explains why a connected infrastructure is crucial for ensuring that AI agents can not only observe but also take action.
In many companies, discussions are currently focused on the best AI model. Very few, however, are talking about the right foundation. A smart model will not become an AI agent that acts with the same level of intelligence if the infrastructure remains isolated.
The agents remain on the threshold of becoming expensive observers, much like employees who are not allowed to act—who, figuratively speaking, have their hands tied behind their backs. They can see the problem, but they cannot fix it. Yet this is precisely where it is decided whether a company remains stuck in «Pilot Purgatory» or succeeds in making the leap to becoming an Agentic Enterprise.
From Observer to Actor
An example illustrates the problem: An agent on the shop floor reports a material shortage. In a traditional on-premises system, operations would come to a standstill because the agent would be unable to assess supplier risks, compare terms, or place orders. All of this data would be beyond the agent’s reach. The potential, however, is much greater, because an AI agent can do far more than just observe and report—it can even do more than just suggest solutions.
In a connected system, the same agent in a cloud-native setup forwards the request to a procurement agent. The agent hasn’t become any smarter; it simply now has the capacity to take action in such situations. It can act rather than react, and this is precisely where the potential of the Agentic Enterprise lies. Companies structured in this way can respond more quickly and comprehensively to a challenge or bottleneck. Orders can be processed more efficiently, and customer relationships can be further developed.
Architecture Over Hype
Ultimately, the motto is «Architecture over Hype.» The foundation has to be right; otherwise, even the best model will turn into a bureaucratic monster. If a company programs and builds a separate interface to SAP, CRM, or internal tools for every new use case, it starts from scratch every single time. And eventually, the IT department will be suffocated under the mountain of accumulated technical debt. It will be trapped in its own trap, which becomes increasingly difficult to control and manage.
The solution to this dilemma lies in a central AI gateway that—once set up—is used by every new agent as a central docking station. The advantages include easy administration and the assurance of maintaining full control over operations and AI costs.
Governance Structures Are Still Rarely Resilient
The benchmark study «Generative AI in Large Enterprises» highlights precisely this challenge. The decision-makers surveyed reported that 60 percent of productive setups have robust governance structures and gateways in place. When it comes to AI governance, it’s also evident that while many companies have broad rules and compliance frameworks in place, robust decision-making structures—such as an «AI board» or a «central leadership role»—are significantly less common.
The competitive advantage offered by AI does not lie in the choice of model; it lies in the surrounding infrastructure. The data foundation is not merely a backend decision; it is the difference between an AI observer and a truly autonomous AI agent.
