Expectations for artificial intelligence have shifted from precise generative outputs to autonomous agent actions in critical business processes. However, the pace and breadth of enterprise AI deployment have fallen short of expectations. The bottleneck is not model capability, compute access, or a lack of innovation.

What prevents Chief Information Officers (CIOs) from achieving their return on investment (ROI) targets is the constraint imposed by traditional enterprise operating models. Now is the time to rethink this model for the AI era, and trust is the key.

Traditional enterprise operating models have no room for AI

Getting started with AI is not difficult, which is why the world is expected to spend$4.5 trillion on AIthis year. Yet CIOs remain reluctant to hand over entire workflows to agents without human oversight. The challenge is not finding humans to participate in supervision, but determining which decisions can be safely delegated to AI agents. Weighing this issue slows progress, as teams often stall when moving AI pilots to production due to concerns about risk and reversibility. As a result, AI fails to achieve end-to-end process automation, and its ROI reaches only a fraction of expectations.

The reason lies in the fact that typical enterprise operating models revolve around people, processes, and technology. Unfortunately, many organizations view AI merely as another tool in the technology toolbox. MIT has found thatnearly all enterprise AI projects fail. These organizations adopt different vendors and different AI approaches, and although their paths vary, they all fail in the same way. Forcing AI into traditional operating models (people, processes, technology) does not work.

However, the right approach is not to discard the old model, but to extend it by embedding AI natively. Just as enterprises today strategically decide where and how to apply model components to meet needs, adopting an AI-native approach ensures that people, processes, and technology can all effectively leverage AI to achieve goals and deliver results.

Delegating the right resources to the right challenges

CIOs have already invested heavily in systems such as ITSM, SSO, cloud, and DevOps. A large portion of IT budgets goes to systems that keep the organization running. These systems, even when they have APIs, mostly remain in silos. As vendors roll out more AI-enhanced products, these silos only become smarter, not broken down. The persistent gaps between silos still require manual filling, where people stitch together data, gather context, and coordinate with other people and systems to advance processes.

Just as human employees are delegated more strategic tasks, AI must also prove its value and trustworthiness to achieve goals. Truly moving toward an autonomous enterprise requires delegating those "coordination gap" tasks to AI agents. The question facing CIOs is not which AI to choose, but what AI can be trusted to handle. Trust is the decision vector.

Under this framework, risk and responsibility determine which tasks can be delegated to AI, falling into one of three risk-weighted options:

  • Human-led processes: Decisions involving high risk, low precedent, or irreversible elements remain under human control. AI provides assistance, but the final decision-making authority always rests with humans.
  • Human-supervised processes: Lower-risk decisions can be delegated to AI agents, but they must act under human supervision, follow established rules, and log all actions. As AI gradually learns and trust in outcomes grows, the required supervision decreases. Humans govern agent work through overall strategy rather than overseeing each AI decision individually.
  • Agent-led processes: Low-risk, high-volume decisions are fully delegated to AI agents, which think, decide, and act autonomously without human supervision.

This framework overturns the traditional approach to AI scaling. The common current practice is to pilot AI on a single process, declare success, and then push it into production, only to fail due to the near-infinite combinations of processes, data, and decisions. A better path is to start with low-risk, routine operations, validate success, and then gradually elevate AI to higher-risk, higher-stakes tasks. Trust is not granted during pilots but is earned incrementally through successive successes.

Deploying AI agents within the delegation framework

In working with hundreds of enterprises, we have found that about 80% of IT tickets can be resolved autonomously by agents, such as guiding users to the right information or performing simple low-risk operations. This is a blueprint for applying the delegation framework to IT and the entire enterprise: start with routine operations (such as chatbots guiding users to relevant articles, resetting passwords, and guiding software updates), then gradually elevate AI to more strategic work.

Since risk is the decision vector, governance becomes central to build-versus-buy decisions. Deployment speed matters, but fast-tracking to eventual failure is not success. The options are as follows:

  1. Build in-house: Offers full control but is time-consuming and diverts resources from other projects. Scaling security and governance is also challenging, and AI talent is expensive and not a core enterprise competency.
  2. Native AI: Deploys extremely quickly and integrates natively with solutions, but locks context and data within existing silos. Additionally, choosing and maintaining a single vendor's AI across cross-platform automation raises delegation issues.
  3. Hyperscale cloud providers: Easy to adopt, offering powerful AI capabilities and services, but integrations and solutions must be built from scratch. Their resource requirements are similar to in-house builds, and responsibility falls on the enterprise itself.
  4. Specialized enterprise platforms and solutions: Pre-built, customizable, and governance-by-design, combining the speed, control, and integration advantages of other options.

When weighed against the delegation framework, the choice becomes clear. The fastest path to success is a solution that offers full control and can scale across applications and from low-risk to high-risk processes at the enterprise level.