AI expectations have moved from pinpoint generative output to autonomous agentic action across mission-critical processes. Yet, enterprises are not deploying AI as quickly or as widely as predicted. The bottleneck is not model capabilities, access to compute, or a lack of innovation.
What’s keeping CIOs from meeting ROI goals is that their hands are tied by the legacy enterprise operating model. It’s time to rethink it for the AI era and trust is the key.
The enterprise operating model has no place for AI
Getting started with AI is easy, which is why the world is expected to spend $4.5 trillion on it this year. Yet, CIOs remain reluctant to hand off entire workflows to agents without keeping humans in the loop. The challenge is not finding humans to fit into the loop; it is determining which decisions can be trusted to AI agents. Weighing this question slams the brakes on progress as teams push AI pilots into production and are blocked by concerns over risk and reversibility. Instead of automating an end-to-end process, AI tops out at a small fraction of the expected ROI.
The reason is that the typical enterprise operating model revolves around people, processes and technology. Unfortunately, too many organizations are treating AI as simply another tool in the technology toolbox. MIT found that nearly all corporate AI initiatives fail. Those organizations are taking different approaches with different vendors and different AI solutions, yet are all failing despite the divergence. Jamming AI solutions into the legacy operating model—people, processes, technology—is not working.
The move, however, is not to throw out the old operating model, but to expand it by incorporating AI into the model natively. And, just as enterprises now strategically determine where and how to apply the model components to a need, taking an AI-native approach ensures people, processes and technology all utilize AI effectively to accomplish goals and deliver outcomes.
Delegating the right resources to the right challenges
CIOs have invested heavily in systems: ITSM, SSO, cloud, DevOps and more. The bulk of IT’s budget is in those systems required to keep the organization running. Most of those systems are siloed, even with APIs. As vendors push more AI-enabled products, those silos just get smarter. They don’t break down. The persistent gaps between silos are still filled by human workers stitching together data, gathering context and coordinating with other workers and systems to push processes along.
Similarly to how a human worker is delegated more strategic tasks, AI has to prove its value and that it can be trusted to accomplish goals. Truly moving closer to becoming an autonomous enterprise requires delegating those “coordination gap” tasks to AI agents. The question CIOs face is not which AI to choose, but what AI can be trusted with. Trust is the decision vector.
In this framework, risk and responsibility determine which tasks can be delegated to AI, fitting one of three, risk-weighted options:
- Human-owned processes: Decisions that involve high risk, low precedent, or irreversible elements remain in human hands. AI assists; humans always make the final call.
- Human-supervised processes: Lower-stakes decisions can be delegated to AI agents that act under human supervision, using approved rules and logging every action. As AI learns over time, trust in the outcomes grows and less supervision is required. Humans govern agentic work by overarching policies rather than meticulous oversight of each AI decision.
- Agent-owned processes: Low-risk, high-volume decisions are entrusted completely to AI agents that think, decide and act without human oversight.
This framework flips the approach to AI scalability. Today, it is common to pilot AI on a single process thread, deem it a success and then doom it to failure as it is rolled out to production and near-infinite process, data and decision permutations. The better approach is to apply AI to low risk, routine actions, prove success and then promote AI up to higher risk, higher stakes actions. Trust isn't bestowed during a pilot—it's earned, one proven success at a time.
Deploying AI agents within a delegation framework
Working with hundreds of enterprises, we have found that approximately 80% of IT tickets can be resolved autonomously using agents to point users to the right information or take simple, low-risk actions. That’s the blueprint for applying the delegation framework across IT and the enterprise: start with routine operations (chatbots that point users to relevant articles, reset passwords and walk workers through software updates) and then promote AI to more strategic work.
Since risk is the decision vector, governance becomes the crux of the build-versus-buy decision. Speed of deployment plays a part, but a fast route to ultimate failure is not success. The options are familiar:
- Build it internally: Offers full control, but takes time and pulls resources away from other initiatives. Security and governance at scale are also a challenge and AI expertise is expensive and not a core competency.
- Native AI: Very fast and native to the solution, but locks context and data inside existing silos. It also creates delegation issues when choosing and maintaining one vendor’s AI for each cross-platform automation.
- Hyperscaler: Easy to adopt and offers compelling AI capabilities and services, but integrations and solutions must be built from scratch. It has many of the same resource needs as building solutions internally and the responsibility falls on the enterprise.
- Purpose-built enterprise platforms and solutions: Pre-built, customizable and governed by design, plus the speed, control, and integration benefits of the other options.
The choice becomes clear when weighed against the delegation framework. The fastest path to success is the one that offers full control to scale enterprise wide, across applications and from low risk to higher risk processes.
