Doug Shannon is a global leader in digital transformation, specializing in AI, GenAI and intelligent automation.
Enterprises should own what makes their business unique while remaining free to use whichever AI model best fits the work.
For the last few years, I have heard the same questions from enterprise leaders: “Which model should we use? OpenAI? Anthropic? Gemini? Should we move to the newest one, and what happens if something better comes out next month?”
My answer has increasingly become the same. Stop trying to pick the model that will still be winning a year from now. By the time a large organization evaluates one, runs it through security, negotiates access and puts it into production, another model is already getting the headlines.
Waiting for the next model is not an AI strategy. I recommend building the enterprise so the model can change without forcing the business to change with it.
The Technology Is Rarely The Hardest Part
I have spent more than 20 years working across infrastructure, M&A, automation, machine learning and AI. Earlier in my career, I worked in environments with tens of thousands of systems. Later, automation put me directly into the processes underneath them. That work taught me that the technology everyone sees is rarely the hardest part. The real work starts when you connect technology to how people actually get things done.
I remember sitting with people while documenting processes for automation and finding that the process on paper was not really the process at all. Someone had a spreadsheet nobody else knew about. Another person knew an exception that happened every month but had never documented it. Sometimes the person doing the work could not explain every decision because experience had made those decisions almost automatic.
The Real Process Lives Between The Boxes
That experience shaped how I look at processes. Most processes are not really processes. They are workflows dressed up as processes. The boxes and arrows show the steps, but they rarely capture the small decisions people make between them. When do I wait? When do I question something? Whom do I call? Which exception is normal? When does something look right on paper, but experience says otherwise?
Those micro-decisions are easy to overlook because we make them without thinking. You probably do it every time you drive. You know when to ease off the accelerator without calculating the exact distance to the car ahead. Your eyes move between the road and mirrors without someone telling you when to look. As you enter a turn, you position the car, adjust the wheel and look toward where you want to go. Most of us could not describe every small decision we just made. We would simply say we drove the car.
The same thing happens inside an enterprise. An employee pauses an invoice because something looks unusual. A manager knows who can resolve an issue faster than the official escalation path. Someone recognizes that a customer request meets the rules but needs another look. Much of that never appears on a process map, but remove enough of it and you discover how much of the business people were carrying in their heads.
What The Enterprise Should Actually Own
Once we can see those decisions, we can decide what to do with them. Some belong in the process. Some are institutional knowledge that should be captured. Some can be automated or supported by an agent. Others should remain human judgment.
This is what I mean by owning the enterprise. Keep control of your data, context, processes, identity, permissions, institutional knowledge and decision history. Bring outside intelligence to that understanding instead of moving the understanding outside with it.
I use AI this way myself. I use it to co-think, challenge an idea, find something I may have missed and move faster through work I already understand. My experience provides the context, and I am responsible for questioning what comes back. AI helps me go further, but I do not hand it the wheel.
Model Agnostic Does Not Mean Model Indifferent
Enterprises can work the same way. Being model agnostic should mean more than swapping one API for another. Know what information needs to leave the organization, how much context needs to travel with it and whether it needs to leave at all. Some work can remain internal. Other work may benefit from a frontier model. If something safer or better comes along, use it without rebuilding around another provider.
I expect this approach to extend to agents. Enterprises may eventually maintain marketplaces of agents that are verified, tested, vetted and approved for particular departments. A finance agent should not have the same access as one used in manufacturing or customer service. Know which Model Context Protocol (MCP) servers they connect to, which command-line interfaces (CLIs) they can invoke, which systems they can touch and what authority they have.
Current OpenAI research shows people are delegating longer and more complex work to agents, including researchers working with multiple agents concurrently. But we learned years ago not to let software move freely through an enterprise simply because somebody found it useful. We should not forget that lesson now that software can reason.
Employees Are Still The Source Of Context
My years in automation also taught me that employees compensate for technology long before technology compensates for them. They work around broken processes, disconnected systems and missing information every day. Now we are asking those same people to understand technology that changes almost weekly.
That is why I treat employees as the first customers of AI. You need to enable them to understand it, empower them to use it and embolden them to question it. They carry context the model does not have, while AI can help them see beyond what they already know.
My recommendation is straightforward: Understand the micro-decisions that make your business work. Keep your context and institutional knowledge close. Enable, empower and embolden the people who carry it. Build enough separation from the model that you can use the best intelligence available without rebuilding your business around it.
The models will change. Your enterprise should not have to reinvent itself every time they do. That is what I mean by own the enterprise and rent the intelligence.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?