When it comes to AI adoption, the key issue is often not resistance to AI or a lack of tools, budget, or ambition: it’s ownership.
When leaders encourage employees to experiment with AI tools (as they should), employees naturally find ways to make their work easier. They build AI-powered workflows or agents that improve how things work, and these workflows often become part of how the business runs without people realizing.
To begin with, these ‘shadow operations’ are usually not a problem. An AI workflow or agent might be a quick fix for one person or one team. But the moment other people start depending on it, something changes. What began as an experiment has evolved into important organizational infrastructure. And it is often without clear ownership or a plan for when something goes wrong.
That gap is becoming increasingly important. Research from Make’s Mid-Market AI Adoption Journey found that 88% of organizations now use AI in at least one business function, yet most remain in experimentation or piloting rather than operating at scale. AI use is already widespread, but the operating model around it is still catching up.
When agentic automation becomes infrastructure
When leaders tell me that their teams are using AI well, I tend to ask one simple question: ‘if the employee who built the workflow or AI agent your teams rely on left tomorrow, what would happen to it?’ All too often, the answer is some variation on: ‘We’ll figure it out.’ That is today’s AI adoption gap in a nutshell: it’s not about budget or tools or AI fears, it’s about ownership over workflows being formalized.
The answer is not to drag every automation and AI agent back into IT. Ownership should sit close to the business process, with the people who understand it best, while central teams provide the enablement, visibility, security and guardrails needed to scale.
The tipping point is when other people start depending on a workflow. If its failure affects someone beyond the person who built it, informal ownership is no longer enough: it needs a named owner, clear accountability and a plan for what happens when something goes wrong.
A mid-sized problem
For mid-sized companies, the problem is particularly acute. They are often large enough to have multiple teams, recurring processes and real dependencies between functions, but may not have the transformation infrastructure or specialist teams of a large enterprise. That means AI workflows and agents can spread quickly, and the consequences can be significant when nobody knows who owns them.
But that same position is also an advantage. Mid-sized businesses are large enough for improvements to recurring workflows to make a meaningful difference, while often being small enough to make decisions, run pilots and change processes faster than a large enterprise. The challenge is preserving that speed as AI adoption becomes more operational and widespread.
Our research found that only 45% of organizations report having a formal AI strategy. That is where the gap starts to show. Experimentation is moving faster than the structures companies have in place to manage it.
The aim should not be to add overly complex governance to every experiment from day one. That would kill exactly the kind of bottom-up innovation companies should be encouraging. Instead, leaders need to recognize the subtle point at which an experiment becomes a process that needs ownership.
Visibility before control
The best response is not another heavy layer of governance. It is visibility. The strongest approach combines bottom-up experimentation with enough central insight to understand what is becoming important. That could be a hybrid model: teams generate and prioritize ideas locally, while common standards and guardrails provide structure and consistency across the organization.
Leaders therefore need to ask some basic questions. What AI-enabled workflows and agents are actually running? Which automated processes have become business-critical? Who owns them? Who else depends on them? What data do they touch? And what happens when they fail?
As organizations adopt more complex AI and agentic workflows, this becomes harder for any one person to keep track of. Eventually, no individual can hold the entire automation landscape in their head. That is where visibility really matters and, as an old boss of mine used to say, “if you can’t see it, you can’t manage it.”
Out of the shadows
When employees create their own AI-enhanced workflows, that should be welcomed and fully embraced. The danger comes when those experiments become critical business infrastructure without the organization noticing.
The next phase of AI adoption is about bringing those workflows into the light: making them visible, giving them clear owners and applying reasonable governance as their importance grows. Otherwise, leaders may discover the hard way just how much the business has come to depend on something they cannot easily see.
