Shadow AI is hiding
in your processes.

There’s a strange contradiction happening inside businesses right now.

We’re giving individuals access to some of the most powerful technology ever created, encouraging them to become more productive, automate their work, and make better decisions.

And somehow, we’re surprised when we start losing visibility into what’s actually happening across the organization.

Most conversations about Shadow AI focus on security. Employees putting sensitive information into ChatGPT, using unapproved applications, or building automations that IT doesn’t know exist.

Those are legitimate concerns. But I think we’re missing something much bigger.

“The real danger of Shadow AI isn’t what employees are putting into AI tools. It’s what organizations are losing between them.”

AI is an enterprise-level solution, even if you don’t have an enterprise-level team implementing it.

You can have five employees or fifty thousand. The moment AI starts influencing how work gets done, how information moves, or how decisions are made, you have an organizational issue to solve.

And here’s where it gets interesting.

The more useful AI becomes, the harder it is to see.

It starts small:

• Someone in sales uses it to prepare proposals.
• A project manager uses it to summarize development updates.
• A developer creates an agent to automate testing.
• A department head builds an assistant to turn project reports into executive summaries.

Unlike traditional enterprise software, AI doesn’t announce itself as a new system. None of these decisions require a meeting, a purchase order, or even a conversation with IT.

And frankly, most of these people are doing exactly what we’ve encouraged them to do: finding ways to work smarter.

Now imagine how all those individual improvements interact. Think about how information moves through your organization.

Picture a pyramid.

At the bottom are the people doing the work. They know every task, every challenge, every decision, and every workaround. Above them are the managers, who need to understand progress, blockers, and whether the work is on track. Not every detail.

Above those managers are department leaders, looking at milestones, resources, and outcomes. And at the top are executives who want a relatively simple answer: Are we on track?

Organizations have always worked this way. Every level translates information into something useful for the level above it.

But let’s be realistic about human nature. At almost every level, we’re trying to answer the same question:

“What’s the minimum amount of information I need to share so the person above me feels comfortable that everything is under control?”

Not because we’re hiding anything. We don’t want to overwhelm people. We want to communicate efficiently, demonstrate progress, and get back to our work. We’re busy.

Now let’s look at the Shadow AI cycle.

Every person in that chain did their job. Every AI tool worked exactly as intended.

And yet the executive could be further from understanding what’s actually happening than ever before.

We may be simplifying the process, but we’re also automating the loss of context. Every handoff creates an opportunity for information to disappear.

And eventually, the executive hears that things are going well.

Now imagine each of those translations happening through different AI systems, each operating with its own instructions, available information, and definition of success.

Nobody necessarily lied. Nobody intentionally withheld information. And no AI system had to hallucinate. The systems simply optimized for the tasks they were given: the developer’s agent was trying to complete work, the manager’s assistant was trying to create a useful report, the department head’s AI was trying to communicate progress, and the executive’s assistant was trying to provide clarity.

But who was responsible for maintaining continuity across all of them?

And therein lies the problem. As AI becomes more capable, the disconnect becomes more difficult to recognize.

We spend an enormous amount of time discussing whether AI can complete individual tasks. We spend far less time asking whether those tasks remain connected to the broader organization.

Everything looks increasingly organized. The reports get better. The dashboards look cleaner. The summaries become more concise. All while the underlying work is becoming more fragmented.

This is the part I find most concerning. An organization may have a perfectly reasonable AI policy. It might approve certain tools, restrict access to sensitive information, and establish guidelines for employees. Which is important.

But knowing which applications people use doesn’t necessarily tell you how AI is changing the way the business operates.

• Who created the automation that moves information between departments?
• What assumptions does the agent use when deciding whether a project is on track?
• Where did the executive summary come from?
• Can someone trace a recommendation back to the original work, data, and decisions?
• When something goes wrong, does anyone know where to look?

These aren’t questions reserved for large enterprises with sophisticated technology departments. They’re questions for every business using AI.

AI needs to be open by design.

At Waifinder, we spend a lot of time thinking about what happens between the systems, people, and processes that make a business work.

“AI should make organizations more connected, not simply make individuals more productive.”

That doesn’t mean everyone needs access to every detail. An executive shouldn’t have to read development logs to understand whether a project is progressing. But they should be able to follow the information backward when they need to.

A one-line executive summary should connect to the milestones underneath it. Those milestones should connect to the work being performed. And that work should connect to the underlying systems, decisions, and people responsible for it.

The point isn’t to eliminate the pyramid. It’s to make sure the pyramid doesn’t become a game of telephone.

That’s what Open by Design means to us: building AI solutions where visibility, context, and collaboration are part of the architecture, not features someone tries to bolt on after implementation.

It’s also why we believe organizations need a shared foundation for how systems, information, and AI agents work together. You can have different assistants serving different people. In fact, you probably should. But they shouldn’t all be operating from completely different understandings of the business.

Individual productivity is not the same as organizational progress.

I don’t think the answer to Shadow AI is to tell employees to stop experimenting. Quite the opposite. We want people to find new ways to solve problems. We want developers building agents, managers automating reporting, and executives getting better information.

But we have to stop assuming that making everyone individually more efficient automatically makes the business more effective. Because it doesn’t.

The real opportunity with AI isn’t simply helping each person do their job faster. It’s helping the entire organization understand what is happening, why it’s happening, and what should happen next.