The System Doesn’t Stay Still

We tend to think about introducing AI into an organization as though the organization itself stays still.

It doesn’t.

The technology changes the work. The work changes what people contribute. That changes decisions, expectations, capabilities, and eventually the organization itself. Meanwhile, the technology keeps changing too.

Each change affects what happens next.

W. Edwards Deming spent his career teaching organizations to understand variation, and he drew a distinction that matters here. Some variation is common cause, produced by the system itself, inherent to how the work operates. Other variation is special cause, a signal that something specific has gone wrong.

His warning was about confusing the two. Reacting to common cause variation as though it were a defect makes the system worse. He called it tampering. The organization adjusts, overcorrects, and adds instability to a system that was behaving as designed.

Generative AI produces variation.

The same prompt returns different responses. A person challenges what comes back, adds context the model didn’t have, takes the work somewhere unexpected. That changes the next exchange, which changes the one after it.

That variability isn’t a defect. It’s a property of the system, and some of the value depends on it.

Organizations have spent decades getting very good at eliminating variation, for good reason. Standardized processes, defined roles, controls, decision rights. Consistency is what allows an organization to operate at scale.

Applied to generative AI, that same instinct becomes tampering. Constrain the variation and you remove the thing you deployed it for.

Not all variation is useful. Some of it is error, inconsistency, and slop, and there are places where constraining it is exactly right. The judgment is knowing which kind you are looking at, and that depends on understanding the conditions you are operating in.

The Cynefin framework, developed by Dave Snowden, is useful here. In a clear domain, cause and effect are predictable, so you follow the process. In a complicated domain, cause and effect exist but require expertise, so you analyze and determine the right response. In a complex domain, cause and effect can often only be understood after the fact and the conditions keep shifting, so you probe, observe, and adapt. In chaos, the relationship isn’t discernible at all, and waiting for information makes things worse, so you act to establish enough stability to begin making sense of it.

An organization occupies all four at once. The same company in the same week is running payroll, working a tax position, deploying AI into knowledge work, and responding to something at two in the morning that nobody planned for.

What has changed is the distribution.

Technology, geopolitics, institutions, climate, and markets are all changing at the same time and interacting as they change. AI accelerates that further, altering not only what an organization can do but how quickly its own conditions shift underneath it. More of the work that used to sit in clear and complicated has moved into complex. Chaos arrives more often than it used to.

The problem is that most organizations only have one mode.

When uncertainty increases, the instinct is to produce more certainty. More analysis, more controls, more approvals, more detailed planning. Those are complicated-domain responses, and they are excellent when the problem is complicated.

Applied to a complex problem, they don’t work, and the failure is hard to see from inside. The organization isn’t standing still. It is analyzing and planning with real intensity, waiting for a clarity that isn’t going to arrive, while the conditions continue to move.

That is what exhausts people. Not the uncertainty itself, but working extremely hard in a mode that cannot resolve it.

A complex domain requires something else. People need enough room to sense what is changing, interpret it, try something, learn from what happens, and respond again.

That is where agency becomes essential. Not agency as freedom from constraints, but the ability to act intelligently within conditions that keep moving.

Organizational agency extends that beyond the individual. Recognizing which domain you are standing in is a belief. Being able to respond differently is a structure. Most organizations have neither.

Tightening control when conditions are uncertain feels like the safe choice. It is often what leaves an organization least able to respond.