The Agentic Age

Most of us can read a change before we build anything for it. We come in early, look at what's being asked, and get a feel for it. How big is this? How far do people have to move from the way they work today to the way they'll need to work tomorrow? Who's likely to struggle, and where? That reading is our craft. It shapes the size of the effort, the strategy, the plan, and most of the time it's close to right because we've learned to understand change by looking at the people it impacts.

That's the part I want to talk about, because it's the part AI quietly changes.

When we size a change, we're really sizing what people have to absorb. A new system. A new process. A new expectation. We estimate the distance between where people are today and where they'll need to be tomorrow. That's what our assessments are designed to measure.

The challenge is that part of an AI implementation doesn't move through people at all. Some of the work begins happening alongside them, or instead of them. Our assessments are still aimed at the people, so the part of the work that leaves the human path is the part we're most likely to underestimate. It doesn't appear as an adoption gap because there isn't a person standing there to experience one.

A lot of what used to keep that work safe was never captured in a process map either. It was based on the judgment that people developed over the years. The specialist who read the queue before the floor filled up. The manager who knew an absence had context behind it. The employee who picked up the phone instead of sending the standard message because they could tell this situation was different. None of that was a documented step. It lived in the space between the process and the people doing the work. When the work moved, nobody decided to remove it. It quietly disappeared, and nothing in the implementation marked that it was gone.

The resistance often looks different, too. Usually, when people push back, we can trace it. People don't understand why the change is happening. They haven't been trained. They don't see the value. We know how to work with those situations.

For some AI implementations, it’s another matter. People understand the technology. They've completed the training. They know how to use the tool, and they still hesitate. The question isn't whether they're capable of using it. It's about whether they trust it, whether they know when to rely on it, and what relying on it means for their own judgment and expertise. They're working out, often unaided, where their experience still adds value and how much responsibility they're willing to hand over.

Then adoption arrives, the numbers look healthy, and this is where our routine interpretations can fail us. Usage is usually where we start to look for confirmation that a change has taken hold. But usage can be high while something underneath is quietly breaking. The tool is being used, yet the case that once received a second look now moves through the system on its own because using the technology and applying human judgment were never the same thing. Only one of those shows up in the dashboard.

High usage can mean the change succeeded, or it can mean there are simply fewer people left doing the work. The measure doesn't tell us which.

Then we move into reinforcement, because we know change doesn't sustain itself. Reinforcement works by strengthening behavior. You reinforce something a person does. But where the system now performs the work, there may be no behavior left on that part of the workflow to reinforce. The judgment that protected the difficult cases was never identified as needing protection, so it never became part of the reinforcement plan. It isn't reinforced. Over time, it quietly fades away.

In none of this is the practitioner at fault. Our methods are doing exactly what they were built to do. They're focused on the people moving through the change. What's different is that part of the change has moved where people no longer are. Our instruments continue to report on the human side of the implementation, while some of the most important shifts are happening elsewhere. We're not measuring incorrectly. We're measuring the part we can still see.

That's the gap I keep coming back to. Another template won't close it. A checklist only helps once you already know what you're looking for. The harder work comes earlier. It's learning to recognize the parts of the change that have left the human path. It's deciding what has to be preserved and what can safely go. And it's making those hidden forms of judgment visible enough to hold up in a room where people ask what it costs. That isn't another deliverable. It's a different way of reading change, and I don't think our profession has fully developed it yet.

The season ends with Maren working backward from a failure nobody knows how to see until it's already happened. She pieces together a different way of reading change because the methods she already knows can only explain the part of the implementation that still moves through people. The hard part is learning to see what leaves the human path before it quietly carries away the judgment that kept the difficult cases safe.

She builds that way of reading because she has to. We don't.

The failure she uncovers isn't confined to the story. It's the kind of failure more organizations are likely to encounter as AI takes on a larger role in how work gets done. The difference is that it no longer has to be discovered one project at a time, after something important has already been lost.

That's what this season has been building toward. Not simply recognizing that traditional change methods leave part of the implementation unseen, but developing a way to read that missing part. A way to identify the judgment that lives outside documented processes, to tell what can safely change from what still needs human oversight, and to make those hidden risks visible before they turn into costly outcomes.

In the story, the method and the tools didn’t exist. Now they do.