Agentic Engineering is All About Change Management
Your Agentic Transformation Is More Human Than You Think
Right now, many organizations are trying to move toward the new shining goal, usually marketed as Agentic Engineering. For many companies, this is already a strategic goal. More accurately, though, we should call it agentic work, because the principles apply to almost any kind of white-collar work. It can be legal, consulting, research, or any other workflow where work moves through a system and decisions must be applied in practice. In that sense, an agent is an active side in a process. It is something that can do work and even create other agents. Every agent in a given system has input into that system and can also change its state.
But first we still need a clear definition of what an agent is. An agent is a model, whether an ML model, an LLM, or another kind, that doesn't just take input and produce output. Usually these models produce a probability, a decision, text, an image, or video. An agent goes beyond that. It plans and executes actions like calling APIs, gathering files, sending emails, and in the end carrying out an action. It doesn't only make decisions - it applies them in practice. The moment a model starts producing actions, not just outputs, we're already dealing with a different kind of system.
That is where the real shift begins.
The Technical Illusion
The problem is that we usually think about agentic workflows technically - through integrations, MCPs, and orchestration. But the change is bigger than that. The moment you bring AI into a business in this way, you're making a significant change in how that business functions. It is no longer just a tool that helps someone think, write, or analyze. It becomes an active side in a workflow. It takes input from the system, it changes the system, and it applies decisions in practice. Once that starts happening, the business is already moving into a new way of working - whether it fully realizes it or not. An agent does not carry responsibility for its actions. It is not deterministic in the approaches it will take. And it is, by its nature, indescribable in the approach it took to arrive at a decision or a course of action.
That is why introducing agentic workflows is, in practice, a change management challenge. Change management is the structured process of helping people, teams, and organizations move from their current way of working to a new one - with minimal disruption and the highest chance of success in the new paradigm. This is exactly the situation here. The moment agents enter modern work, the organization isn't simply adding technology. It is changing how work is done. If that change isn't recognized for what it is, then changes are still happening - only they are not managed.
That is why agentic transformation is more human than many people think.
But to understand just how human it is, we need to look at what change management actually involves. It is not one thing. It is a set of interlocking challenges, each of which shows up in a distinct way when agents enter the picture.
1. The Awareness Problem
No change can be managed if the people affected by it don't understand that it is happening, or why it matters. This is the first element of the Prosci ADKAR model - Awareness - and it is where most agentic initiatives silently fail before they begin.
In many organizations, the decision to introduce agents is made at the strategy level. A leadership team sees the potential, sets a direction, and funds a technical team to build it. The people whose daily work will be restructured often learn about the change late, through a demo or a pilot that is already underway. They are presented with a finished system and asked to adopt it. This is not awareness. This is announcement.
The distinction matters because awareness is not about information. It is about understanding. When the British textile industry began mechanizing in the early nineteenth century, the workers who resisted - the Luddites of 1811 to 1816 - were not ignorant of the technology. They understood it better than most. What they lacked was any credible explanation of why the change was happening in a way that destroyed their livelihoods without offering a path forward. Their grievance was not with the loom. It was with the absence of a managed transition.
The same failure mode is already visible in agentic transformation. Teams are told that agents will "augment" their work. But what does that mean in practice? Which parts of their workflow will change? Which decisions will still be theirs? What will their day look like in six months? If no one can answer those questions clearly, then awareness hasn't been achieved - regardless of how many town halls have been held.
2. The Sponsorship Gap
Change management research has consistently shown that the single strongest predictor of whether a transformation succeeds is the quality of executive sponsorship. Not the existence of a sponsor - nearly every initiative has one on paper - but the active, visible, sustained involvement of senior leaders throughout the process.
John Kotter's work at Harvard Business School, beginning with his landmark 1995 article and later expanded into his eight-step model, placed this at the foundation. His first step was establishing a sense of urgency. His second was building a guiding coalition - not a steering committee that meets quarterly, but a group of influential people who actively champion the change and remove obstacles as they arise.
In agentic transformations, the sponsorship gap takes a specific form. Technical leaders often sponsor the platform. They secure budget, hire engineers, and build the infrastructure. But the change that needs sponsoring is not the platform. It is the reorganization of work that the platform makes possible. That sponsorship must come from the business side - from the people who own the processes, manage the teams, and are accountable for outcomes. When the only visible sponsor is the CTO, the rest of the organization reads a clear signal: this is a technology project, not a business transformation. And they respond accordingly - with polite indifference.
3. Resistance as Information
Every framework for managing change acknowledges resistance. Most treat it as an obstacle to overcome. The more honest ones treat it as information to interpret.
When people resist the introduction of agents into their workflows, they are rarely resisting the technology in the abstract. They are responding to something specific. Sometimes it is a loss of autonomy - tasks they once owned are now delegated to a system they don't control. Sometimes it is a loss of identity - if the thing that made them valuable was their judgment in a particular domain, and an agent now exercises that judgment, then what is their role? Sometimes it is a legitimate concern about quality - they have seen the agent produce confident, plausible, and wrong outputs, and they don't trust it with consequential decisions.
Each of these is a different kind of resistance, and each requires a different response. The loss of autonomy is an organizational design problem. The loss of identity is a psychological one. The concern about quality is a governance one. Treating them all as "resistance to change" and responding with more communication or more training misses the point entirely.
William Bridges, whose work on transitions has shaped the field since the 1980s, drew a distinction that is especially useful here. Change, he argued, is situational - a new system, a new process, a new structure. But transition is psychological - the process people go through as they let go of the old way, move through a period of uncertainty, and eventually internalize the new way. Organizations can mandate change. They cannot mandate transition. And when they try, they get compliance without commitment - people who use the new system while quietly maintaining shadow workflows that preserve the old one.
In agentic transformation, shadow workflows are not just inefficient. They are dangerous. If a team is nominally using an agent to triage incoming requests but a senior analyst is quietly re-reviewing every decision because they don't trust the system, then you have the worst of both worlds: the cost of the new system and the overhead of the old one, with neither operating effectively.
4. The Identity Question
Of all the dimensions of change management, identity is the one most consistently underestimated - and the one that matters most when agents enter knowledge work.
Knowledge workers derive professional identity from what they know and what they can do with that knowledge. A senior paralegal's value lies in their ability to read a contract and identify risk. A financial analyst's value lies in their ability to interpret data and form a judgment. A researcher's value lies in their ability to synthesize sources and construct an argument. When an agent begins performing these same tasks - not perfectly, but well enough to handle the routine cases - the question that surfaces is not about efficiency. It is about meaning.
This is not new. When the mechanized loom entered the textile industry, it did not merely replace labor. It dismantled a craft identity that had been built over generations. The croppers of Yorkshire were not unskilled workers. They were artisans whose expertise lay in finishing cloth to precise standards. The machine didn't need that expertise. The social consequences were not technical - they were existential.
In the context of agentic work, the identity question takes a specific shape: if I am no longer the person who does this work, then who am I in this organization? The answer cannot be "you oversee the agent," because supervision is not an identity. It is a task, and often an unsatisfying one. The organizations that navigate this well will be the ones that can articulate a genuine answer - one that redefines expertise around the things agents cannot do: setting direction, making value judgments, managing ambiguity, and holding relationships that require trust.
5. Governance Before Architecture
One of the most underappreciated aspects of change management is the role of governance - not in the compliance sense, but in the sense of clearly defining who decides what, who is accountable for outcomes, and how exceptions are handled.
In traditional workflows, governance is often implicit. A senior team member reviews a junior team member's work. A manager approves an expenditure above a certain threshold. A legal counsel signs off on a contract before it is sent. These structures are informal, embedded in relationships and organizational culture, and they work precisely because everyone understands them without needing to articulate them.
Agents break this implicit governance. When an agent drafts a client communication and sends it for review, who is reviewing it - and against what standard? When an agent triages support tickets and escalates only the ones it deems high-priority, who is accountable for the tickets it didn't escalate? When an agent negotiates a preliminary price with a supplier based on parameters it was given, who owns the outcome if those parameters were incomplete?
These are not edge cases. They are the central questions of agentic work, and they must be answered before the system is deployed - not after the first failure. The history of enterprise software is littered with cautionary examples. The massive ERP implementations of the 1990s - SAP, Oracle, PeopleSoft - failed not because the software didn't work, but because organizations hadn't resolved the governance questions the software forced into the open. When departments that had operated independently for decades were suddenly required to share definitions, processes, and accountability, the result was not efficiency. It was conflict. The integration was technical. The failure was organizational.
Agentic systems introduce the same governance pressure, with an added complication: the agent's reasoning is opaque. You cannot audit an agent's decision the way you can audit a transaction in an ERP system. You can see the inputs and the outputs, but the path between them is probabilistic, context-dependent, and often irreproducible. This means that governance for agentic work must be designed differently - less around tracing decisions and more around defining boundaries, thresholds, and escalation paths in advance.
6. Communication as Architecture
In most transformation efforts, communication is treated as a support function. There is a change management team that produces newsletters, FAQs, and talking points. This is necessary, but it is not sufficient - and in agentic transformation, it is especially inadequate.
The reason is that agentic work changes the communication structure of the organization itself. When an agent sits in the middle of a workflow - receiving inputs from one team, producing outputs for another, escalating to a third - it becomes a node in the communication network. But unlike a human node, it does not participate in the informal communication that holds organizations together. It does not mention in a hallway conversation that a client seemed unhappy. It does not flag in a standup meeting that a pattern of requests looks unusual. It processes what it is given, according to the logic it has, and it moves on.
This means that organizations adopting agentic workflows must deliberately redesign their communication architecture to compensate for what agents cannot do. The informal channels that used to carry critical information - the aside after a meeting, the quick message to a colleague, the intuition that something feels off - must be formalized, at least partially, because the agent will not carry them.
The sociologist Diane Vaughan documented something analogous in her study of the Challenger disaster. NASA's formal communication channels had flagged the risk of O-ring failure in cold temperatures. But the organizational culture systematically normalized the deviation. Information existed in the system. It simply could not travel to the place where it would have changed the decision. The failure was not informational. It was architectural. The same risk applies when agents absorb workflow steps that previously carried implicit communication: the information that used to travel with the work stops traveling, and no one notices until something goes wrong.
7. The Capability Shift
Change management has always recognized that people need new skills to work in new ways. What is different about agentic work is the kind of skill that matters.
In previous technology transitions, the capability gap was usually about operating the new system. When spreadsheets replaced ledger books, accountants needed to learn how to use software. When email replaced memos, professionals needed to learn a new communication medium. The skill was concrete and teachable.
The capability required for effective agentic work is less concrete. It involves knowing how to decompose a problem into components that an agent can handle. It involves understanding what an agent is likely to do well and where it is likely to fail. It involves reviewing probabilistic outputs with appropriate skepticism - neither dismissing them reflexively nor accepting them uncritically. And it involves a form of meta-cognition: the ability to think about your own workflow as a system of tasks, some of which can be delegated and some of which cannot.
This is a fundamentally different kind of training. It is not about learning to use a tool. It is about learning to work alongside a collaborator whose strengths and limitations are unlike anything most professionals have encountered. The closest historical analogy may be the introduction of junior professionals into senior workflows - the transition from doing the work yourself to delegating it to someone who will approach it differently than you would, and whose judgment you must calibrate over time. Except that in this case, the junior colleague has no memory between tasks, no career incentive to improve, and no ability to tell you when your instructions are unclear.
8. Sustaining the Change
The final and most neglected dimension of change management is reinforcement - ensuring that the new way of working persists once the initial energy of the transformation fades.
Kotter identified this as the most common point of failure: organizations that declare victory too early, before the change is embedded in culture, incentives, and daily practice. The initial deployment of an agentic workflow is not the end of the transformation. It is the beginning of a period in which the organization must actively sustain the new way of working against the gravitational pull of old habits.
When the economic historian Paul David studied the electrification of American factories, he found that the full productivity payoff arrived roughly thirty years after the technology was available. The lag was not because the technology was immature. It was because organizations needed a generation to truly reorganize around what electricity made possible - to rethink floor plans, retrain workers, redesign processes, and build new management practices around a fundamentally different way of powering work. The organizations that electrified early but didn't reorganize gained little. The organizations that reorganized deeply gained everything.
Agentic transformation will follow the same curve. The early adopters who deploy agents without reorganizing their work will see modest gains. The organizations that treat the agent as the beginning of a deeper change - one that touches roles, governance, communication, identity, and culture - will see transformative ones. But that deeper change takes time, sustained attention, and a willingness to manage a process that is, at every stage, more human than technical.
The Real Work
The organizations that will succeed in this transition are not the ones with the best models or the most sophisticated orchestration layers. They will be the ones that recognize what is actually happening: a fundamental change in how work gets done, who does it, and how decisions flow through the system.
They will invest in awareness before deployment. They will secure sponsorship from the people who own the work, not just the people who build the platform. They will treat resistance as a diagnostic signal, not a barrier to overcome. They will confront the identity question honestly. They will design governance structures before they design agent architectures. They will rebuild communication channels to carry the information that agents cannot. They will train for judgment, not just for tool use. And they will plan for sustainability, knowing that the hardest part of any change is not starting it - it is making it stick.
The technology is the catalyst. The change management is the work.
Every major technological shift in history has followed the same pattern: the technology arrived fast, and the human reorganization arrived slow. The gap between the two is where value is lost, trust is broken, and transformations fail. Agentic engineering will not be the exception. The question is whether, this time, we manage the gap - or simply fall into it again.