The Real Work of Decentralization
There is a paradox at the center of most AI and digital transformation efforts, that leaders often miss. The more an organization wants to decentralize innovation — to push experimentation closer to the business, to speed up decisions, to give teams real authority — the more work the center must do. Not to control what teams are doing, but to enable them to do it well. Many organizations have this backwards, and the consequences show up in ways that are predictable once you know what to look for.
The announcement of empowerment is not the same thing as the design of it. Without this design, leadership is merely declaring decentralization without building the architecture that makes it functional. As a result, they do not end up with agile, innovative teams. Rather, they end up with teams that are uncertain about where their authority begins and ends, leaders who are anxious about what they cannot see, and a bureaucracy that fills in the space between what was promised and what was built.
The Gap Between Intent and Structure
Most organizations pursuing AI or digital transformation describe the same set of goals. They want experimentation that happens faster and closer to the problems. They want decisions made by people who understand the work, not routed up through layers that slow everything down. They want innovation that is distributed across the business rather than owned by a central function.
The operational reality in most of those same organizations tells a different story. Decision rights remain concentrated. Risk tolerance is managed at the top. Approvals that were supposed to be eliminated have simply migrated to new bottlenecks. Teams that were told to move are still waiting — not because they are resistant, but because nobody told them clearly what they were authorized to do.
This is not primarily a culture problem, though culture gets blamed for it frequently. It is a design problem. The organization signaled what it wanted without building the structures that would make it possible. And teams, absent clear parameters, do what any reasonable group of people does under ambiguity: they default to caution and wait for signals that never come.
Three Failure Modes Worth Naming
The gap between decentralization as intent and decentralization as design produces three distinct failure modes. They are worth examining separately because each one has a different cause and a different remedy, even though they tend to compound each other in practice.
The first is responsibility without clarity. Teams receive the mandate to innovate with AI, run pilots, experiment, but they do not receive the map. What risks are acceptable? What data can be used, and under what conditions? Which decisions require review, and which do not? Without answers to those questions, teams are not empowered. They are exposed.
The result looks like hesitation and over-caution from the outside, but from the inside it is something more rational: people protecting themselves from consequences they cannot fully anticipate. General Stanley McChrystal described a similar dynamic in rebuilding the Joint Special Operations Command. Distributing authority only produced results when it was paired with what he called shared consciousness — a common operating picture that allowed every part of the organization to understand the intent behind decisions, not just the decisions themselves. Without that shared understanding, distributed authority produces distributed uncertainty.
The second failure mode is invisible experimentation. When leadership cannot see what teams are doing with the autonomy they have been given, fear fills the gap. This is not irrational on leadership’s part. AI experimentation carries real risks around data practices, system integration, and downstream effects that individual teams may not be positioned to assess on their own. When those experiments are not visible, leaders respond the way most people respond to uncertainty: they compensate with control.
Approval processes that were supposed to be streamlined get reinstated. The bureaucracy that the transformation was supposed to replace grows back, usually faster than it was removed. Controlling every action does not equate with governance; in fact, it dilutes it. Visibility from lightweight mechanisms make experimentation understandable without making it burdensome, such as sprint demos, or risk tagging that surfaces cross-functional impacts early. These are not substitutes for judgment; they are how organizations maintain situational awareness without centralizing control.
The third failure mode is local optimization that destabilizes the system. Decentralized teams naturally optimize for their own context — their workflows, their pain points, their key performance indicators. In most operational environments, that is manageable. In AI and digital transformation, it frequently is not, because AI changes the dynamics of upstream and downstream processes in ways that a single team cannot always see from where it sits.
Repenning and Kieffer, whose Dynamic Work Design framework was developed across more than two decades of work in real organizations, describe this as one of the central challenges of operational improvement: local solutions that appear to resolve a problem while relocating it somewhere less visible. The answer is not to prohibit local experimentation. It is to pair local autonomy with enough system-level visibility that cross-functional impacts surface before they compound.
What the Center Has to Do Differently
If these three failure modes have a common root, it is this: organizations decentralize responsibility without decentralizing clarity. They distribute the work without distributing the understanding that makes the work coherent.
Correcting that requires a different concept of what leadership’s role is in a decentralized model. The work is not to review and approve. It is to design the conditions under which good judgment can be exercised without a senior leader in the room for every decision.
McChrystal’s framework is instructive here. Empowered execution holds that once every part of the organization is working from the same operating picture and understands the intent behind decisions, authority can be pushed to the person closest to the action. The senior leader’s job is not to make every call; it is to ensure that the people making calls have everything they need to make them well. That shift is harder than it sounds. It requires leaders to invest in clarity before they distribute authority — not after problems surface.
Wiseman’s research on what she calls the Investor in Multipliers describes the same principle from a different angle. Where a micromanager jumps in and retains ownership, the Investor defines what is expected, provides the resources and support the team needs, and then holds people accountable for the outcome. The distinction is not between hands-on and hands-off — it is between leaders who design for accountability and those who substitute their own judgment for it.
Collins argued in Good to Great that disciplined people who deeply understand an organization’s purpose can make sound decisions without layers of bureaucracy — not because rules don’t matter, but because their judgment is grounded in the intent behind them. When following the policy runs at odds with doing what is right, the person who understands why the policy exists can navigate that tension; the person who only knows the rule cannot. This is why defining intent is not just a communication task. It is the foundation on which good judgment travels through an organization without senior leadership attached to every decision.
That means articulating boundaries explicitly — where team authority ends, what categories of decision require escalation, what risks the organization is and is not willing to accept — and establishing escalation criteria specific enough to be useful, so that teams are not left inferring from silence what the rules are.
It also means creating the connective tissue that allows a decentralized organization to function as a coherent whole rather than a collection of independent efforts. It’s more about orchestration than oversight. And it means monitoring system-level effects even as local experimentation proceeds — tracking cross-functional impacts, watching for bottleneck migration, and maintaining enough of a whole-system perspective to catch the problems that local teams cannot see from where they work.
The Practical Upshot
The organizations that navigate this well are not the ones that move fastest to distribute authority. They are the ones that do the design work first — defining intent, establishing boundaries, creating the visibility mechanisms that allow decentralized teams to operate as a coherent system rather than a collection of independent efforts. That work is not a constraint on innovation. It is what makes innovation sustainable.
AI transformation is, at its core, a business transformation. The technology surfaces the data and the analysis; what it cannot supply is the judgment about what to do with it. That judgment — applied consistently, across boundaries, under pressure — is only possible in organizations where the alignment and trust have been built deliberately. And that work starts at the center.
Bearing Check
As your organization distributes more authority to innovate, is the design work keeping pace? Are you providing the clarity of intent, the boundaries, and the visibility mechanisms that allow teams to make confident decisions without senior leadership in the room — and that give senior leadership the confidence to let them?
Bearing Check is published biweekly by Klarhet Compass. Each issue draws on research and real experience to explore what makes complex organizations move well — and what gets in the way.