Companies are redrawing reporting structures before redesigning the system through which work actually moves.
The structure is not the system
A new wave of reorg is coming. It is being driven by AI and most companies will choose to restructure the wrong thing. Mercer’s 2026 workforce research found that 98% of executives plan to make significant organizational design changes within the next two years. Most expect a meaningful share of their workforce to be redeployed or reskilled because of AI. Boards are asking every CEO the same question: what does our organization look like when AI reliably delivers a real share of the work?
In company after company, the answer is taking the same shape: flatten here, consolidate there, rename a few business units, announce the new structure at a town hall. Eighteen months later work moves the same way and encounters the same bottlenecks, except now there are fewer people and everyone is expected to be more efficient because they have an AI license to build presentations, write communications, and summarize meetings.
The danger is that the reporting structure may not be a root cause of workflow inefficiency. Under every org chart sits the system that truly determines the effectiveness of work: the way demand gets surfaced, understood, prioritized, broken down, moved through processes, carried across teams, and reported back in signals leadership can act on. Every organization has a work management system. Almost none deliberately designed it.
Control was never what leaders thought it was
Here is the unspoken fear underneath every AI reorg conversation: if teams work fluidly, if agents execute work without a human watching every step, if structure stops being fixed, what purpose does leadership fulfill?
Bureaucracy is not the enemy. An organization needs bureaucracy the way a body needs a skeleton. What it needs is the minimum amount required to operate. Gary Hamel and Michele Zanini found that employees in large companies spend roughly a quarter of their time on internal compliance that creates nothing of value, and about 80% say bureaucracy slows decisions they already know how to make. Bureaucracy was already expensive when humans did the work. It is fatal when work autonomously moves at lightning speed through a bottlenecked system.
The design principle that replaces layered control is old and battle-tested: intent flows down, accountability for execution is decentralized, and work progression information flows up. Leaders who fear losing control usually discover they never had it. They had commitments and decisions built on stale PowerPoint decks and status meetings that left them uninformed and reactive. Real control is a clean line of sight from intent to outcome. Clarity.
The system becomes executable
When a work management system is explicit, it stops being tribal knowledge and becomes something people and AI agents can pick up and run. Intake becomes a mechanism an agent can triage. Prioritization becomes logic an agent can apply and a human can override. A definition of done set at the organization level becomes inherited context instead of a lesson every team has to relearn.
Most organizations get none of this value because they use AI backwards. They demand output and rarely build the instructions, context, specifications, and catalogs that make the output trustworthy. The discipline that pays is the unfashionable one: feed the system the institutional knowledge, decision rules, voice, and definitions before asking it for anything.
AI amplifies whatever system it is plugged into. Good inputs, better outputs. Broken in, broken out, just dangerously faster. If goals do not say why the work matters, if priorities live in the loudest voice in the room, and if the process exists only in people’s heads, AI will automate the confusion and scale it.
Open loop, closed loop
AI makes it cheap to generate a full backlog from a roadmap in an afternoon. Organizations that spent years learning to build products iteratively are quietly sliding into open-loop execution: build everything, barely show anyone, get feedback never. An open-loop system runs without checking its own results, so nothing corrects the course.
The answer is not a choice between defined and adaptive. It is both at once. Trained inputs on one side: specifications, catalogs, brand guidelines, definitions, and institutional knowledge. Loops in the middle that check every output against those definitions and send the misses back around. Repeatable, high-quality outputs on the other side, in service of the desired outcome.
State and adhere to the definitions. Let the plan breathe. Build a slice, show it to someone real, and let the feedback inform what happens next. Quality stops depending on someone remembering to check because the checking is built into the system.
- Full backlog from a roadmap
- Build everything
- Barely show anyone
- Feedback never returns
- Trained inputs and definitions
- Build a slice, show someone real
- Check output against definitions
- Send the misses back around
The right order, and the window
The sequence for reorganizing in the AI era is the opposite of the one most companies are following. Identify the points of intake. Map the existing work systems. Define where people, automation, augmentation, and AI assistance can integrate. Design the minimum version on purpose. Decentralize execution around prioritized goals. Run metrics up instead of meetings. Then let the integration points and data reveal whether the existing hierarchy still makes sense.
The end state is the client’s own people running the system without us. We embed the intake mechanism, prioritization schemas, goal structure, and reporting spine, then build the internal capability to operate them and turn the keys over. An operating model you cannot hand back is not an operating model. It is a dependency with better branding.
AI has put nearly every organization in flux at the same time, and flux is the best window in which to change operating models. The organizations that emerge stronger will not be the ones with the boldest new structure from reducing headcount. They will be the ones that used the window to deliberately design the system everyone else is still running by accident.
-
Start here Identify points of intake Where demand actually enters
-
Map existing work systems See what is already running
-
Define integration points People, automation, augmentation, AI
-
Design the minimum version On purpose, not by accident
-
Decentralize execution Around prioritized goals
-
Run metrics up Instead of meetings
-
The chart comes last Let data test the hierarchy Does the existing structure still make sense?
See the system first. Design the minimum. Let the evidence redraw the hierarchy if it needs to move.
Share With Your Team
Choose who it is for. You send it yourself from your own email; Agileon never writes to your colleagues.


