The AI-Native Organization Will Not Look Like Today’s Organization With AI Added

An organization chart contains assumptions about how work gets done. It groups activities that benefit from shared expertise, locates authority where judgment is expected to reside, and creates management roles to coordinate what individuals and teams cannot resolve alone. Reporting cycles, approval structures, and performance measures reinforce those choices.

When AI changes the cost of accessing expertise, developing analysis, or coordinating work, some of those assumptions deserve to be reconsidered. An assistant can help an employee operate more effectively within an existing role. It cannot, by itself, determine whether that role still contains the right activities, whether decisions sit with the right people, or whether management attention is directed toward the right problems.

In AI Transformation Happens at the Workflow Layer, we argued that the workflow connects AI capability to business outcomes. The next question is what happens when several important workflows change, but the organization around them does not.

Normatic's view is that sustained changes in work eventually require a corresponding examination of the operating model. An AI-native organization, as we use the term, is one whose design reflects demonstrated AI capabilities and their limitations. Its distinction lies in how it allocates work, knowledge, and authority—not in its age, its technology budget, or the number of assistants it deploys.

Organizations embody the constraints they were built to manage

Organizational structures serve several purposes, including accountability, incentives, professional development, and the resolution of conflicting interests. They also help manage the cost of knowing what to do and finding someone who knows more.

Luis Garicano's research on knowledge hierarchies gives this latter function a precise economic interpretation. His model explains how a hierarchy can organize problem solving when acquiring knowledge and matching problems to knowledgeable people are costly. More common problems can be handled closer to production, with unresolved problems referred to specialists. This is a theory of one important organizational mechanism, rather than a complete account of why hierarchies exist. Hierarchies and the Organization of Knowledge in Production

AI creates possibilities for changing that mechanism. Where reliable assistance makes relevant knowledge available at the point of work, an employee may be able to resolve a problem that previously required a referral. Where AI can assemble context across systems, a manager may spend less effort establishing what happened before exercising judgment about it.

The organizational implication extends beyond time saved. If a handoff existed partly to obtain information that can now be accessed directly, the reason for that handoff has changed. If a management meeting existed mainly to reconstruct a shared view of performance, its purpose may need to change too.

Other reasons for those arrangements may remain. Independent challenge, separation of responsibilities, negotiation, and trust cannot be assumed away because information moves faster. The opportunity is to distinguish arrangements that still serve a necessary purpose from those maintained primarily by an earlier constraint.

A role is more than its remaining tasks

Adding AI to existing jobs can create substantial value. It is also a natural way to begin learning: the employee retains context, responsibility, and established relationships while experimenting with a new capability. The limitation appears when the existing job description becomes a permanent boundary on what can change.

A job is a bundle of activities, expertise, relationships, and authority. If AI changes several activities within that bundle, merely subtracting the automated tasks can leave an incoherent role. A specialist may be freed from routine preparation but still lack the authority to act on the analysis now available. Another employee may inherit a greater concentration of difficult cases without the training, support, or time needed to handle them well.

Evidence from collaborative work makes this question tangible. In the published version of The Cybernetic Teammate, Fabrizio Dell'Acqua and colleagues report a field experiment with 791 Procter & Gamble professionals working on product-innovation challenges. Individuals using AI matched the performance of teams without AI, and AI-assisted proposals combined technical and commercial perspectives more evenly. These findings concern the experimental innovation work; they do not demonstrate that whole teams can be replaced or that long-term organizational performance would improve through such a change. The Cybernetic Teammate

Our interpretation is that AI can change which contributions need to be assembled through separate people at each stage of a problem. Some work may become feasible within a broader role. Other work may benefit from deeper specialization, with experts concentrating on difficult interpretation, creative exploration, relationships, or consequential decisions. The appropriate bundle depends on what the redesigned work actually requires.

This makes work design a question of capability and responsibility, not simply workload reduction. Moving routine activity to systems may increase the importance of the human work that remains. It can also increase its intensity. A role consisting almost entirely of exceptions requires different support from one in which demanding cases are interspersed with predictable activity.

More analysis does not determine where authority belongs

AI can expand the amount of analysis, monitoring, and decision support an organization can produce. Human attention and the authority to commit resources remain constrained. If every new signal travels through the same escalation chain, increased analytical capacity can create more managerial congestion.

There is no universal structural answer. Research by Nicholas Bloom, Luis Garicano, Raffaella Sadun, and John Van Reenen distinguishes technologies that make information easier to use from technologies that make communication easier. In American and European manufacturing firms, better information technologies were associated with greater autonomy and wider spans of control, while communication technologies were associated with less autonomy. The study predates generative AI, but it provides evidence against treating all technological change as a single force toward either centralization or decentralization. The Distinct Effects of Information Technology and Communication Technology on Firm Organization

For AI, we see both possibilities. Better local understanding may justify giving an employee more authority within clear limits. Better visibility across an enterprise may justify moving some decisions upward when their consequences span multiple businesses. The same organization may pursue both changes for different decisions.

A useful distinction is between the ability to produce a recommendation and the right to act on it. A technically capable system does not acquire business authority merely because its answer is persuasive. Delegation requires explicit boundaries, reliable performance, escalation arrangements, and a person accountable for the result.

Management roles therefore need to be examined by the work they perform. A manager who mainly collects and relays information may have more scope to change than one whose contribution centers on coaching, resolving competing priorities, or maintaining critical relationships. A wider span of control becomes credible only if the resulting demands on judgment and support remain manageable. Removing routine coordination does not prove that a manager can responsibly lead more people.

Repeated workflow changes expose a different organizational logic

Consider an illustrative customer-operations organization with separate teams for inquiries, billing adjustments, and service recovery. Each team has its own queue, performance targets, and supervisors. Employees refer cases because information and authority are distributed across those boundaries.

Suppose redesigned workflows demonstrate that AI-assisted employees can reliably assemble the relevant context and resolve a wider range of routine issues, with controlled actions in connected systems. If that change remains confined to one queue, an improved local process may be sufficient. If it recurs across several important customer journeys, the organization faces a different question: whether its existing divisions still provide the best basis for allocating work.

One possible response is a broader customer-resolution role supported by specialist escalation, shared knowledge stewardship, and clearly bounded authority. Specialists would remain responsible for matters requiring depth or independent approval. Supervisors might devote more attention to the quality of resolutions and the development of judgment. Shared services might concentrate more on maintaining common capabilities and resolving complex cases. This is a possible organizational design, not a Normatic client example or a recommendation to combine these functions everywhere.

The change would also affect incentives. Separate targets for handling an inquiry, processing an adjustment, and closing a case can reward activity within each function while obscuring whether the customer's problem was resolved. Broader responsibility requires measures and resource decisions that recognize the whole outcome. Functional expertise can remain valuable even when the primary basis of accountability changes.

The case for redesign becomes stronger when several workflows repeatedly encounter the same structural obstacle: fragmented authority, duplicated expertise, incompatible targets, or ownership that stops at a handoff. Those patterns reveal an operating-model issue that another isolated use case is unlikely to resolve.

The progression is therefore evidential. AI capability changes a task; task changes make a different workflow feasible; repeated workflow changes reveal implications for roles, decisions, and coordination. Organizational redesign follows when those implications are material and sufficiently demonstrated. Some decisions about authority must be made earlier to enable safe experimentation, so this is a logic of learning rather than a rigid sequence.

Accessible knowledge still needs an institutional owner

Making knowledge easier to retrieve changes how expertise is encountered. An employee may no longer need to know which colleague holds a particular document or remembers a previous decision. Onboarding can include access to relevant precedents and explanations within the work itself, reducing some dependence on informal introductions.

But availability does not establish authority. Organizational knowledge includes outdated procedures, contested interpretations, unwritten exceptions, and decisions whose rationale was never recorded. An AI system can make those inconsistencies easier to encounter as readily as it can make useful information easier to find.

Our view is that wider access increases the importance of stewardship. Subject-matter experts may spend less time answering recurring questions individually and more time determining which guidance is valid, resolving contradictions, and identifying where experience has outgrown the documented rule. Institutional memory needs to preserve why a decision was appropriate in its context, not merely reproduce the answer that was given.

This changes the relationship between expertise and management. Knowing whom to ask may become less of a bottleneck; knowing when the available answer should be challenged remains consequential. Managers and experts retain a role in developing that discernment, including through discussion, supervised experience, and exposure to unfamiliar cases.

The organization also needs to recognize this work in staffing and performance expectations. Maintaining shared knowledge cannot be treated as incidental overhead if many workflows now depend on its quality. The more widely a source is reused, the more consequential its errors and omissions can become.

Management systems must evolve with the work

The operating model includes how the organization decides what to pursue and how it learns from the results. AI can affect planning, forecasting, resource allocation, performance monitoring, and the management of transformation itself.

Consider the purpose of a periodic business review. If much of the meeting is spent assembling numbers and explaining discrepancies, better analytical support may free attention for decisions. The opportunity is to examine assumptions, compare options, and commit to action with a clearer understanding of uncertainty. Producing more reports without changing the conversation preserves the original constraint on management attention.

More frequent information does not require constant intervention. An organization may need explicit thresholds for when new evidence warrants a review and when a plan should remain stable. Otherwise, continuous monitoring can produce continual reprioritization. Forecasts and simulations also need to be distinguished from observed results; their precision cannot settle the choices leaders must make about competing objectives.

Resource allocation can change as shared AI capabilities begin to support multiple workflows. Funding each use case independently may underprovide common knowledge, integration, and evaluation capabilities. Conversely, central funding can become detached from whether those capabilities solve a material business problem. The organizational task is to connect shared investment with accountable demand and evidence of value.

Governance belongs within these management mechanisms. A separate AI committee cannot substitute for clear authority in the operating business. The people responsible for outcomes need a defined relationship with those responsible for technical reliability, independent challenge, and the limits of permitted action. Human accountability must remain legible as decisions and activities move across systems.

Redesign should follow evidence, not precede it

None of this establishes that every organization should reorganize immediately. AI capability varies by task and context. Integration and evaluation carry costs. Existing structures may continue to serve essential purposes even after some activities change. Benefits from augmentation can justify investment without a wider redesign.

The stronger claim is conditional: where AI materially changes the economics and reliability of important workflows, preserving every inherited role, boundary, and management mechanism can prevent the organization from realizing the fuller opportunity. That constraint becomes harder to ignore as the changes accumulate.

For Normatic, the AI-native organization is a direction of institutional adaptation rather than a standard blueprint. Its design can differ across functions, and useful changes may begin with decision rights, team responsibilities, or funding mechanisms before they appear on an organization chart. Technical capability supplies possibilities; demonstrated operating performance provides the basis for choosing among them.

The long-term ambition is an organization able to revise how it works as its capabilities evolve, while preserving clear responsibility for its decisions. AI transformation begins in workflows. When those workflows change substantially and repeatedly, leadership must be willing to reconsider the organization built around them.

Sources

• Luis Garicano — Hierarchies and the Organization of Knowledge in Production. Journal of Political Economy, 2000. A theoretical account of hierarchy as a way to organize knowledge and problem solving.
• Fabrizio Dell'Acqua and colleagues — The Cybernetic Teammate: A Field Experiment on Generative AI and Teamwork. Organization Science, published online June 2026. The published version reports 791 participants; earlier working-paper accounts used a different sample count.
• Nicholas Bloom, Luis Garicano, Raffaella Sadun, and John Van Reenen — The Distinct Effects of Information Technology and Communication Technology on Firm Organization. Management Science, 2014. Empirical evidence on technology, autonomy, and managerial spans of control.