Organization design in the age of AI: which roles change

In the age of AI, the roles that change most are those at the interfaces between business and technology, and process and product ownership gain weight. Organization design starts with activities and only then with positions: which tasks a language model or an agent takes over determines which tasks remain with people and how their roles change. This article describes a framework with which senior management and heads of department can plan that change. This article provides an overview and does not constitute legal advice.
From position to task
Classic organization design arranges positions in a structure. When AI is deployed, the task is the right unit of analysis, because language models and agents take over parts of activities and rarely entire positions. A position consists of many tasks. Some can be automated, others continue to require judgment and accountability. Labor economists David Autor, Frank Levy and Richard Murnane established this view as the task-based approach: technology substitutes for routine tasks and complements tasks that require judgment and problem-solving.
Three questions per task suffice for the analysis.
- Repeatability: How regularly and how clearly defined does the task occur?
- Verifiability: Can a result be assessed quickly and reliably?
- Cost of error: What does a wrong result cost, and who is accountable for it?
Tasks with high repeatability and good verifiability are the first candidates for AI. A high cost of error requires a human review step. For high-risk applications under the EU AI Act, for example in recruitment or credit decisions, this human oversight is mandatory.
Which roles change with AI?
The roles that change most are subject-matter experts, managers, product and process owners as well as IT and data functions. The change shows in five role groups.
- Subject-matter experts and clerical staff: The emphasis moves from producing to reviewing and improving results. Quality judgment becomes the core competence. A field study of more than 5,000 customer service agents shows the order of magnitude: with an AI assistant they resolved on average 14 percent more queries per hour, and newcomers as much as 34 percent more.
- Managers: They steer teams in which people and AI applications deliver together, and measure results rather than utilization.
- Product owners and process owners: They own the application together with the workflow and are accountable for the metric.
- IT and data functions: They move from project service provider to operator of a platform that business units use.
- New roles: an AI owner per application in the business unit and a function for quality assurance of results.
How job profiles actually govern responsibility is covered in Rollenklarheit. The wider frame for structure and delivery is set out in the article characteristics and metrics of a high-performing organization.
What new roles does AI create?
AI mainly creates two new roles: the AI owner per application and quality assurance for machine-generated results. A steering role at leadership level comes on top.
- AI owner (AI product owner): the person in the business unit accountable for an application and its workflow. Tasks: set goal and metric, prioritize requirements, approve changes, report the benefit in the portfolio.
- Quality assurance for AI results: the function that reviews samples and analyses error patterns. Tasks: define review rules, measure the error rate, feed back to the owner, halt operations when deviations occur.
- AI accountability at leadership level: a member of the executive team or a Chief AI Officer with a budget and a portfolio mandate. When a dedicated role pays off is covered in Die Rolle des Chief AI Officer.
The prompt engineer as a standalone position has not established itself in most organizations. The skill of instructing applications well moves into the business roles and into the team that runs the platform.
How does AI change management and span of control?
AI changes how span of control is sized, because routine coordination disappears and review work emerges. Managers increasingly steer teams in which people and applications deliver together. The metrics shift from utilization to lead time and quality of results.
A clear rule applies to sizing. In teams with a high share of standardized cases, where applications take over reporting and status checks, the span can grow. In teams with a high share of quality judgment and coaching it stays the same, because this work remains with the manager. The time for it is planned into the organization design.
Middle management is particularly affected, because applications can take over part of its coordination work. The freed capacity only creates impact if the organization deliberately assigns it to new tasks, for example reviewing results and enabling teams.
Defining decision rights in the operating model
As soon as an AI application takes over parts of a workflow, a question arises that the existing organization chart does not address: who is accountable for the result of a machine-generated step? The answer belongs in the operating model. It defines which decisions applications prepare and which remain with people. Why it precedes the structure is explained in Operating Model.
A tiered model helps with the assignment.
- Prepare: The application drafts a proposal, a person decides.
- Decide with review: The application decides within defined limits, a person reviews samples.
- Decide without individual review: The application decides autonomously where the cost of error is low, and monitoring tracks deviations.
The classification per task is documented and approved by senior management. It prevents accountability from shifting unnoticed from people to machines within the workflow.
Typical missteps in role design
The first misstep occurs when an organization creates specialist AI roles before an application is in the workflow. An AI lead in a business unit without an application holds a title but no decision authority. AI accountability at leadership level with a budget and a portfolio mandate, by contrast, belongs at the very start.
The second misstep concerns review. Where applications produce results, the volume of what a person must check grows. If the organization does not factor review time into capacity, a new bottleneck emerges and the review becomes a formality. Quality judgment then loses exactly the value for which the role was created.
The third misstep lies in centralization. A central team provides platform and standards, while accountability for the result of a workflow stays in the business unit. If the two are mixed, every change waits in a queue at the central team. The trade-off between a central team, a center of excellence and decentralized ownership is covered in KI-Organization.
A four-step approach
First, the organization captures the tasks per unit and assesses them against the three criteria from the first section. Second, it assigns each task a level of the operating model. Third, it redesigns the roles: review and ownership functions are created, job profiles updated, and managers given room for quality judgment. Fourth, it tests the new set-up in a defined unit and measures lead time and quality. Only then does expansion follow.
This approach links organization design with the introduction of AI, because every role decision belongs to a concrete application in the workflow. A reorganisation without reference to applications describes roles no one can fill. An AI initiative without reference to roles produces prototypes without users.
In companies with a works council, its involvement belongs in the first step. Changed workflows and new technical systems regularly touch information and co-determination rights, and early involvement shortens the consultation.
Niehues Consulting as a partner for roles and operations
Niehues Consulting combines both sides of this task: organization design with roles and decision paths, and the introduction of AI into operations. The approach follows the Niehues Delivery Model with its phases Understand, Build and Introduce. Three of its review questions matter for the role question. In Understand: who decides, who delivers, who uses? In Build: which roles, decision paths and interfaces do operations require? In Introduce: who owns operations from day one?
Marin Niehues steers programs as delivery manager or product owner and builds organizations that deliver. His projects range from venture-backed scale-ups to DAX 40 groups, with assignments in 20+ countries on six continents. The free initial consultation takes 30 minutes: request a consultation.
Frequently asked questions
Which roles change with AI?
The clearest changes affect subject-matter experts, who move from producing to reviewing, and managers, who steer mixed teams of people and applications. Owners per application in the business unit come on top. IT and data functions develop into operators of a shared platform.
What new jobs does AI create in organizations?
New roles are the AI owner or AI product owner per application, quality assurance for machine-generated results and AI accountability at leadership level. The prompt engineer remains the exception as a standalone position, as the skill moves into business roles.
How do you start organization design for AI?
By breaking activities down into tasks and assessing them for repeatability and verifiability as well as the cost of error. Only then are roles designed. A pilot unit tests the new set-up before it is expanded.
Will AI eliminate jobs?
AI first takes over tasks and rarely entire positions. Where the share of repeatable, easily verified tasks in a position is high, its design changes substantially and demand for it falls. Senior management therefore plans early which new tasks the freed capacity will take on.
How does AI change organizational structure?
Structure follows the new roles: fewer coordination layers where applications take over reporting, and new ownership functions in the business units. A central platform supports the units. Accountability for results stays decentralized.
Does every company need a Chief AI Officer?
Every company needs named AI accountability at leadership level with a budget and a portfolio mandate. Once several applications run across different units, a dedicated role pays off. Below that threshold, a member of the executive team takes on the task.
Does the works council need to be involved when roles change?
Yes, in companies with a works council it regularly has to be involved when workflows change and new technical systems are introduced. Involvement at the start of the task analysis shortens the consultation considerably. The company clarifies the details with its legal advisers.
Who is accountable for the result of an AI application?
A named owner in the business unit is accountable for result and quality. The operating model defines which decisions applications prepare and which they take. Senior management approves the classification.
Let’s talk about your initiative.
In 30 minutes we clarify where your organization stands and which step will pay off most now.
Request a consultation30 minutes · free · no obligation



