Agentic AI changes the structure of work
Traditional economic models often assume that AI automates individual tasks while humans remain responsible for coordination, interpretation and final judgment. Under this view, work changes gradually: some tasks are automated, while workers move toward activities that require contextual reasoning, interpersonal skills and responsibility.
Agentic AI challenges this assumption. As AI systems become able to break complex goals into subtasks, select tools, generate code, evaluate outputs and revise plans, they begin to operate not only within existing workflows, but upon the workflows themselves.
The economic question is therefore no longer only which tasks AI can perform. It is whether AI can increasingly organize, coordinate and integrate the tasks that make up an entire professional role.
This framework is developed in our paper, The Economics of Hybrid Intelligence. Read the full paper below or continue for an accessible overview of the main arguments.
The Economics of Hybrid Intelligence
Suggested citation: Sherson, J., Schroeder, S., Pytlikova, M., Rafner, J., & Lodefalk, M. (2026). The economics of hybrid intelligence. Working paper, Center for Hybrid Intelligence, Aarhus University. Retrieved from https://hi-infinity.ai/research/the-economics-of-hybrid-intelligence/
From task automation to workflow automation
Professional roles are rarely defined by one isolated task. They consist of connected activities: gathering information, interpreting context, producing outputs, checking quality, communicating with stakeholders and assuming responsibility for the final result.
Agentic AI makes the boundaries between task and role automation less clear. Even when a human remains necessary for one part of the process, the surrounding workflow may be reorganized so that fewer people are needed to coordinate and complete it.
Figure: The progression from task-level AI support to workflow orchestration handled completely by GAI.
From general-purpose tech to dynamic GPT
Earlier general-purpose technologies depended on organizations to create the complementary innovations needed for adoption. Firms redesigned processes, workers developed new skills and institutions adapted over time.
AI may differ because it can increasingly contribute to producing these complements itself. It can help write software, create documentation, generate training materials, test interfaces and identify organizational bottlenecks. AI is therefore not only adopted by organizations; it may also accelerate the reorganization of the organizations that adopt it.
Three perspepctives on AI’s economic impact
The economic debate can be understood through three perspectives. A cautious view emphasizes adoption frictions and organizational constraints. A technology-acceleration view emphasizes scaling and increasingly agentic systems. An intermediate market-dynamics view focuses on how expectations and incentives may trigger restructuring before full automation is mature.
Different assumptions, Different dynamics, Different futures.
Figure: Visualization of the different possible AI pathways
These perspectives are not mutually exclusive. AI capabilities may improve rapidly while implementation remains difficult. At the same time, firms may react to expectations about future automation before the technology is operationally mature.
What Current Estimates Tell Us
Discussions on job displacement from AI are everpresent, but what does the current numbers actually say? Which jobs are most likely to be influenced negatively or positiviely?
Which workers and tasks are most exposed?
According to the IMF, AI exposure is concentrated in the private sector and increasingly affects highly educated and high-income workers, rather than mainly manual occupations. Around 60% of Danish jobs are exposed to AI and approximately 20% are at risk of displacement, rising to 25% in the private sector compared with 13% in the public sector.
In the US OpenAI believes they are more exposed than the EU, in the short term, because they have more technology-intensive and knowledge-based jobs. The EU is less exposed overall, because physical and manual work remains relatively protected.
How much work could be disrupted?
Estimates vary widely depending on what counts as disruption. According to OpenAI, 50–90% of jobs could be affected, while its economic estimate places around 14% of EU employment at risk.
The IMF estimates that 60% of Danish jobs are exposed and 20% could face displacement. In its upper scenario, Danmarks Nationalbank assumes that 80% of tasks in exposed sectors could be profitably replaced, while agentic-AI scenarios reach around 82%. However, they do not include AI in their economic forecasts.
Note that estimates of AI’s labor-market impact vary widely because they measure different things, including task exposure, productivity augmentation, technical replacement potential and realized displacement.
Organizations May Restructure for AI That Does Not Yet Exist
Firms do not make employment decisions only on the basis of what AI can do today. They also respond to expectations about future capabilities, competitor behavior and investor demands.
This creates an AI expectation trap: Augmentation evidence may be interpreted as evidence of imminent replacement. Organizations may reduce junior hiring, training and staffing before AI can reliably perform the full workflow.
However, AI making workers more productive does not necessarily mean that AI makes workers unnecessary. The result can become self-reinforcing. Weaker expertise pipelines increase dependence on AI, while workforce reductions are interpreted as proof that automation-first strategies are succeeding.
The Hybrid Intelligence economic framework
To explain why human centrality in an AI economy is a design choice rather than a technological question, the economic framework analyses the AI transition across interdependent societal stages.
- Workflows that build expertise: AI supports search, prediction, drafting and simulation, while humans retain meaningful authority over framing, contextual interpretation and final commitment.
- Markets that reward human contribution: As standardized functional output becomes cheaper, trust, accountability, provenance and responsibility become important sources of differentiation.
- Institutions that sustain investment: Standards, certification, education and procurement can make genuine human contribution visible and economically viable.
- Human touch marketplace: Is the overall take, describing markets in which value is created not only through functional performance, but also identifiable human judgment, accountability and trust.
Together, these levels show how markets and institutions can be designed so that human judgment remains verifiably productive, commercially rewarded and economically sustainable as AI capabilities scale.
Two AI pathways, two different futures
The choices we make today determine whether AI creates abundance without agency – or abundance with human flourishing.
Figure: Visualization of the two possible political pathways for AI integration into society
The two pathways use similar technological capabilities but create different feedback loops. Automation-first strategies can weaken expertise and increase dependency. HI strategies invest in human capability, market trust and better combined performance.
What can decisionmakers do?
Hybrid Intelligence will not become the dominant model through good intentions alone. Decision-makers must create conditions that make genuine human contribution visible, valuable and investable.
- Develop standards and certification that distinguish meaningful human judgment from symbolic oversight.
- Support HI investment and procurement so organizations can develop systems that build expertise rather than only reduce labor costs.
- Protect expertise pipelines through education, training and roles that allow people to develop judgment alongside AI.
We describe this institutional approach as Sensecurity: An extension of the Nordic flexicurity model focused not only on helping workers move between jobs, but on ensuring that human judgment remains structurally valuable in an AI-rich economy.
Read the full paper to dive deeper into The Economics of Hybrid Intelligence Framework, where we propose economic arguments for hybrid intelligence, that can help mitigate job displacement concerns.