Thematic notes on ”How AI Agents May Reshape Operations Management” - Key Insights

Picture: Bing Image Creator

Text: Timo Seppälä, Xinry Liu and Siavash Khajavi

As part of a new Delphi research on how AI agents may reshape OM over the next decade, Xinry Liu and Siavash Khajavi conducted an interview with Timo Seppälä. These notes summarize the main viewpoints expressed by the interviewee. The points below reflect statements made during the interview. The interview was conducted 5th of February, 2026 and it lasted for 45 minutes. 

1. Current AI Systems in Operations Are Mainly “Prediction Machines”

Timo emphasized that most existing applications of AI in organizations focus on prediction, rather than judgment or execution.

He noted that in practice, companies are currently offered “prediction machines,” while “judgment machines” and “execution machines” do not yet exist in a mature form. According to him, current AI architectures are well suited for generating predictions quickly and at scale, but are not designed to handle higher-level judgment or to directly execute operational decisions.

He also highlighted that the ability to generate predictions has increased significantly: what previously took days can now be done much faster and in large numbers. As a result, prediction itself is no longer the main bottleneck. Instead, the unresolved challenge lies in deciding which predictions should be used in practice and how they should be translated into action.

2. From Prediction to Optimization and Execution

Building on this, Timo described a shift in focus from prediction problems toward optimization and execution problems.

He argued that as organizations gain the ability to generate many predictions, the problem becomes one of selecting the “right” prediction for a specific day or context and using it to guide actions. In his words, the core business challenge is moving from prediction toward judgment and execution.

However, he also stated that current AI agents are not yet capable of performing these judgment and execution functions reliably, and that this limitation applies across different areas of operations management.

3. Decision Cycles Are Moving from Days to Milliseconds

Looking ahead, the interviewee stressed that the implications of AI agents for OM should be understood in terms of decision-making time scales.

He contrasted today’s organizational decision cycles—often based on daily learning and adjustment—with future environments where prediction, judgment, and action would need to occur in seconds or milliseconds, particularly in autonomous systems and advanced manufacturing contexts.

He pointed out that existing cloud-based infrastructures are not well suited for such millisecond-level decision loops. As a result, decision-making and error correction would increasingly need to take place at the edge, closer to machines and operational equipment. He also noted that in such fast loops, decisions cannot be overly complex and may need to rely on simplified decision structures and tolerances rather than deterministic “zero–one” logic.

4. Product and Process Design: Toward Hyper-Personalized Products and Services

Within the product and process design domain, the interviewee described a future shift from configurable products toward hyper-personalized products and services.

He argued that personalization would need to be embedded in the “DNA” of products, especially in contexts where products and systems become more autonomous. According to him, AI tools, supported by large-scale data and computing power, could enable closer matching between customer requirements and product or service configurations. He also linked this development to broader trends such as servitization.

5. Hidden Knowledge and the Role of Transcriptions in Operations

Timo repeatedly emphasized that a significant amount of operational knowledge remains hidden within personnel and organizational practices.

He discussed the use of AI-based transcription of meetings and work interactions as a way to make this hidden knowledge more visible and traceable. By transcribing discussions over time, organizations can analyze how ideas evolve, how decisions are made, and how development activities progress. He suggested that this could create an additional layer of organizational knowledge that is currently underutilized across OM functions.

6. From Point Solutions to Continuous Monitoring

Another theme concerned the nature of current digital systems in operations. The interviewee described most existing solutions as point solutions, designed to address narrow tasks.

He suggested that future AI-enabled systems may move toward more continuous monitoring across operations. At the same time, he emphasized that although machines can increasingly “see” and “listen,” current AI systems do not genuinely “think.” What is often labeled as reasoning should not be equated with human thinking capability.

7. Obstacles to AI Agent Deployment: Regulation and Bureaucracy

When discussing obstacles to deploying AI agents in operations, Timo mentioned regulatory and institutional barriers, particularly in the European context.

He argued for technology-neutral regulation and expressed concern about excessive bureaucracy and overlapping regulatory layers. In his view, many regulatory regimes already exist across industries (e.g., safety, privacy, quality, and security), and adding separate AI-specific regulation risks slowing down innovation and deployment.

He contrasted Europe with other regions, where, according to him, experimentation with new technologies faces fewer procedural constraints, and suggested that regulatory complexity contributes to slower adoption in Europe. 

8. Responsibility and Accountability: Toward Multi-Party Contracts

On responsibility for failures in autonomous AI systems, the interviewee argued that traditional bilateral contracts are insufficient when multiple actors are involved. He suggested that responsibility may need to be addressed through multi-party contractual arrangements, potentially supported by smart contracts and distributed ledger technologies to enable traceability in highly autonomous systems.

Closing Perspective

Within the context of this Delphi-style inquiry into the future of AI agents in operations management, the interviewee’s perspective highlights both technological and institutional challenges. While AI systems have already become embedded in prediction and monitoring tasks across OM activity areas, significant limitations remain in judgment, execution, real-time decision-making, regulatory alignment, and responsibility allocation. The interview underscores that the future development of AI agents in OM is shaped not only by technical capabilities, but also by organizational design, regulatory frameworks, and contractual arrangements.

Keywords: Delphi Research, AI-Agents, Operations Management, Desicion Cycles, Organizational Design, Regulatory Frameworks, Contrctual Arrangements