For decades, the existence of firms was a given in economic models. We analyzed supply and demand, markets and competition, but rarely questioned why organizations themselves existed in the first place. Coase’s seminal work, “The Nature of the Firm,” posits that firms exist primarily to minimize transaction costs, which include the costs associated with searching for information, negotiating contracts, and enforcing agreements (Dollery & Leong, 1998). Williamson expanded on this by introducing concepts such as bounded rationality, opportunism, and asset specificity, which further elucidate why firms are structured in particular ways to manage complexity and uncertainty more effectively than market transactions alone (North, 1990). Douglass North’s contributions to institutional economics highlight the role of institutions in shaping economic behavior and reducing transaction costs, emphasizing that both formal and informal rules are crucial for economic performance (Caballero & Soto-Oñate, 2016). These foundational ideas are not just academic relics; they are becoming increasingly relevant in the age of artificial intelligence, particularly as we grapple with the rise of AI agents.

The Legacy of Transaction Cost Economics

Coase’s fundamental argument was that firms exist to minimize transaction costs. These costs, encompassing everything from searching for information to negotiating and enforcing contracts, are often lower within a firm’s boundaries than on the open market. Williamson further elaborated on this, highlighting the role of bounded rationality (our limited cognitive abilities), opportunism (the tendency for individuals to act in their own self-interest), and asset specificity (investments tailored to a particular transaction that lose value elsewhere) in driving organizational form. In essence, firms provided a structure to manage complexity and uncertainty more efficiently than a collection of individual market transactions. Douglass North further expanded on this concept by emphasizing the role of institutions—both formal rules and informal norms—in shaping economic behavior and reducing transaction costs within a broader societal context.

AI: Technology and Business Model

In the contemporary landscape, the rapid advancement of AI technologies, particularly through the development of transformers and foundation models, presents both opportunities and challenges for firms. The distinction between AI as a technology and AI as a business model is critical; the latter pertains to how AI capabilities are utilized to create and capture value. The economics of platforms, which is deeply rooted in transaction cost theory, becomes increasingly relevant as organizations explore how to integrate AI into their operations. Scholars have examined how modularity and design rules can facilitate this integration, suggesting that a well-structured platform can enhance efficiency and reduce costs (Aggarwal & Zhao, 2009).

The Allure of AI Agents

One prominent business model emerging around AI is that of AI agents. These autonomous software entities are designed to automate tasks, mirroring in some ways earlier technologies like AppleScript, but with the promise of far greater ease of use and accessibility. For individuals and small businesses, the potential benefits are clear: streamlined workflows, increased efficiency, and the ability to automate complex processes without specialized coding skills. This echoes the democratization of technology envisioned by scholars like Eric von Hippel, who studied user innovation.

The Corporate Conundrum: Productivity or Chaos?

However, the picture becomes more complex when we consider the adoption of AI agents within larger organizations and even governments. The initial allure is undeniable: empower every employee with an army of personal AI assistants and watch productivity soar. But this optimistic vision might be overlooking critical factors rooted in the very theory that explained why firms exist in the first place.

Herein lies the potential pitfall. As employees create and deploy their own specialized agents, a number of challenges arise:

1. Atomization and Turnover: For instance, the atomization of processes may result in a legacy of digital “cruft,” where individual agents become idiosyncratic and difficult to transfer or adapt, particularly in the face of employee turnover (Kay, 2014). This scenario mirrors the concerns raised by transaction cost economics regarding the efficiency of internal versus external capital markets, where fragmented systems can lead to duplicated efforts and decreased coordination (Dollery, 2001; Hart, 1995).
2. The Illusion of Efficiency: The proliferation of agents might create a fragmented organizational landscape, with different departments using different tools and approaches. This could lead to duplicated effort, conflicting processes, and a general decrease in coordination, ultimately hindering overall productivity, not raising it. For example, consider a scenario where the sales and customer service departments each develop their own agent-driven systems for managing customer interactions. Without proper integration, these systems might generate conflicting information or lead to redundant communications, ultimately frustrating customers and undermining the company’s reputation.
3. The Platform Paradox: From Lower Costs to Lock-in: Moreover, the reliance on external platforms for deploying AI agents can create a new form of lock-in, where organizations become dependent on specific platforms, thereby increasing their external transaction costs and diminishing their internal coherence. This phenomenon reflects observations about centralized control in network-based systems, where the platform provider assumes a gatekeeping role, potentially stifling organizational adaptability. The integration of numerous independent agents can lead to increased entropy within the organization, complicating management and coordination efforts (Marty & Warin, 2023).
4. Integration and Entropy: In a traditional firm, management and organizational structure create integration and reduce entropy, making internal transactions simpler. Individual agent creation is more akin to artisanship or even distributed markets, as explained by Von Hippel. As mentioned before, the coordination of all these agents, the entropy, and the integration efforts will be more and more absorbed by the platform. What is left for the organization? To make an analogy with physics, the proliferation of AI agents within an organization can be likened to increasing the Brownian motion of particles within a system. Each agent, acting independently and driven by local optimization goals, introduces a degree of randomness and disorder. Without a countervailing force to maintain order and coherence, the organization’s internal entropy increases. This can lead to a gradual erosion of organizational structure and a decline in overall efficiency. The organization, once a well-defined entity, begins to resemble a chaotic, decentralized network, its boundaries blurring as it becomes increasingly intertwined with the external platform. This can eventually lead to the dissolution of the organization as it currently exists. At the very least, it will turn it into a complex system, rendering most current management paradigms obsolete.

The New Gatekeepers

In essence, while AI agents seemingly lower the cost of automating tasks within the firm, they might inadvertently increase the organization’s dependence on the external platform. This creates a subtle shift in power, transforming the platform provider into a new kind of “gatekeeper” in the digital age. The organization’s internal transaction costs might be reduced at the micro-level, but at the cost of increased external transaction costs and a loss of organizational coherence and long-term adaptability. The organization is at risk of being dismembered by the platform.

Alternative Models and Solutions

To address these challenges, organizations must consider alternative models for agent development and deployment. Establishing internal platforms can provide a standardized framework for agent creation, fostering greater coordination and reducing fragmentation. Hybrid models that balance centralized control with individual autonomy can also be effective, allowing for the development of core agents that meet common needs while empowering teams to create specialized solutions. Additionally, implementing processes for auditing agent usage and managing their lifecycle can help mitigate the risks associated with digital cruft and ensure alignment with organizational goals.

Conclusion: A Call for Strategic Foresight

The rise of AI agents presents a fascinating case study in the enduring relevance of transaction cost economics. While the potential for increased productivity is real, organizations must proceed with caution. A blind embrace of agent-driven automation could lead to a fragmented, platform-dependent future, undermining the very rationale for the firm’s existence. The challenge for leaders today is to develop a strategic approach to AI adoption, one that leverages the power of automation without sacrificing organizational integrity. This requires a deep understanding of both the technological capabilities of AI and the economic principles that govern organizational structure and success, including those of institutional economics. We must ask ourselves: are we building a future of empowered individuals and organizations, or are we inadvertently creating a new form of digital feudalism, where the platform reigns supreme? The answer will depend on the choices we make today. And those choices are strategic and business model ones, not just technological ones.

By carefully considering the potential risks and adopting a proactive, strategic approach, organizations can harness the power of AI agents while preserving their core strengths and ensuring their long-term success in the evolving digital landscape. It is not just about maximizing productivity through automation, but also about maintaining organizational coherence, adaptability, and strategic autonomy in an increasingly complex and interconnected world. And those challenges are not new, they are just exacerbated by a powerful technology, AI, and by the current dominant business models of AI.

References

1. Aggarwal, R. and Zhao, S. (2009). The diversification discount puzzle: evidence for a transaction‐cost resolution. Financial Review, 44(1), 113-135. [https://doi.org/10.1111/j.1540-6288.2008.00212.]
2. Caballero, G. and Soto-Oñate, D. (2016). Why transaction costs are so relevant in political governance? a new institutional survey. Brazilian Journal of Political Economy, 36(2), 330-352. https://doi.org/10.1590/0101-31572016v36n02a05
3. Dollery, B. (2001). New institutional economics and the analysis of the public sector. Review of Policy Research, 18(1), 185-211. https://doi.org/10.1111/j.1541-1338.2001.tb00973.x
4. Dollery, B. and Leong, W. (1998). Measuring the transaction sector in the australian economy, 1911–1991. Australian Economic History Review, 38(3), 207-231. https://doi.org/10.1111/1467-8446.00031
5. Hart, O. (1995). Firms, contracts, and financial structure. Oxford University Press. https://doi.org/10.1093/0198288816.001.0001
6. Kay N. (2014) Coase and the Contribution of ‘The Nature of the Firm’, Manage. Decis. Econ., 36; pages 44–54, doi: 10.1002/mde.2705.
7. Marty, F., & Warin, T. (2023). Multi-sided platforms and innovation: A competition law perspective. Competition & Change, 27(1), 184-204. https://doi.org/10.1177/10245294221085639
8. North, D. (1990). Institutions, institutional change and economic performance. Cambridge University Press. https://doi.org/10.1017/cbo9780511808678
9. Williamson, O. (1981). The economics of organization: the transaction cost approach. American Journal of Sociology, 87(3), 548-577. https://doi.org/10.1086/227496