Tacit knowledge and agentic AI

As part of the development of its own offering and the definition of the associated business model, Cobelty conducted a prospective study in a context where the scarcity of specialized skills in international technological standardization and the aging of experts are major challenges for a significant number of organizations: standards -setting and -development bodies, companies, and institutional players. More generally, this study can be extended to organizations whose activities and value proposition are based on intellectual know-how. In knowledge-intensive sectors—engineering, research, design, consulting, and cultural production—the transfer of tacit knowledge (know-how, experience) and rare skills is becoming increasingly difficult to organize, with scarcity also fueling the hunt for skills and expertise, exacerbating asymmetries and even creating new ones. This scarcity of expert human capital particularly undermines the continuity of activities that need to be organized over the long term, structured, and coordinated within complex sectors and ecosystems. It also raises the question of the preservation, formalization, transmission, and dissemination of knowledge, not only at the organizational level but also at the economic level, i.e., taking into account sovereignty stakes and associated capacities. Just as Cobelty has engaged in this exercise in relation to the definition and structuring of its own offering, many organizations will also need to do so. In order to share the benefits of this reflection and its experience, Cobelty has undertaken more extensive forward-looking work related to the deployment of agentic AI systems aimed at exploiting and increasing intellectual know-how, without forgetting the intellectual property dimensions, which will be the subject of future posts.

It is in this context of capacity reinforcement and even rebuilding that the rise of human-agentic artificial intelligence collaboration systems appears to be a technological response that is as promising as it is ambivalent. Indeed, by being able to act relatively autonomously, learn from experience, and cooperate with humans in complex environments, these agents are redefining the ways in which information and knowledge are exploited, produced, and circulated. They help to capture and formalize some of the tacit knowledge of experts, thereby facilitating intergenerational transmission and accelerating learning. But their integration also raises critical issues: what dimensions of knowledge can actually be automated? What forms of judgment, intuition, or discernment remain specifically human? How can collaboration between humans and AI be organized without destabilizing the cognitive and institutional balances within organizations and outside them, within their ecosystems?

When considering transformations, Friedrich Hayek’s 1945 theory on the knowledge problem provides a relevant analytical framework. Hayek showed that knowledge useful for economic coordination is dispersed, local, and tacit, and that decentralized mechanisms (such as the market) ensure its mobilization. However, in the era of transformative artificial intelligence (TAI), which must be considered in terms of its consequences and not just its capabilities, this dispersion is partially reconfigured: certain forms of tacit knowledge become codifiable and actionable by digital systems.

In the very recent volume “The Economics of Transformative AI – K. Agrawal, A. Korinel, E. Brynjolfsson,” Erik Brynjolfsson and Zoë Hitzig describe precisely this shift in their chapter “AI’s Use of Knowledge in Society.” By analyzing TAI as a vector for restructuring the sharing of knowledge and decision-making power, they offer particularly insightful keys to understanding the transformation of knowledge-based organizations. Their analytical framework helps us to think of human-AI collaboration not only as a response to the scarcity of skills, but also as a knowledge governance issue: how can we preserve collective intelligence and knowledge transfer in a context where the capacity to act and reason tends to be distributed between humans and machines? Brynjolfsson and Hitzig provide interesting insights into the continuity of intellectual know-how, training, and new forms of cognitive cooperation in the era of agentic artificial intelligence.

By revisiting Hayek’s knowledge problem, the authors show that AI does not eliminate the dispersion of knowledge, but rather changes its dynamics: knowledge becomes partially codifiable, storable, and transferable through computational architectures that reconfigure modes of collaboration, coordination, decision-making, and learning. In organizations where intellectual know-how is the main productive resource and which are faced with a growing shortage of skills and an aging workforce, AI opens up new possibilities. Without replacing humans, agentic AI solutions, developed to be capable of contextual reasoning and a degree of autonomy, appear to be epistemic partners that can help ensure cognitive continuity for organizations. They make it possible to record, formalize, and reproduce certain tacit knowledge, thereby supporting intergenerational transmission and the acculturation of new professionals. The challenge is not to replace human expertise, but to extend its scope and sustainability in work environments where intellectual capital has already become a critical asset.

However, cognitive automation also carries risks. By codifying and centralizing knowledge, organizations expose themselves to structural dependence on AI systems and a possible erosion of situated expertise: that which resides in human judgment, experience, and intuition. Knowledge governance then becomes a strategic issue, aiming to maintain a balance between codification and embodiment, between algorithmic efficiency and human intelligence. On the other hand, it introduces a risk of cognitive power concentration, where control of data, models, and intelligent workflows becomes a factor of informational domination, both at the company level and at the level of economic ecosystems. The increased centralization of knowledge is also changing the ways in which humans and AI cooperate. Agentic AI, by integrating functions such as planning, coordination, and supervision, tends to “occupy” positions traditionally reserved for human decision-makers. While this redistribution may improve the consistency and responsiveness of processes by increasing capacity, it also threatens cognitive diversity and local deliberation—two essential components of collective intelligence and innovation capacity.

Thus, the development and implementation of agentic AI calls for deep reflection on the reconfiguration of models for organizing intellectual work. They transform human-machine collaboration into a space for the co-evolution of skills, where value no longer lies solely in mastery of expertise, but in the ability to orchestrate cognitive complementarity between human and artificial agents. This is the essence of the approach developed by Cobelty and the thinking behind it.

Extending the perspective of Brynjolfsson & Hitzig, we can conclude that the challenge facing knowledge-intensive companies is not only technological, but fundamentally epistemic and institutional: it is a question of inventing forms of governance capable of preserving the living, interpretative, and transmissible nature of human knowledge, while integrating the learning and action power of agentic AI. This new balance determines both the sustainability of cognitive capital and the resilience of organizations in the face of accelerated transformation of knowledge and influences.