- Yale Professor K. Sudhir argues that many businesses are underestimating a critical factor when deploying AI Agents: the implicit rules that run an organization but do not exist in formal data or official processes.
- The article opens with the example of a wealthy client requesting an update to their investment account beneficiary. The AI system handles it flawlessly: classifying the request, completing the paperwork, and sending an automated confirmation. However, an experienced financial advisor would recognize this as a signal that the client is considering moving their assets to another institution.
- A month later, the client actually leaves the firm.
- No AI made a mistake, yet the business still made the wrong decision because it missed the implicit signals that humans typically perceive.
- Sudhir calls this invisible layer of capability the “implicit organization.”
- According to him, two operating systems coexist simultaneously in every business.
- The first operating system consists of formal processes, policies, organizational charts, and documented data.
- The second operating system is the network of tacit knowledge, personal motivations, experience, and judgment that employees use every day.
- The implicit organization performs three functions that formal systems struggle to replace: coordination, motivation, and self-restraint.
- Coordination occurs when employees exchange information on their own initiative without being required to do so by a formal process.
- Motivation stems from culture, professional reputation, and social relationships within the company.
- Self-restraint is the ability to stop when feeling that something is wrong, even when one possesses the formal authority to execute that action.
- Sudhir argues that current AI Agents can access data and processes very quickly, but they do not automatically inherit these implicit factors.
- When businesses replace employees with AI, they are not just changing who performs the task; they are also eliminating many underlying motivation and control mechanisms that operate silently beneath the surface.
- The author cites the example of credit underwriters. An AI can optimize loan approval speed but cannot inherently understand the relationship risks or the industry experience that experts accumulate over many years.
- Ramp is mentioned as a positive example, designing AI Agents to handle the majority of requests but routing difficult cases to human judgment. Those human decisions are then used to improve the system over time.
- McKinsey with its Lilli system and Estée Lauder with ConsumerIQ have succeeded in partially digitizing organizational knowledge.
- However, the author emphasizes that knowledge retrieval does not equate to judgment capability.
- Multi-AI agent systems are particularly prone to failure because they lack a shared history, lack experience, and have no single entity taking ultimate accountability.
- Sudhir cites research showing that the failure rate of multi-agent systems can range from 40% to 80%.
- He believes businesses have only three choices: inserting AI into current processes, redesigning based on old processes, or redesigning based on a thorough understanding of the implicit organization.
- The third option is seen as the only effective path forward.
- Another long-term risk is that AI could disrupt the training pipeline for young talent. Many entry-level jobs are precisely where employees learn judgment, absorb corporate culture, and accumulate experience to become future leaders. If AI completely replaces those roles, businesses may lose the pipeline of human resources capable of managing AI in the future.
- Sudhir recommends that businesses must proactively redesign apprenticeship, supervision, and AI-critique systems instead of solely focusing on automation.
- 📌 Conclusion: The greatest challenge of AI Agents is not the technology, but the invisible factors that help businesses operate effectively every day. K. Sudhir calls this the “implicit organization,” comprising experiential knowledge, professional motivation, and judgment capabilities that no system fully records. When AI replaces humans, businesses risk inadvertently eliminating the very mechanisms protecting them from flawed decisions. According to the author, companies that succeed with AI will not be those that automate the most, but those that understand best what cannot be automated.
AI Agents can follow every process perfectly but still cause businesses to lose major customers
Related Posts
VIET NAM CONSULTING AND MEASUREMENT JOINT STOCK COMPANY
Contact
Email: info@vietmetric.vn
Address: No. 34, Alley 91, Tran Duy Hung Street, Yen Hoa Ward, Hanoi City
© 2026 Vietmetric