The Thinking Company

According to McKinsey, 88% of the participants in its 2025 study said their organisation regularly used AI in at least one function. Among the organisations already using it, 7% said they had fully deployed it and integrated it across the enterprise.

BCG found that 5% of the more than 1,250 companies in its study were achieving value from AI at scale, while 60% reported scarce material gains. McKinsey and BCG study different samples and outcomes; from different angles, their figures describe the distance between adopting AI and turning it into an enterprise capability.

In your own company, professionals who already use assistants may coexist with teams that build agents, processes that incorporate automations and suppliers that add AI to their products. Leadership needs to understand which decisions are changing, which results are improving and what capability stays inside the organisation after each initiative.

An agent can consult information, use tools, coordinate with other agents and act on systems. As its capabilities expand, so do the consequences of a wrong criterion, an excessive permission or a decision that cannot be reconstructed.

In a Deloitte survey of more than 500 business and technology leaders in the United States, 5% considered their processes very well prepared for agents and 15% said they had scaled coordinated multi-agent systems across functions. 70% reported difficulties in trusting and governing agents. These are perceptions and situations declared by the participants, not a universal measure of business readiness.

That is why every step forward needs a purpose, an accountable owner, limits of action, verification proportionate to the risk, traceability and the capacity for human intervention. The knowledge that feeds agents also needs criteria for currency, quality and access.

Principle of the method: an agent's autonomy must remain within the company's capacity to understand it, verify it, govern it and stop it.

A company coordinates knowledge held by people, clients, processes, systems and suppliers. Part of that knowledge is documented. Another part appears when a professional resolves an exception, interprets a difficult situation or corrects a decision.

Agentic AI makes it possible to extend that working capacity. The Thinking Company method establishes how to connect agents and people with reliable knowledge, rules, memory, verification and governance. In this way, useful experience can move from an isolated action to a capability that other teams use and improve.

Roche offers an example in research: the company describes how it connects biological and chemical experiments with its AI models in its Lab-in-the-Loop system. The results of experimentation help its scientists test new hypotheses. Its publication describes the system and its goals; that source does not give a figure for the return achieved.