Date:
23 October 2026
Authors:
Vittorio Checchia, Managing Partner Private Sector BIP
Organizational knowledge and AI: building judgment without hardening corporate memory
Less experience for those joining, more entrenched patterns for those staying: two risks organizations need to manage together.
Artificial intelligence makes accessible in seconds knowledge that used to require research, expert colleagues and time. For organizations the question shifts from where knowledge sits to how it is used to decide, and to what happens to judgment when answers arrive faster than they can be evaluated.
The InSummit 2026 survey offers three qualitative findings. People entering organizations today have fewer chances to gain field experience, because many training activities have moved to machines. That experience is considered decisive in building judgment, the capability needed when facing situations that are not codified or decisions made under uncertainty. And AI widens the gap between the two, because it makes ever more sophisticated results reachable without having gone through the experiences that traditionally built the judgment needed to assess them. The risk is that the quality of outputs grows faster than people’s ability to understand them, verify them and take responsibility for them. The system answers so quickly that it leaves no time to evaluate, critique or understand. This effect is known as experience debt.
The debt and its counterweight
Judgment is formed through cases lived, often in the most repetitive tasks of the early years, such as preparing a file or checking someone else’s work. These are the tasks AI performs best, and delegating them is rational, but a rung is lost on the ladder from junior to senior. It is like the app that sets a runner’s pace: it helps, and it does not replace the training that teaches you to recognize fatigue. Meanwhile AI lowers the cost of producing content and raises the cost of verifying it, and verification takes experience. According to Istat, lack of skills is the obstacle cited by nearly six in ten companies that evaluated AI without adopting it.
The mirror-image risk is inertia. When past procedures and decisions are built into a platform, the system reads new situations through the patterns it was given. Memory feeds the answers, answers feed decisions, decisions return as memory: the organization grows more consistent with itself and less able to see what is changing. “We’ve always done it this way” stops being an attitude and becomes an architecture, replicated at scale.
Debt and inertia show up through simple signals: how long it takes newcomers to reach autonomy on non-standard decisions, review of system answers that is purely formal, built-in procedures nobody has re-examined for years.
An adversarial memory
Organizations need to synchronize three speeds: knowledge, which spreads fast, judgment, which matures slowly, and reality, which changes constantly. The proposed answer is a three-step cycle: turn experience into memory, challenge the memory, feed the outcome back into new memory. The second step is the one that is missing, and it is where AI can help, as a tool that looks for exceptions, points to practices that did not work and to the conditions under which a procedure stops being valid.
The cycle requires four roles. The expert brings depth of evaluation and makes the constraints and assumptions behind past decisions explicit. The newcomer brings distance from the system and can ask why things are done this way. AI amplifies both, if it is designed to challenge as well as answer. The organization has to decide what to make available, what to keep or abandon, and to build a culture where whoever questions a practice is not penalized. Agents should be treated as junior colleagues: they make mistakes without context, and context is the experts’ tacit experience, to be passed on through feedback groups.
Generational turnover is the test bench. When people leave, experience built over decades goes with them, and exit plans deserve the same care as hiring plans: handover alongside, identifying in good time the skills that were never written down, explaining the reasons behind choices as well as the results.
So far we have designed systems to keep organizations from forgetting. Today we must design systems that also help them recognize when what they remember is no longer valid. AI can play a decisive role: not just retrieving the answers of the past, but comparing them with new data, spotting exceptions, highlighting contradictions and suggesting which beliefs deserve to be questioned.