Manufacturing sites have always had people who try to make their own work a little easier. When repetitive work piles up, they build Excel macros. Engineers who are comfortable with code write Python scripts or small programs for themselves. AI is increasingly taking that role today, but what happens on the shop floor is not as different from the past as it may seem.
Because people who understand the operation solve their own problems, the effect can be immediate. A task that used to take hours can shrink to minutes, and checks that were repeated by hand can be automated. But whether those tools have truly become part of the team's way of working often becomes clear only after the person who built them leaves.
Tacit knowledge does not disappear. It moves into code.
The Excel files and code remain after a person leaves. The formulas are still there, and so are the thresholds. What is often missing is why those values were chosen, when exceptions were made, and what the engineer was looking at when the decision was made. After enough time passes, someone else may still be able to read the code without being able to understand the judgment behind it.
The tacit knowledge that used to live in someone's head has not been eliminated. It has simply moved into software that one person created. Productivity may improve immediately, but the knowledge truly remains in the organization only when other people can understand and modify it without the original author.
Tacit knowledge also has an expiration date
There is another problem. Manufacturing knowledge is not information that can be documented once and then reused forever in exactly the same way. A judgment that was correct several years ago is not guaranteed to be correct today. Equipment condition changes, materials and products change, and process conditions and metrology methods keep changing as well. Even when a symptom looks similar, its cause may be different now.
That means neither blindly reusing old knowledge nor discarding it simply because it is old is a good approach. We need to understand the conditions under which a judgment was made, compare them with the present, reuse what still applies, and reconsider what has changed.
Past experience needs context before it can be reused
If all that remains from a process event is that a correction value was changed, the next engineer inherits only a rule. To reuse the experience, we also need to know what product was being produced, what condition the equipment was in, what changes appeared in the data, and how the outcome changed after the decision.
Experienced engineers already use knowledge this way. When a new problem appears, they do not simply copy an old answer. They first ask, “Something similar happened before, but what is different this time?” The value of past experience is not only in remembering the answer. It grows when enough context remains to compare that experience with the current situation.
This is also a data-analysis problem
As the number of tools, products and processes grows, making these comparisons entirely from memory becomes increasingly difficult. Engineers need to find whether a similar problem occurred before, align the conditions from then and now, and determine which differences affected the result. The more experience an organization accumulates, the more reference cases it has—and the harder it becomes to manage them through memory alone.
This also makes the role of AI clearer. AI can do more than retrieve old knowledge. It can identify the conditions behind a past decision, compare them with current data, and surface what has changed. The final decision may still belong to an engineer or to existing analytical logic, but AI can greatly expand the range of comparisons that people previously had to make in their heads.
What it means to preserve knowledge
Digitizing tacit knowledge in manufacturing matters. But if recording becomes the goal in itself, the result may simply be another outdated manual or another piece of ownerless code. People need to be able to understand why a decision was made and, as time passes, verify whether it is still valid.
Research supported by MIT's Initiative for New Manufacturing on AI adoption in engineering and manufacturing also points to two related problems: expert knowledge often remains with individuals, and knowledge captured once can become outdated as materials and manufacturing methods change.