Tacit knowledge is what you know but cannot fully say. Michael Polanyi's formulation is the one that stuck: we can know more than we can tell. It is the knowledge that lets an experienced clinician feel that something is wrong before the tests confirm it, and a good editor know a sentence is off before articulating why. It is acquired by doing, transmitted by proximity. It is the form of knowledge most exposed by AI, because it is the form least likely to be in any training corpus.
Definition
Tacit knowledge: knowledge that resists full articulation, acquired through experience and practice rather than instruction, and typically transmitted through shared work rather than documentation. Contrasted with explicit knowledge, which can be written down, and therefore copied, taught and trained on.
Where it comes from
Polanyi introduced the idea in Personal Knowledge (1958) and The Tacit Dimension (1966). His examples are ordinary and hard to dismiss: recognising a face among thousands without being able to describe how, riding a bicycle without being able to state the balancing rules.
The concept was taken into management by Nonaka and Takeuchi in the 1990s, who made it central to how organisations create knowledge, and argued that the tacit-to-tacit transfer happening through apprenticeship and shared work is a primary mechanism of organisational capability.
Why AI puts it under pressure
Language models are trained on what has been written down. That is, by construction, the explicit portion of human knowledge. They are extraordinarily good at it, and that is what makes the boundary matter.
Three consequences follow.
The explicit portion of a job commoditises fastest. Whatever could be documented is now cheap. What remains scarce is disproportionately the part nobody wrote down.
Tacit knowledge is what verification runs on. Knowing an answer is subtly wrong, before you can say why, is a tacit judgement. That is why verification is expertise applied rather than a procedure that can be delegated to someone junior with a checklist.
Its transmission route is the one being automated. Tacit knowledge passes through shared work: the junior doing the task badly, the senior correcting it, the accumulated exposure to cases. Remove the junior task and you have not just removed work, you have removed the channel. That is the mechanism underneath the missing rungs, and the reason better documentation cannot fix the loss.
The boundary keeps moving
The boundary is not fixed, and this research has been wrong about such boundaries before. A great deal of what was considered tacit, medical pattern recognition and stylistic judgement among it, has turned out to be at least partly learnable from enough examples. Assuming any particular capability is permanently beyond a model is not a safe position.
There is also a fair objection to the concept itself: tacit is sometimes used to mean genuinely inarticulable and sometimes to mean not yet articulated, and the two have very different implications. Much of what organisations call tacit is simply undocumented, which is a solvable problem rather than a fundamental one.
Stop defending the preserve
The useful move is to notice that tacit knowledge is generated by a process, and organisations are dismantling the process while assuming the stock will last.
Explicit knowledge can be bought, copied and trained on. Tacit knowledge has to be grown, in people, through repetitions, over years. An organisation that automates the repetitions ceases to produce tacit knowledge rather than converting it into the explicit kind, and the effect will be invisible for about five years, which is roughly how long the existing stock lasts.
Related SuperSkills research
On the transmission failure, the missing rungs and the missed reps. On the organisational stock, capability debt. On what it does to verification, who owns verification. On the boundary, what stays human.
Key sources
- Ericsson, K. A. et al. (1993). The role of deliberate practice in the acquisition of expert performance. Psychological Review, 100(3).
- Macnamara, B. N. and Maitra, M. (2019). Revisiting Ericsson. Royal Society Open Science, 6(8).
- Autor, D. (2024). Applying AI to Rebuild Middle Class Jobs. NBER Working Paper 32140.
- Budzyń, K. et al. (2025). Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy.
About this definition
Tacit knowledge is Michael Polanyi's concept and is not a SuperSkills coinage. Rahim Hirji is the author of SuperSkills (Kogan Page, 2026). Reviewed quarterly.
Cite this
Hirji, R. (2026). What is tacit knowledge? The SuperSkills Intelligence Company. Last reviewed 26 August 2026. thesuperskills.com/research/what-is-tacit-knowledge