Your Industry Knowledge Is the Moat, Because It Was Never Written Down

There is a reasonable worry that domain expertise is the next thing to commoditise. If a model can read every published paper, every industry report, and every forum thread in your field, what is left that you know and it does not?

Quite a lot, as it turns out, and the reason is structural rather than temporary.

The publication gap

Models learn from what was written down. In most industries, the operationally important knowledge was never written down, for three reasons. It was tacit, so nobody thought to record it. It was commercially sensitive, so nobody was allowed to. Or it was embarrassing, so nobody wanted to.

The third category is larger than people admit and disproportionately valuable. Nobody publishes a case study called 'how our documentation process quietly loses 3% of billings'. Nobody writes a report titled 'why our lane rates are eight points above market and we did not notice for two years'. These problems are extremely common, extremely expensive, and almost entirely absent from the written record.

What tacit knowledge actually looks like

Tacit knowledge is usually a pattern-matching shortcut that the person holding it cannot fully explain. A site superintendent who can walk a floor and know which trades will cause the handover delay is not applying a documented method. They are recognising a configuration they have seen forty times.

“It feels like common sense because everyone you work with knows it too. It is not generic. It is local, and the locality is what makes it valuable outside.”

That recognition is worth encoding, and encoding it is the entire opportunity. The person cannot articulate the rule, but they can label examples, which is enough. Show them thirty situations and ask which ones worry them and why, and a rule emerges that exists nowhere else.

Why this does not erode quickly

It feels like common sense because everyone you work with knows it too.

The obvious counter is that this knowledge will eventually get captured. It will, in aggregate, over long periods. But the mechanism for capturing it requires someone to write it down, and the people who hold it have no incentive to publish it and every incentive to build with it.

There is also a replenishment effect. Industries keep changing, which keeps generating new tacit knowledge faster than the old knowledge gets documented. The operational reality of running an outpatient clinic in 2026 differs from 2021 in ways that no published source captures yet, because the people living it are busy living it.

The extraction problem

The hard part is not that the knowledge is inaccessible. It is that the person holding it does not perceive it as knowledge. It feels like common sense, because within their context it is common sense. Everyone they work with knows it too, which makes it feel worthless.

This is the single most common reason experienced professionals do not build anything. They assume that because their immediate colleagues share their understanding, the understanding is generic. It is not generic. It is local, and the locality is exactly what makes it valuable outside.

Turning it into something

The extraction method that works is specific rather than abstract. Do not ask yourself what you know. Ask yourself what you have watched go wrong repeatedly, what you check first when you inherit a mess, and what you would tell a competent replacement in their first week that they could not read anywhere.

Those three questions produce a list. The list is your defensible material. Everything else in a venture, including the software, is now cheap enough to be a commodity, which means the list is the whole business.

Read what AI cannot automate for the longer argument, or start with the insight engine if you want the structured version of the extraction.