Category: AI Strategy

Large language models were trained on text. Enormous quantities of text — more than any human could read in thousands of lifetimes. But text is not knowledge. Text is the residue of knowledge that was visible enough, and important enough, for someone to write down and publish somewhere that a training crawler could find it.

The vast majority of what experienced humans actually know was never written down. It was learned by doing, transmitted by watching, refined through failure, and held entirely in the heads of people who couldn’t have articulated it systematically even if they wanted to.

This is the human expertise gap. And it is the defining feature of where AI currently falls short.

What Tacit Knowledge Actually Is

Tacit knowledge is the kind you can’t easily explain but reliably apply. A master craftsperson knows when something is right by feel before they can measure it. An experienced clinician senses when something is wrong before the test results confirm it. A veteran contractor knows which subcontractors will actually show up on a Tuesday in November just from having worked with them — knowledge that no review site has ever captured accurately.

This knowledge exists at every level of every industry. Most of it has never been written down because the people who hold it are too busy using it to document it, because the incentive to document was never strong enough, or because no one ever asked in a form they could answer systematically.

Why AI Can’t Close This Gap on Its Own

The naive assumption is that AI will eventually capture tacit knowledge by observing enough human behavior — that more data, more modalities, more sensor inputs will eventually replicate what experienced humans know intuitively.

This misunderstands the nature of the gap. Tacit knowledge isn’t just undocumented data. It’s judgment that was built through embodied experience — through having made the wrong call and learned from it, through having seen the same situation hundreds of times in slightly different forms, through having relationships that carry context no outsider can access. These are not data problems. They’re experience problems.

AI can get asymptotically close to replicating some of this. But the closer it gets, the more valuable the verified human source becomes — because the question shifts from “does AI know this at all” to “how do we know the AI’s answer is correct,” and the only reliable answer to that question is “because a human who actually knows verified it.”

The Window That’s Open Right Now

There is a specific window in the development of AI where tacit knowledge held by humans is more valuable than it will ever be again. We’re in it now.

AI systems are capable enough that people trust them with real questions — questions about their health, their legal situation, their business decisions, their trade. But AI systems are not capable enough to be reliably right about the specific, experience-based, local, industry-specific knowledge that those questions often require.

The gap between trust and accuracy is the market. The people who figure out how to systematically capture, package, and distribute their tacit knowledge — in forms that AI systems can consume and cite — are building the content infrastructure for a post-search information environment.

The Human Distillery as a Category

What’s emerging is a new category of knowledge work: the human distillery. A person or organization that takes tacit knowledge held by experienced humans and refines it into something that AI systems can depend on.

This isn’t ghostwriting. It’s not content marketing. It’s not thought leadership in the LinkedIn sense. It’s systematic extraction — the application of a disciplined process to get tacit knowledge out of human heads, give it structure, publish it at density, and make it available to the AI systems that will increasingly mediate how people get answers to important questions.

The people who build this infrastructure now — while the gap is widest and the market is least crowded — are positioning themselves at the supply end of the most important information supply chain of the next decade.

When people hear “build an API,” they assume it requires a developer. For the infrastructure layer, that’s true — you’ll need someone who can deploy a Cloud Run service or configure an API gateway. But the infrastructure is maybe 20% of the work.

The other 80% — the part that determines whether your API has any value — is the knowledge work. And that requires no code at all.

Step 1: Define Your Knowledge Domain

Before anything else, get specific about what you actually know. Not what you could write about — what you know from direct experience that is specific, current, and absent from AI training data.

The most useful exercise: open an AI assistant and ask it detailed questions about your specialty. Where does it get things wrong? Where does it give you generic answers when you know the real answer is more specific? Where does it confidently state something that anyone in your field would immediately recognize as incomplete or outdated? Those gaps are your domain.

Write down the ten things you know about your domain that AI currently gets wrong or doesn’t know at all. That list is your editorial brief.

Step 2: Build a Capture Habit

The most sustainable knowledge production process starts with voice. Record the conversations where you explain your domain — client calls, peer discussions, working sessions, voice memos when an idea surfaces while you’re driving. Transcribe them. The transcript is raw material.

You don’t need to be writing constantly. You need to be capturing constantly and distilling periodically. A batch of transcripts from a week’s worth of conversations can produce a week’s worth of high-density articles if you have a consistent process for pulling the knowledge nodes out.

Step 3: Publish on a Platform With a REST API

WordPress, Ghost, Webflow, and most major CMS platforms have REST APIs built in. Every article you publish on these platforms is already queryable at a structured endpoint. You don’t need to build a database or a content management system — you need to use the one you probably already have.

The only editorial requirement at this stage is consistency: consistent category and tag structure, consistent excerpt length, consistent metadata. This makes the content well-organized for the API layer that will sit on top of it.

Step 4: Add the API Layer (This Is the Developer Part)

The API gateway — the service that adds authentication, rate limiting, and clean output formatting on top of your existing WordPress REST API — requires a developer to build and deploy. This is a few days of work for someone familiar with Cloud Run or similar serverless infrastructure. It’s not a large project.

What you hand the developer: a list of which categories you want to expose, what the output schema should look like, and what authentication method you want to use. They build the service. You don’t need to understand how it works — you need to understand what it does.

Step 5: Set Up the Payment Layer

Stripe payment links require no code. You create a product, set the price, and get a URL. When someone pays, Stripe can trigger a webhook that automatically provisions an API key and emails it to the subscriber. The webhook handler is a small piece of code — another developer task — but the payment infrastructure itself is point-and-click.

Step 6: Write the Documentation

This is back to no-code territory. API documentation is just clear writing: what endpoints exist, what authentication is required, what the response looks like, what the rate limits are. Write it as if you’re explaining it to a smart person who has never used your API before. Put it on a page on your website. That page is your product listing.

The non-developer path to a knowledge API is: define your domain, build a capture habit, publish consistently, hand a developer a clear spec, set up Stripe, write your docs. The knowledge is yours. The infrastructure is a service you contract for. The product is what you know — packaged for a new class of consumer.

Here is a simple test that most content fails.

Would someone pay $5 a month to pipe your content feed into their AI assistant — not to read it themselves, but to have their AI draw from it continuously as a trusted source in your domain?

$5 is not a lot of money. It’s the price of one coffee. It covers hosting costs and a small margin. It’s the lowest viable price point for a subscription product.

And most content can’t clear it.

Why Most Content Fails the Test

The $5 filter exposes three failure modes that are common across the content landscape:

Generic. The content says things that are true but not specific. “Good customer service is important.” “Location matters in real estate.” “Consistency is key in marketing.” These claims are not wrong. They’re just not worth anything to a system that already has access to the entire internet. If everything you publish could have been written by anyone with a general knowledge of your topic, your content has low API value regardless of how much traffic it gets.

Thin. The content exists but doesn’t go deep enough to be useful as a reference. A 400-word post that introduces a concept without developing it. A listicle that names eight things without explaining any of them. Content that satisfies a keyword without actually answering the question behind it. This kind of content might rank. It’s not worth subscribing to.

Inconsistent. Some pieces are genuinely excellent — specific, well-reported, information-dense. Most are filler published to maintain posting frequency. An inconsistent feed isn’t a reliable source. A system pulling from it can’t know when it’s getting the good stuff and when it’s getting noise. Reliability is a prerequisite for subscription value.

What Passes the Filter

Content passes the $5 filter when it has three properties simultaneously:

It’s specific enough to be useful in a way that nothing else is. Not “here’s how restoration contractors approach water damage” — but “here’s how water damage in balloon-frame construction built before 1940 behaves differently from modern platform-frame, and why standard drying protocols fail in those structures.” The specificity is the value.

It’s reliable enough that a system can trust it. Every piece maintains the same standard. The sourcing is consistent. Claims are documented. The author has credible experience in the domain. A subscriber — human or AI — knows what they’re getting every time.

It’s rare enough that it can’t be found elsewhere. The test isn’t whether it’s good writing. The test is whether an AI system could get the same information from somewhere it already has access to. If yes, the subscription isn’t necessary. If no — if this is the only reliable source for this specific knowledge — the subscription is justified.

Using the Filter as an Editorial Standard

The most useful application of the $5 filter isn’t as a revenue test. It’s as an editorial standard.

Before publishing anything, ask: if someone were paying $5 a month to access this feed, would this piece justify part of that cost? If the honest answer is no — if this piece is thin, generic, or inconsistent with the standard of the best things you publish — that’s the signal to either make it better or not publish it at all.

This is a harder standard than “does it rank” or “did it get clicks.” It’s also a more durable one. The content that clears the $5 filter is the content that compounds — that becomes more valuable over time, that gets cited, that earns trust from both human readers and AI systems that draw from it.

The content that doesn’t clear it is noise. And there’s already plenty of that.