The internet was supposed to change everything. And it did — for about ten years.

The web democratized access to information at a scale with no historical precedent. Anyone could publish, learn, reach a global audience from a dorm room. Then it consolidated. Search became Google. Social became Facebook. Commerce became Amazon. The open web funneled into a handful of mega-corporations that captured most of the value.

Access was democratized. The benefits were re-centralized. The information was free; the distribution was not.

What's Different This Time

LLMs are triggering a second democratization — but at a different layer.

The internet democratized access to information. LLMs are democratizing what you can do with that information — the ability to act on knowledge, build from it, apply it in context.

That's fundamentally harder to consolidate.

You can centralize a search index. You can build a social graph with network effects so powerful that leaving means leaving your relationships. These are winner-take-all dynamics, and they're why the first democratization re-consolidated.

But you can't centralize domain expertise. You can't aggregate the fabrication shop owner's twenty years of scheduling intuition, or the consultant's understanding of how their clients actually make decisions, or the operator's knowledge of which three fields in a thirty-field form determine the outcome.

That knowledge — specific, contextual, earned through experience — lives in millions of heads. Not in a data center. Not in a training set. And for the first time, it can become software.

The Bottleneck Moved

For decades, the bottleneck was building. Ideas were cheap; turning them into software was expensive. So the people with the deepest domain knowledge were locked out, and the people who could build didn't have the domain knowledge.

The result was platforms. Vendors surveyed the market, averaged everyone's needs, shipped the mean. Good enough for most. Perfect for nobody.

LLMs moved the bottleneck. Building is no longer the constraint — the cost per outcome dropped by orders of magnitude. The bottleneck now is knowing what to build. Not in the product-manager sense. In the deep, contextual, only-comes-from-experience sense.

The fabrication shop owner who knows why standard scheduling tools fail — because they don't account for how material availability cascades through job priorities specific to metal fab. The operator who knows which steps are essential and which are legacy overhead. That knowledge was always valuable. It just wasn't actionable because the cost of turning it into software was too high.

Now it's the most actionable resource in the industry.

Why This Won't Re-Consolidate

The internet consolidated because distribution has network effects. Domain expertise doesn't.

A scheduling tool that perfectly fits one shop's workflow isn't more valuable because a thousand shops use it — it's less valuable, because accommodating a thousand workflows means losing the specificity that made it good. The value is in the fit, not the scale.

That's a structural barrier to re-consolidation. The big platforms can't aggregate this value because aggregation is what destroys it. Turn a bespoke tool into a platform and you're back to every feature being a liability. Back to the mean.

The winners won't be another wave of billion-dollar platforms. They'll be individuals and small teams with domain expertise and taste who now have the tools to turn what they know into software.

The Compound Effect

When you build software that encodes your domain expertise, every improvement compounds your understanding. Your compounding loop belongs to you — not to a vendor whose roadmap you can't control.

The AI that helps you build doesn't just make the first tool cheaper. It makes the tenth tool faster. Every problem you solve creates patterns for the next one. The domain knowledge deepens. The tools sharpen. The gap between what you can build and what a generic platform offers widens over time.

Web platforms created compounding loops that benefited the center — more users, more data, better algorithms. With domain-specific building, the loop benefits the edges. The people closest to the actual work.

What This Means

There's handwringing about whether AI will concentrate more power in Big Tech. Those questions assume the consolidation pattern is inevitable.

It's not. The internet consolidated because distribution could be centralized. LLMs are enabling creation that can't be — because the raw material is domain expertise distributed across millions of people, and it only becomes software when those people have the tools to express it.

Those tools now exist. The people picking them up aren't just developers — they're operators, consultants, shop owners, specialists who've been waiting their entire careers for tools that match how they think.

The first democratization gave everyone access to information and asked them to use someone else's tools to act on it. The second gives them their own.

That's not a prediction. It's already happening.