In 1997, I was teaching myself HTML. Not at a school. Not in a class. In my bedroom, on a machine that took four minutes to boot, with a 28.8 modem screaming its way onto the internet. I was fourteen years old and I thought I had discovered fire.
My girlfriend at the time was doing the same thing. We would trade pages back and forth — bad layouts, broken tables, colors that should never have been put next to each other. But we were building things. Putting them on the internet. Making something out of nothing with a text editor and a browser.
And I remember thinking: in a few years, everyone will know how to do this.
It seemed obvious. HTML was not hard. You could learn the basics in an afternoon. The tools were free. The internet was everywhere. Why would anyone not learn this?
They Did Not Learn It
My niece was born in 1998. One year after I wrote my first tag. She grew up with the internet in her pocket. She uses it more than I do — more hours, more apps, more platforms. If you asked her the difference between HTML and CSS, she would look at you the way I look at someone asking about carburetor tuning. The words register. The meaning does not.
And she is not unusual. She is normal. She is the overwhelming majority.
Twenty-nine years later, HTML is still the backbone of every page on the internet. And almost nobody knows how to write it. Not because it is too hard. Because most people never needed to close the gap between using a thing and understanding how it works.
I think about this every time someone tells me Claude Code is going to change everything.
The Moat I Did Not See
I have been using Claude Code daily for months. I have published over a hundred posts through it, managed multi-site deployments, rebuilt infrastructure, shipped features I would have needed a team for. It has fundamentally changed how I work. I wrote about the 43:1 cost ratio. I wrote about operating at a different altitude. I believe every word of it.
And I caught myself making the same mistake I made in 1997.
I assumed this was obvious. That once people saw what was possible, they would close the gap themselves — learn the tool, adopt the workflow, start operating like a team of one. That the power was so self-evident, adoption would be automatic.
It will not be.
Not because the tool is bad. Not because people are incapable. Because the distance between "this is possible" and "I have done the work to make it mine" is a canyon most people will never cross. Not out of laziness. Out of the same reasonable logic that kept my niece from learning HTML: she never needed to. The world gave her interfaces that hid the complexity. She clicked buttons. The buttons worked. Why would she look underneath?
The Curve Is the Moat
Here is what I did not appreciate at the beginning: the learning curve is the advantage.
Claude Code is not a button. It is a conversation with a system that requires you to know what you want, describe it clearly, and make judgment calls when the output is not right. That is not a trivial skill. It is the skill of directing work — of being the conductor, not the audience. And it requires something that cannot be installed or downloaded: enough domain knowledge to know when the model is wrong.
When I sit down and tell Claude to update RSS feeds across four XML files, add a blog entry with the right category tags, and commit the changes — I am not doing something anyone could do by reading a tutorial. I am applying twenty-nine years of knowing how this stuff works under the hood. The tool amplifies that knowledge. It does not replace it.
That is the part the "AI will democratize everything" narrative misses. Democratization requires the recipient to meet the tool halfway. And most people — reasonably, understandably — will not.
OpenClaw and the Power User Gap
I wrote a while back about OpenClaw — an agent framework that lets you build entire operating systems out of plain-English prompts. CRM, email triage, morning briefings, security monitoring. Twenty-six prompts that replace a small company's worth of software.
It hit 200,000 GitHub stars. The developer got acqui-hired by OpenAI. The project was everywhere for a few weeks.
And then it was not. Because the people who could actually use it were the people who already understood what a CRM does, how email routing works, what a webhook is, and why you would want a local-first architecture. The people who needed it most — the ones drowning in manual work — did not know where to start. Not because the prompts were in English. Because knowing what to ask for requires a map of the territory the tool is designed to traverse.
That is the pattern. Every generation of tooling follows it. HTML. WordPress. No-code platforms. Now agentic AI. The promise is always "anyone can do this." The reality is always "anyone who already understands the underlying domain can do this faster."
Why I Am Not Worried Anymore
For a while, this bothered me. I thought: if everyone gets Claude Code, what is my edge? If the tool is available to all, the advantage disappears. The moat drains.
But the moat was never the tool. The moat is the twenty-nine years of pattern recognition that tells me what to build, how to describe it, and when the output is wrong. The moat is the domain knowledge that turns a conversation with a model into a production-quality result. The moat is having sat with enough friction, for long enough, that the friction became fluency.
My niece has had access to free HTML tutorials for her entire adult life. She has never written a tag. Not because the information was gated. Because the motivation to push through the learning curve was never there. The abstraction layer — the apps, the platforms, the buttons — was good enough.
Claude Code will have its abstraction layer too. Simpler interfaces will emerge. Buttons will appear. Most people will use those and never look underneath. And that is fine. That is how technology has always worked.
But the people who look underneath — who learn to direct the model, who develop taste for what good output looks like, who build the instinct for when to trust and when to override — those people will operate at a level that looks like magic to everyone else. The same way a developer in 2026 looks at a well-built website and sees the structure, while everyone else just sees the page.
The 1997 Lesson
I was wrong in 1997. I thought HTML would be universal. It became infrastructure — invisible, essential, and understood by a small fraction of the people who depend on it every day.
I suspect Claude Code, and tools like it, will follow the same arc. Not a revolution that lifts everyone equally. A lever that multiplies the capability of anyone willing to learn how to use it. The gap between those who adapt and those who do not will not shrink. It will widen. And the people on the adapt side will build things that would have required entire teams a year ago.
That is not elitism. It is pattern recognition. The same pattern, playing out on a thirty-year loop.
The Protocol: The tool is available to everyone. The willingness to sit with it until it becomes fluent is not. That was true for HTML in 1997. It is true for Claude Code in 2026. The moat is not access. It is adaptation. And adaptation is a choice most people will reasonably decline — which is exactly what makes it valuable for those who do not.
