The AI gold rush has produced a lot of merchants and very few warlords.
Take, for example, the warlord. When a warlord conquers a country, he does so not only with an army, but with supreme force of will. He picks ground he can defend, fortifies it, and makes it so costly to take that nobody bothers trying. He understands terrain. He knows which valleys are traps. He doesn’t spread his forces thin chasing every glint of light on the horizon. By the time the warlord rides to war, he has already made defeat impossible. His victory was established long before the first blade was drawn - in the depth of his knowledge of the terrain, the strength of his position, the quality of the ground beneath his castle. The war itself is a formality.
Then there’s the merchant. The merchant follows the crowd, sets up where the traffic is, competes on price, and moves when the crowd moves. The merchant has no territory. No depth. No permanence.
The AI industry is full of merchants. They set up shop wherever the latest model API gives them an opening. They compete on UI polish and marketing spend. They have nothing that can’t be replicated by a motivated engineer in a weekend. When the model providers absorb their features - and they will - the merchants fold up their tents and disappear.
The old Internet at least had borders. SEO rankings, domain authority, brand recognition - defensible positions that took years to build. The AI landscape has no such stability. A model update can make your entire product obsolete overnight. An open-source release can destroy a market you spent years building. The ground shifts while you’re trying to lay foundations.
This is why depth matters more than it ever has.
Meditation
Before you chase anything - before you build anything - stop.
New model every week. New framework every month. Some kid on Twitter just shipped in a weekend what took you six months. The hype machine runs on your fear of being left behind, and fear makes you stupid. It makes you reactive. It makes you build the wrong thing fast instead of the right thing deliberately.
In programming we have loops and scopes that control the flow of data, and the mind works in exactly the same way. You’re running loops right now that you’re not conscious of - assumptions about what the market wants, beliefs about what AI can and can’t do, the gnawing feeling that if you don’t ship this quarter you’ll miss the window. Only through stepping back from those loops can you view your venture with critical and honest objectivity. Can what I’m building be replicated by anyone with a credit card and an afternoon? If the answer is yes, you don’t have a business. You have a temporary configuration of APIs.
Don’t ponder too long though, for a good project never really takes its shape until long after release. After all, what fun would it be if you didn’t feel like you were building the wings of the aircraft in mid-flight?
The Stampede
This is the AI gold rush, and the forty-niners are riding out before they know the terrain.
A carpenter found gold flakes in the American River and 300,000 people flooded into California. Most went home broke. Some died in the mud. The people who actually got rich were selling shovels. Sam Brannan, California’s first millionaire, never panned a single nugget - he bought every pan in San Francisco at 20 cents and sold them for $15. Levi Strauss sold trousers. The ones who understood that the real opportunity wasn’t in the gold but in the infrastructure around it - they’re the ones whose names we still know.
Everyone’s an AI company now. The analytics startup pivoted to “AI-powered analytics.” The WordPress agency is offering “AI transformation.” The junior developer with a weekend project is an “AI engineer” on LinkedIn. They’re pouring in from every direction, and the surface gold is already gone.
The wrapper startups - the ones who raised a seed round on the back of a nice UI bolted onto the GPT API - are dying in waves. The features that made them special six months ago are now built into ChatGPT, Claude, and Gemini. When your entire product is a system prompt and a pretty interface, you’re not a company. You’re a feature request waiting to be fulfilled by someone bigger than you.
Builder.ai - Microsoft-backed, valued at $1.2 billion - filed for bankruptcy. Their assistant “Natasha” was supposed to let anyone build an app without writing code. Turns out that promise was somewhat more difficult to deliver than it was to pitch. This isn’t the exception. This is the pattern.
Most of the AI industry has already lost. They rode out before they understood the ground, built on territory they didn’t own, and called it strategy. What follows is for the ones who intend to win before they fight.
Building Your Castle
Most people want the demo. Build the layer under the demo. This is often the most trying time for any venture, so brace yourself.
Your castle is the depth of your understanding and the infrastructure you’ve built from it. It’s what remains when the APIs change, when the models commoditise, when the hype cycle moves on. It can’t be bought with funding or replicated with a prompt.
If your entire stack is a prompt template hitting a third-party API, you have a tent pitched in someone else’s field, and the landlord can raise the rent or evict you whenever he pleases. The real work - the unsexy, demo-unfriendly, thankless work that nobody wants to do - is building the layers between the model and the problem. Orchestration. Evaluation. Failure handling. State management. Eighty percent of enterprise AI implementations fail within their first year, not because the models are bad, but because nobody bothered to build the foundation. They skipped the grinding and went straight to the canvas, and the painting fell apart.
In 2026, a castle is not branding and it is not prompt engineering. It’s control of the workflow. It’s evaluation. It’s failure handling. It’s knowing the domain well enough to tell when the model is wrong, not just when it errors. It’s distribution you actually own, trust you’ve actually earned, and infrastructure that still matters when the model vendors roll the feature into the next release.
The generalist AI company is already dead, most of them just haven’t stopped twitching yet. “We do AI” is not a position. It’s not even a sentence worth finishing. Know your territory so well that you can feel when something is wrong before you can articulate why. That kind of depth takes years to build and cannot be shortcut with a larger model or a bigger budget.
Let your castle be visible. Open source your tools, write about your architecture decisions, share the reasoning behind your approach. This isn’t giving away your advantage - it’s demonstrating it. When people can see the depth of your thinking, they trust you before you’ve spoken. When they can’t, you’re just another logo. The best engineers want to work with people confident enough in their craft to let others examine it. That confidence attracts talent, and talent is the only moat that compounds.
Rallying the Troops
A warlord rallies his troops with fear and promise of spoils untold, but as builders we offer each other something better - the chance to do work that actually matters. You don’t need a big team. You need people who think in systems.
The hardest problems in AI are not model problems. The model is smart enough. The hard problems are everything around it - the thousand ways a production system breaks when you stop watching the demo. These are systems engineering problems, the same kind of thinking that goes into real-time protocols and distributed databases. The people who’ve been solving these problems for decades have a massive advantage right now, whether they realise it or not.
Don’t hire people who are excited about AI. Hire people who are obsessed with the problem you’re solving. The AI-excited will chase the next shiny thing. The problem-obsessed will stay and build the castle.
Going to War
On the Internet, success is measured in conversions, revenue and real estate. Total annihilation of the competition is just a bonus.
When anyone can generate a blog post, a marketing site, or a demo video in minutes, demonstrated competence is the only weapon left. Not the appearance of competence. Actual depth, visible in the work itself. The work can’t be faked, and increasingly that’s the only signal that matters.
Taste in code architecture. Taste in product decisions. Taste in what to build and - critically - what not to build. You can’t automate judgement. You can’t prompt your way to discernment. You develop it through years of building, breaking, and rebuilding. When everything looks the same, the thing that doesn’t is the only thing anyone notices.
Gold rushes end. They always end. The California Gold Rush lasted seven years. The surface gold was gone within two. The AI gold rush started in late 2022. We’re three and a half years in. What remains after a gold rush is the infrastructure - the railroads, the banks, the cities - built by people who were thinking in decades while everyone else was thinking in quarters.
Every tool I’ve ever used has changed. C++ to TypeScript. Raw sockets to WebRTC. REST to MCP. The tools will keep changing. That’s the nature of the game. But the principles underneath don’t change, because they’re not about technology. They’re about how you think, how you see, and whether you have the discipline to establish victory before you ever set foot on the battlefield.
In this world of quick-fix pop culture and black-hat morals, genuine quality has become something of a rarity, but what’s rare is also valuable. Make the depth visible. Open source where it proves your thinking. Publish the reasoning, not just the result. Don’t get suckered by hollow promises and hype. Go deep, and don’t let the noise of a hundred thousand panicking forty-niners shake your conviction, because conviction is the one thing you can’t buy, you can’t fake, and you can’t generate with a prompt.
The gold rush will end. The castle endures.