The current AI moment looks like chaos from the outside. From the inside, it’s just a product development cycle: raw, messy, and completely normal.
To understand what’s actually happening in AI right now, you have to stop watching it like a consumer or an investor and start thinking like an engineer.
What the rest of the world is experiencing as a confusing storm of hype, breakthroughs, and disappointment is, to anyone who has actually shipped a product, deeply familiar. It’s the development cycle. It’s the long, ugly middle stretch between “we have an idea” and “we have something mature that works.” Engineers live in that stretch every day. The only thing different this time is that the entire world is standing inside the lab with us, watching the prototypes spark and burn out in real time.
Some people are trying to invent something themselves. Most are trying to apply what developers have already built. But what those developers have built isn’t a finished product either. It’s a prototype of an idea, pushed out into the open for anyone to use. The thing you’re reaching for is still an experiment; it just happens to be one with a download link. Either way, everything you’re touching today is raw. These are Edison’s first lightbulbs: dim, fragile, dying fast. Nobody looked at those early bulbs and concluded that electric light was a scam. They understood they were looking at step one of something. We’re at step one. Act accordingly.
The spreadsheet mindset is the wrong mindset
Here’s where a lot of people have already lost without realizing it: the ones trying to slap a price tag on research and development, tally the costs in an Excel sheet, and ride the whole thing to quick riches.
That’s the wrong mindset, full stop. R&D doesn’t hand you a clean return on investment while you’re still inside it. That’s not a flaw in the process. That is the process. You spend, you experiment, you throw away most of what you try, and somewhere down the line a working product falls out the other end.
This is exactly why the spreadsheet guys shouldn’t be running technical organizations. They optimize, they cut, they “rationalize.” What they don’t do is create. They’re very good at destroying things and very bad at building them, because they fundamentally don’t understand the world they’ve been put in charge of. You don’t reach an invention by cutting costs.
None of this means price tags are forbidden forever. There’s a right time for them: when the product is actually finished. At that point you can count both what it cost to develop and what it will cost to produce at scale, and the numbers finally mean something. The mistake is reaching for that calculation while you’re still in the part where nobody even knows what the product is yet.
And here’s the catch the spreadsheet crowd almost always misses: sometimes producing something at scale is its own research problem. Apple didn’t just design the first iPhone and hand it to a factory. They had to invent new manufacturing processes to build it in the volumes they needed. The research didn’t stop when the prototype worked. Getting from “one of these exists” to “millions of these exist” was a second wave of R&D, with its own dead ends and its own costs that nobody could have written down in advance.
Copying is easy. Everything after is slow.
Take n8n as a concrete example. The core idea, visual workflow automation, wasn’t new. That category already existed; the concept had been done before. And that’s the point: copying an idea is easy, and it happens fast. Anyone can clone a concept.
What’s hard is everything that comes after. The next steps, the real differentiation, the thing that turns a copy into a product with its own gravity: all of that happens through trial and error, and trial and error takes time. You can’t budget it into a single quarter. You can’t shortcut it. You just have to grind through the dead ends.
Most people never see that part, because most people only use and test what developers have already put in front of them. Their entire sense of “what AI can do” is bounded by whatever someone else shipped last month. They mistake the edge of the available tools for the edge of what’s actually possible. Those are very different edges.
“It failed” usually means “I got impatient”
So now the people who showed up expecting instant miracles, the ones who wanted the magic to arrive fully formed and for free, are drifting off and announcing that the whole thing has failed.
It didn’t fail. Their fantasy failed. They were never really evaluating the technology; they were evaluating their own patience, and it ran out. The tools are exactly where you’d expect tools to be this early. The disappointment says far more about the expectation than about the engineering.
Why the giants pay insane money for tiny startups
Watch what the serious players actually do with their money. Why do people like Elon, and plenty of others, buy startups at prices that make no sense on paper?
Not for the idea. Not for the product. For the talent. There are countless cases where an entire company gets bought purely to land one person on a team. That should tell you where the real value lives in this field. It isn’t in the pitch deck or the patent. It’s in the small number of people who genuinely understand how to build here. Everything else is replaceable. The right human is not.
Nobody ships Edison’s fifteenth bulb
Now watch what’s happening on top of all this. People are starting to talk seriously about security, about hardening these tools and dropping them straight into production systems that real businesses depend on.
Picture the equivalent. Imagine someone grabbing Edison’s fifteenth lightbulb prototype, the one still flickering and burning out on the workbench, and wiring it into a building’s electrical system as if it were finished, certified hardware. It sounds insane. It is insane. And it’s exactly what a lot of people are doing with AI right now.
The order matters, and there’s only one order that works. First you build a product that actually holds together. Only then does it make sense to ask how you secure it and how you run it in production at scale. Trying to do all three at once, on top of a prototype that’s still changing under you every month, isn’t ambition. It’s building on sand. And getting to the point where this stuff is genuinely ready for production, properly secured and properly deployed, could easily take more than ten years.
You have to think in decades
Here’s the part almost nobody wants to hear: with AI, you have to think in terms of at least ten years.
“Overnight success” is, almost without exception, ten years of brutal work and wild investment that you never saw. And the uncomfortable truth underneath that is simple. Very few people are actually capable of grinding hard for ten straight years. That’s precisely why there are so few genuinely successful people in the world. It was never mainly about talent, timing, or a clever idea. It’s about who’s still standing, still building, after everyone else got bored and left.
Put a number on it and it’s stark: something like ninety percent will fail, and maybe ten percent will make it through. The ten percent are the engineers and the innovators, and they’re not the smartest people in the room. They’re the ones willing to grind through sleepless nights, keep going long after it stops being fun, and cut out the noise around them so they can stay locked on a single direction. Focus is the whole game. Almost everything else is a distraction dressed up as an opportunity.
Ask the people who actually built something hard, and you’ll hear the same confession over and over: if they had known at the start how difficult it would be, and how expensive, they would never have begun in the first place. The true cost only becomes visible from the far side, once it’s already been paid. That’s the cruel joke buried inside every demand for a cost estimate up front. If anyone could really produce that number at the beginning, it would be frightening enough to kill the project on the spot, and the thing would never get built at all. A certain amount of not knowing what you’re in for is exactly what lets anything ambitious begin.
Nothing real gets built without sacrifice. If you’re not willing to pay that, the technology was never the thing standing in your way.
Boom or real?
A few years ago I went to NVIDIA’s GTC, and I came home with an answer to the only question worth arguing about here: is AI a boom, or is it real?
What I understood is that it’s exactly what you think it is. It comes down entirely to the direction you’ve chosen. For the people building in the right direction, it’s a real and lasting shift. For the people who aren’t, it’s an empty bubble. Same technology, two completely different realities, and the thing that decides which one you get is you.
And that’s the trap. If you pick the wrong direction, you will see exactly what you already believe. You’ll collect your evidence that it’s all hype, declare yourself right, and walk away, while the people pointed the other way quietly keep building the very thing you just wrote off.
So the question of boom or bust was never really about AI. It was about you, and where you decided to point yourself. Stop asking when it’s going to pay off. Pick a direction worth ten years of your life, cut the noise, and go find out who’s still building when it does.