How Vibe Coding Is Changing the Software Industry: A 2025 to 2026 Reflection

Someone once told me Naval Ravikant said that one in 400 people knows how to code, but all 400 of them know of a problem they wish they could solve. I have no idea if he actually said it. Doesn't matter. The point stuck with me. If the fix for your problem requires code, and you can't code, you're stuck. That leaves 399 people sitting on problems they can't touch.

Vibe coding broke that wall down.

I remember when it showed up last year. People hated it. "That's not real coding." "It can't do this, it can't do that." By summer everyone was doing it anyway, even though the tools were rough. Replit, Lovable, Cursor, Bolt (to name a few), all of them had problems. Then somewhere around October to December, things actually clicked. That was the first time I, a complete novice, built something real. By the first quarter of this year it had gone mainstream. Anthropic could do it, OpenAI could do it, everything could do it. You just plugged in your idea and watched it happen.

So now that the first half of 2026 is almost done, I keep coming back to that 400 number. How many of them can code their own way out of a problem today? Probably not a huge jump. Maybe 30, maybe 50. But going from 1 to 30 is a 3,000% improvement. Sit with that for a second.

Here's the part I actually wanted to write about, though. It isn't the headcount. It's how the whole software world has flipped.

We used to pay somewhere between $14,000 and $18,000 a year for software subscriptions, often just because we needed one feature out of the whole thing. So we built our own. It was painful. Took us two months, and I won't pretend it was fun. But we wiped out that $14,000 a year completely. Now we pay under $80 a month for a server that handles 60 or 70 users. Our company is 11 people, Financial Services company. We will never hit that ceiling, and there's no extra cost if we get close. You could not do that the old way.

The best part came after. What took us two brutal months to build the first time, we rebuilt as version two in under three days. Under three days. And version two wasn't just the same thing again. It cleaned up all the spaghetti code, came with a proper PRD, security, risk profiling, authentication, backups, the works. Faster and better at the same time, which never used to happen.

If a no-name company like ours can save $14,000 a year at the very least, think about what that does across the entire SaaS industry. The effect is going to be enormous.

That's what gets me looking forward. Summer of 2025 was a fight. Summer of 2026 is a different world. So what does summer of 2027 look like? My guess is that the number of people who can take a problem, code it, and actually deploy it lands somewhere between 100 and 200 out of 400. Every quarter these tools get sharper. It's relentless.

I'm not blind to the risks. The data question worries me, and I think there's a real chance of an AI bubble. But I'm still optimistic. Local models, the tooling around them, the UI and UX that lets you run them on your own laptop, all of that is improving fast, and I expect it to keep accelerating over the next few months.

Good times. Let's see how the rest of the year plays out.