As frontier models keep getting better, more and more people are building, or "vibe coding," their own apps and posting them on social media. That has started one of the loudest debates in tech right now. One side says "software engineering is easy now." The other says "anyone can start a million-dollar company with AI agents as their employees."
There's some truth in both. But I think they miss what actually makes a product great. Let's go through them one by one.
"Software engineering is easy, and AI will replace it"
To judge this, we first have to ask why software gets written at all.
Software exists to automate something, so you don't have to keep doing it by hand. You write the logic, test it, deploy it, and put it out into the world, hoping it works the way you planned.
The easiest way to see this is with an example. Say you run a restaurant. You do the most basic thing a restaurant does: you serve food to customers. Now you want to grow, so you start delivering food to people's homes.
People can place orders on their phones, but they can't pay on their phones. If they have to come to the restaurant to pay, that kills the whole point of home delivery. You could let customers hand cash to the delivery person, which works if you have people you trust, but it isn't very safe. The last option is going door to door collecting payments yourself.
We live in the age of software, though. You can build a system where customers pay straight into your bank account when they order. Now you're not handling every payment by hand, you're not worrying about cash, and the whole thing scales.
AI can write that payment system for you in minutes. That part really is easier than ever. But writing the code was never the hard part. The hard parts were:
- noticing that payment, not cooking or delivery, was the thing stopping your business from growing,
- deciding whether to build it yourself or use someone else's system,
- and keeping it running when real customers do unexpected things: payments fail, people ask for refunds, someone tries to cheat.
AI builds what you ask for. It doesn't tell you what to ask for.
Writing code is getting cheaper. Knowing what to build, and why, is not.
"Anyone can build a million-dollar company with AI agents"
This one is half true. The cost of building has collapsed. One person with good tools can now ship what used to take a small team months.
But when everyone can build, building stops being the advantage. Thousands of new apps launch every day, and most of them never get a single real user. They work perfectly, and nobody needs them.
What separates the products that win from the rest has nothing to do with code.
People don't know what they want until they see it
There's a line often credited to Henry Ford: "If I had asked people what they wanted, they would have said faster horses."
People were happy with horses. They didn't sit around wishing for cars, because they couldn't picture one. If you had built exactly what they asked for, you'd have built a faster horse.
Go back to the restaurant. None of your customers asked for online payments. They asked, "Can you deliver?" Online payment was what made delivery actually work, but you only found that out by trying delivery, watching where it broke, and fixing it.
That is the real work of building a product. People describe their problem in their own words, and it's your job to figure out what they actually need. You only get there by building something, putting it in front of people, watching what they do, and changing it.
AI makes each step of that loop faster. But it can't stand in your restaurant and notice why customers stopped ordering.
So what's actually changing?
Building software is getting easier, and that's a good thing. More people can turn their ideas into something real.
But the advantage has moved. It used to be "Can you build it?" Now it's "Do you know what to build?" The people who make great products will be the ones who stay close to the people they're building for, test their ideas quickly, and keep iterating until the thing works, even when nobody could have told them what "the thing" was.
Software is becoming easier. Great products are still hard. They always have been.
