I don't think software is dead
Pieter Levels posted yesterday that software is dead. Everyone builds what they need when they need it, he said, or their AI builds it for them. A lot of people got annoyed. I read the thread because I wrote about this last month, and I wanted to know if I'd been too careful.
I don't believe software is dead. I do believe the boring kind stops getting paid for.
The short version:
- The chat will make a simple tool or file when you ask, and throw it away when you are done.
- Boring apps, like a basic notes app or a simple tracker, stop getting paid for.
- People will ask the chat to do the small job instead of buying another app.
- What still gets paid for has to keep working after the chat is closed, costs money when it is wrong, has a real person who answers, or follows a law.
The day before, a post about AI taking apps apart and rebuilding them got about a million views. That same weekend, someone turned a game they already owned back into the project files in half an hour. Levels quoted the million-view post and made it bigger. The headline did the work. The smaller point is the one worth arguing about.
What I agree with
They ask for the job, not the software
People are not going to say "build me software." They'll say "edit this photo," or "are sales up this quarter across our stores?" The chat edits the photo. If they want to do it themselves, it gives them something like Photoshop. If they want the number to stay current, it connects to the stores and keeps checking. They won't think they built anything. They'll think they got the job done.
They pay once
And they pay once. $20 a month to Claude or ChatGPT, instead of $20 to Adobe, $20 to Canva, $20 to Microsoft 365, $20 to Grammarly, and $20 to Squarespace. Levels says he's already cancelled almost every SaaS bill he had. What he still pays for is the AI, a server, storage, backups, YouTube, Spotify, power, and taxes.
He's arguing against his own income
He sells software, and he said this hurts him, because people will cancel his apps too. He also said he doesn't fully know where he fits. I believe that. It's a strange thing to argue against your own income in public.
The chat already sold the result
Last month I wrote that if someone can copy most of your product with Claude, the code isn't the business. Yesterday was that same point from the buyer's side. I was still picturing a company that sells the result. For a spreadsheet, a sales chart, or a customer tracker, there might not be a company. The chat already sold the result inside the $20 plan. The dashboard I would've tried to sell is an afternoon in Claude.
What's actually left
This morning he wrote down what he thinks is left. Nikita Bier had said a clever technical trick isn't enough to protect a business anymore. Levels's list: data nobody else has, a product that gets stronger because other people use it, a way to reach people, hardware, money, and rules and licenses. That's close to the list I already had. I should've included hardware and rules.
People still pay for creative work
Benja made the split I keep coming back to. Nobody cares who made their PDF reader. They just want the file to open. A book is different. You could have one written just for you and still want a specific person's book. Creative work is what people still pay a person for. The boring tools are what the chat takes.
What I don't agree with
Most people aren't building software
The reply that traveled about as far as his post was from Taras. He said Levels's normal person already hangs around tech people, a regular person has no idea what this argument is about, and teenagers aren't as good with technology as people assume. I agree with Taras. Johnny said the same thing in steps. Most people still don't use AI. Most of those don't pay. Most of those who pay still use it like search.
His examples are close to people who build
Levels's examples were a personal trainer, a restaurant that stopped paying Squarespace, and a 76-year-old dad who likes ChatGPT. I don't doubt those people. They're also one conversation away from someone who builds software for a living. The typical person I know still uses the chat like search that can write.
A one-time file is not a cancelled app
He called the new software temporary, and that word is doing too much work. A PDF the chat made is not a cancelled app. A dashboard that has to keep running is a different thing, because the other tools change and the number can't be wrong. He says the AI looks after its own projects and sends bugs back in to get fixed. I've watched that work until the bug is boring, or the situation gets messy. Then a person has to care. That care is still a job. Sometimes it's still something you can charge for.
That's him, not most people
I haven't cancelled my apps. He has. That's a difference between the two of us. It isn't how most people spend their money yet.
More people building is not more people getting paid
Matt said he sees about a hundred times as many people building apps as a couple of years ago, and he doesn't see more of them making money. I agree with him, and I think Levels treated that as a side point. Taking a game apart doesn't prove people will stop paying for software. It proves more people can make software. Those are not the same thing. If your plan is to be the tool the AI calls, Shopify or Stripe or some smaller pipe, the AI will look for the cheapest option or do the job itself. Your customer is trying not to pay you. That can be a business. It's a worse one than it sounds in a pitch.
So what am I looking for
Same question as September. I'm not looking for something AI can build. AI can build almost anything. I'm looking for a job someone will still pay for after a $20 chat can do the boring version.
If the chat can do it, it was never the gap. Before I build, I look for four things I can actually check:
- A mistake costs money: Payroll, an order that spends money, a form the government reads. A guess is cheap. Being right is the work.
- Someone has to answer: The chat can make the file. It doesn't pick up the phone, and it doesn't take the blame.
- A rule says who is responsible: A law, a license, or a contract. The chat can't skip that.
- I know the customer: I can say how they find it. More people are building, and that has not made more of them get paid.
A few things matter later, once people are actually using it. I don't go hunting for them on day one, because I don't have them yet:
- Data that only exists because people already use the product.
- Other people using it, so it gets harder to leave.
- A machine, a store, or a shipment the software is tied to. The software can be copied. The machine can't.
Levels's warning is that "we're not there yet" is a slow plan when the models jump. I think "not yet" is still fair for customers. It's a bad reason to spend a year on the app they'll drop the month they show up.
I don't know what I'd build instead. He said he doesn't either. I'm still looking, and I could be wrong about the timing. I'm more sure about the order. The boring software goes first.