Is software dead?

00:00:00: Software is dead.

00:00:02: That's a statement I've seen a couple

00:00:04: of times over the last couple of days.

00:00:05: It's a quite heated debate, obviously,

00:00:08: because the underlying question, I think,

00:00:10: of course,

00:00:11: is whether software development is dead.

00:00:14: Or if paid software is dead,

00:00:16: and there are a lot of strong

00:00:18: opinions there, and I have some too,

00:00:20: and some nuances too. So let's dive in.

00:00:22: This, of course, was a post by levelsio,

00:00:24: Pieter Levels. You may know him from X.

00:00:26: He got some strong opinions,

00:00:28: and he's stating that it's now

00:00:30: the era of personal software,

00:00:32: and everyone just builds what they need

00:00:33: whenever they need it. I think. Now,

00:00:36: I definitely disagree with this statement,

00:00:39: and he also kind of clarified it

00:00:41: thereafter. I'll get back to that.

00:00:43: Now,

00:00:44: a couple of answers there, mostly relate

00:00:46: to him being a hypocrite or something.

00:00:49: But then there is interesting

00:00:52: responses like this one.

00:00:53: Don't you think most people are

00:00:54: just lazy and not eager to change?

00:00:56: I get the premise of micro apps, but I

00:00:58: feel like people won't have the patience

00:01:00: to deal with going back and forth with

00:01:01: AI to solve bugs or tune to their liking

00:01:03: if their current apps do

00:01:05: the job without an issue.

00:01:06: And I strongly agree here. I do not think

00:01:10: that software is dead because normal

00:01:12: people will just building their own

00:01:14: Photoshop clones or anything like that.

00:01:17: Totally not the future I see happening.

00:01:20: But one thing I do see happening,

00:01:22: which is what Pieter Levels

00:01:24: then also stated thereafter,

00:01:26: is that they're not doing it in Claude

00:01:28: Code in a terminal like tech people.

00:01:30: But their AI chat app is already doing

00:01:33: it for them without them even realizing.

00:01:36: They're building personal software,

00:01:37: like generating PDF reports, full web app

00:01:39: dashboards, and even entire web apps.

00:01:42: And that's the part I see.

00:01:44: Now, still with a lot of nuance,

00:01:45: though, but I do see that, of course,

00:01:49: normies—and I hate that term; it's so

00:01:51: disrespectful—but normies will ask ChatGPT

00:01:56: to remove the background of an image,

00:01:58: and behind the scenes, ChatGPT may write

00:02:01: some Python code or whatever and do it.

00:02:04: Stuff like that.

00:02:04: The same for a PDF report. Behind the

00:02:06: scenes, ChatGPT writes HTML code and then

00:02:10: writes more code that converts that to

00:02:12: a PDF or prints that as a PDF document.

00:02:15: That's already possible today.

00:02:17: Obviously, now it's worth noting though

00:02:19: that as Andrej Karpathy also mentioned.

00:02:22: The vast majority of people are not

00:02:24: really using AI. Only a very small group

00:02:28: is actually using it, and even there,

00:02:31: not a lot of people are paying.

00:02:34: You may know that chart that only like

00:02:36: two percent of U.S. households or

00:02:38: something like this are paying for AI.

00:02:40: So adoption is super slow,

00:02:43: but that is the direction I see. That

00:02:45: people will ask for problems to be solved,

00:02:48: and those AI assistants will be able to

00:02:50: solve a lot of problems. Now. Again,

00:02:53: and I know people on the internet hate

00:02:54: that, but you have to differentiate.

00:02:57: There will be normal people who

00:02:59: just want to remove the background,

00:03:01: just generate a PDF, and yeah,

00:03:02: AI assistants will do that for them now.

00:03:04: Can you quickly just tell me how much did

00:03:07: these people pay for software before?

00:03:09: Oh, right, it was nothing. You did not pay

00:03:12: for a PDF converter before. Maybe you did.

00:03:16: Most people didn't. They were searching

00:03:18: the web until they found some scammy

00:03:20: website that would do it for them.

00:03:22: They would upload their bank statements

00:03:25: to some weird website where whatever

00:03:28: happens with that data, and they would get

00:03:31: the combined PDF or whatever out of it.

00:03:34: And nowadays they do it with AI,

00:03:36: where whatever may happen with that data,

00:03:38: I guess. So not much changes there.

00:03:40: Now you know who was or is paying

00:03:43: for software, professionals, and

00:03:46: if it's your hobby. Now, games, by the

00:03:48: way, are a different kind of software.

00:03:50: I don't see people vibing their own

00:03:53: games because part of a nice game is

00:03:55: not knowing what's going to happen when

00:03:57: and how to perfectly beat the game.

00:03:59: But that's a different story.

00:04:00: But if we think about other hobbies.

00:04:02: Let's say

00:04:03: you're a hobby photographer,

00:04:04: or you're a hobby video creator.

00:04:08: Maybe you paid for some Adobe products.

00:04:11: You paid for the Photoshop.

00:04:14: You paid for Premiere Pro.

00:04:16: Something like that.

00:04:18: I don't see that those people will

00:04:20: necessarily vibe code their own

00:04:22: alternatives because guess what?

00:04:23: If you do that, you have to maintain it.

00:04:27: You have to fix all those annoying

00:04:29: bugs you'll encounter over time.

00:04:31: That may be fine for certain tasks,

00:04:33: but if it's your hobby,

00:04:34: if you're working a lot with it, if you're

00:04:36: interested in the more advanced features.

00:04:38: If you're maybe also interested in

00:04:40: getting new features you didn't expect

00:04:42: in the future, then you really want

00:04:45: someone else developing that for you.

00:04:47: Now prices may go down though because

00:04:50: now it's easier than ever to launch your

00:04:53: Photoshop alternative, and indeed I'll

00:04:55: get back to that thing and this entire

00:04:57: reverse engineering trend which we have

00:04:59: right now and the downsides of that.

00:05:01: But yeah, prices may go down,

00:05:03: but I definitely see those people not

00:05:05: vibe coding their own alternatives now.

00:05:08: There will be some middle ground here.

00:05:09: It's absolutely possible that

00:05:11: you had certain things you did,

00:05:14: which were very specific,

00:05:16: and now you just ask ChatGPT or Claude to

00:05:19: build you a tool that does that one thing,

00:05:21: which you do over and over again.

00:05:23: On its own, so that it builds you

00:05:25: a tool for one specific task.

00:05:26: I definitely see that this could

00:05:28: happen right now already, but soon I

00:05:32: don't see those hobbyists and so on

00:05:35: switch away from paid software necessarily

00:05:37: now. Long term on the horizon,

00:05:40: nobody knows what happens. Right,

00:05:41: with AI, that is impossible to predict.

00:05:45: There's too much going on, so

00:05:46: it's impossible to say what will be the

00:05:48: case or what will be true in ten years

00:05:51: or even five years. But short term,

00:05:53: yeah, I don't see that now.

00:05:56: Normal people, hobbyists are, of course,

00:05:59: there. They're not the most interesting

00:06:01: market for the reasons I stated.

00:06:03: You know what is interesting?

00:06:05: Companies, businesses,

00:06:07: and here again we have to differentiate.

00:06:09: Large corporations—they will be slow.

00:06:12: There is a lot of inertia here, so they

00:06:15: will be slow in adopting AI solutions.

00:06:20: They will likely not vibe code all their

00:06:22: code or all their software anytime

00:06:24: soon for a broad variety of reasons:

00:06:26: compliance, aligning different teams,

00:06:30: the maintenance burden,

00:06:31: all that fun stuff, and even more.

00:06:34: Small business owners, different thing.

00:06:36: They will definitely also pay for some

00:06:38: software for mission critical parts

00:06:40: where they really don't want that bug to

00:06:42: surface right when they need the software.

00:06:45: But for those annoying side tracks,

00:06:49: I definitely see that. Or for other

00:06:51: things like your website, and I mean,

00:06:53: small business owners before were likely

00:06:56: using something like Wix or Squarespace.

00:06:58: Now they can use ChatGPT with ChatGPT

00:07:01: sites, or they can use Lovable.

00:07:03: So they're just switching.

00:07:05: They weren't primarily interested in

00:07:08: software built by other developers

00:07:09: there before. Well, I guess Wix,

00:07:12: but Lovable is also built by developers.

00:07:14: If you look at it like this, so that is

00:07:16: some switch I see here, and I do also see,

00:07:19: by the way, a nice opportunity here.

00:07:21: If you're a developer, a freelancer,

00:07:23: let's say. Offering to your customers that

00:07:26: you are building software, whatever it is,

00:07:29: with AI, but also with you,

00:07:32: the human in the loop, that can be

00:07:34: a nice business use case right now.

00:07:37: Because I think there will be lots

00:07:39: of business owners and generally

00:07:40: people that are looking to lower

00:07:44: their costs or get a result quicker,

00:07:46: but that don't want to go

00:07:48: all in on AI necessarily.

00:07:49: They may not want to use Lovable.

00:07:51: They may want a developer who uses

00:07:53: Lovable or whatever Claude Code instead.

00:07:56: So I think that, by the way,

00:07:57: is a nice business opportunity

00:07:59: right now and in the near future.

00:08:01: Again, long term, nobody knows.

00:08:04: Now, what about software getting cheaper?

00:08:06: There being more alternatives, of course,

00:08:09: competition is definitely ramping up,

00:08:11: and we have projects like this,

00:08:15: REA repository, reverse engineer anything

00:08:18: which is in the end an MCP you can give

00:08:20: to your AI so that it can easily inspect

00:08:23: binaries and reverse engineer software.

00:08:26: We had this

00:08:28: ArtCraft thing where people rebuilt

00:08:31: the entire Adobe suite. Broken.

00:08:34: I only played around with it, but yeah,

00:08:38: these are things that are happening.

00:08:40: I think

00:08:41: it's easier than ever to build software.

00:08:44: Good software still

00:08:46: tricky. With vibe coding alone may not

00:08:49: be the path to get there, but again,

00:08:52: who knows what better models in the

00:08:54: near future, or definitely in a

00:08:56: couple of years, will be able to do.

00:08:57: But right now, not necessarily there yet.

00:09:00: And if you're thinking about using a piece

00:09:03: of software, you may not want to go for

00:09:05: the reverse-engineered, vibe-coded ones

00:09:07: because nobody may be maintaining them.

00:09:10: You don't know exactly if you can trust

00:09:13: these software products now. Of course,

00:09:16: part of the truth also is there's a lot

00:09:18: of paid shitty software out there too.

00:09:22: I'm not looking at Adobe here,

00:09:23: of course, but there is software out

00:09:25: there with shitty business practices or

00:09:29: where the software itself sucks, and

00:09:31: I think it's a good thing that software

00:09:34: is easier to create, especially by

00:09:37: people who know what they're doing.

00:09:38: Especially if you're not just vibe coding,

00:09:40: but if you're leveraging your

00:09:42: knowledge as a developer,

00:09:43: and that can ultimately be a nice way of

00:09:46: entering a niche or a market that

00:09:49: was way harder to enter before,

00:09:51: because you can probably build your

00:09:53: Premiere Pro clone in a not shitty version

00:09:57: and maybe gain some customers there.

00:09:59: So I think right now, in short term,

00:10:02: this is a very good thing.

00:10:04: Now, long term on the horizon,

00:10:06: as AI does everything, if that is where

00:10:09: we're going, and nobody knows, of course,

00:10:12: what's the remaining moat, which software

00:10:15: will survive in the sense of people

00:10:18: will still pay for it, and I guess

00:10:19: it's not just software, by the way.

00:10:21: You can ask this question general:

00:10:23: which businesses will survive in

00:10:25: a future where AI does a lot?

00:10:27: And one or a couple of things you

00:10:29: hear frequently there are data.

00:10:31: If your business, your software,

00:10:32: whatever it is, depends on data,

00:10:35: only you have. Community.

00:10:37: If you can get people together, content.

00:10:41: If you are creating content that is

00:10:43: original and hard to copy or hard to

00:10:46: recreate by AI in the way you do it.

00:10:49: That all is something that could work,

00:10:52: and there probably are more moats.

00:10:54: But ultimately, I think we just have to

00:10:57: accept and, as much as it's possible,

00:11:00: embrace the fact that we don't know

00:11:03: where we're heading. And this is hard.

00:11:05: I totally get it. And by the way,

00:11:07: we're talking about my business here too.

00:11:08: We're talking about our jobs.

00:11:10: We're talking about businesses.

00:11:11: My business. This is hard. But the

00:11:13: reality is that nobody knows what will be

00:11:17: the case in two or three years and so on.

00:11:19: We can only plan for right now. Right now,

00:11:21: there are many opportunities here,

00:11:23: and we don't know if the data

00:11:25: part will matter in the future.

00:11:27: We don't know if the community part will

00:11:28: matter, or if we all get more lonely

00:11:30: because of AI. Because AI does it all,

00:11:33: and we rely on it so much that we

00:11:34: don't wanna get into any community.

00:11:37: We don't know if we get so used to

00:11:38: AI-generated content that that's the

00:11:41: only content we want to consume.

00:11:43: We don't know it.

00:11:44: So the best thing we can do is try to

00:11:46: see opportunities we have today and

00:11:48: plan for today and the near future.

00:11:50: As unsatisfying as that is,

00:11:52: but to get back to that initial statement,

00:11:55: software is absolutely not dead.

00:11:57: We'll have more code and more software

00:11:59: than ever, in one form or the other.

00:12:02: Software development also not dead right

00:12:04: now. Also not dead in the near future.

00:12:07: Long term.

00:12:08: Nobody knows,

00:12:09: but being able to adapt and use these

00:12:12: tools is the best thing we can do.

00:12:14: I guess.