Software is dead? Pieter Levels' comment section disagrees first

Pieter Levels (@levelsio) posted a beach-walk video on X, opening with: “What's left to do when AI can build everything?” He answered himself: “Many things actually!” In the video he wears a black T-shirt and a black cap, standing in front of beach vegetation, with the sea and several low buildings in the distance. The sea breeze is strong. This image, placed next to “software is dead,” feels a bit out of place.

The mismatch comes from a post he made last month. At the time he wrote: Software is dead. This is now the era of personal software: everyone builds what they need, when they need it. The original post link is here: https://x.com/levelsio/status/2109262337500705094. Someone in the comments asked directly: then why don't you shut down your own SaaS? Pieter replied: enterprises are already paying billions of dollars in AI token costs for this. Another reply was more specific: a commenter said AI wrote almost all the software his company sells, the company's monthly revenue is over $500,000, and he is no longer sure what his job is.

But the comment section was not convinced. Phil No Chill said enterprises still need to buy software, or hire people to create and maintain internal software, and in a few years enterprises will use 1,000 times more software than today. Peter Yang raised enterprise-level questions: security, compliance, and data privacy—can they be solved with vibe code? He doesn't know. Pedro Coelho Silva split the timeline into two layers: software development looks fast at the top, but adoption is extremely slow further down; governments and traditional companies need 5 to 10 years; the physical world is even slower, and it may take another 10 to 20 years for AI to penetrate it. Michal Langmajer offered another categorization: indie developers, geeks, and startups may change first, while large companies and the general public may not.

These replies pull “software is dead” back down to earth. What AI lowers is the cost from idea to a runnable prototype. What enterprises buy is a whole bundle of responsibility: who maintains it when something goes wrong, how permissions are managed, how audits are passed, and where data is stored. Personal software will increase, but general-purpose SaaS will not go to zero because of it. A more likely stratification is that lightweight, one-off, on-demand generated small tools will become more common, while enterprise software that is compliance-heavy, integration-heavy, and responsibility-heavy will continue to exist. Saying “everyone can build” is easy; saying “everyone can maintain it, bear the consequences, and make colleagues comfortable using it” is hard.

In the video Pieter also mentioned audience and content. CodeGlitch said in a reply that he pointed out in this video that audience and content are advantages, and @marclou does not accept that advantage. This observation is more interesting than whether software is dead. Once building becomes cheaper, distribution becomes more expensive. Someone said after watching the video that they felt inspired and wanted to start vlogging too. One person stands by the sea talking, while the comment section discusses enterprise procurement and compliance—the scene itself says a lot.

Pieter posted a YouTube link in a reply: https://youtube.com/watch?v=B6FxojoTHCc. He said he started vlogging because he was influenced by @PeterSantenello, @girdley, @theo, @marclou, @robj3d3, @frederickjames, and others. Someone in the comments asked what equipment he uses: DJI Osmo Pocket 4, 4P, or Action 6; someone said long walking-and-talking videos are good; someone joked that they had lost a wedding ring on the beach and asked him to help find it. The software topic eventually came back down to these details.

So what is left after AI can build everything? The comment section has already listed a set: enterprise procurement, security compliance, data privacy, slow adoption, the physical world, audience, content, distribution, and trust. Code is only one item. Pieter said there are still many things, and that is not false. As for “software is dead,” for now it looks more like a judgment within a timeline than an epitaph.