Pieter Levels Says "Software Is Dead." Tech Replies

Pieter Levels said software is dead. Two days later he put a yearly price on a dataset and aimed it at AI agents.

On 11 October 2026 the post was two lines. “Software is dead.” Then: “It’s now the era of personal software and everyone just builds what they need whenever they need it I think.”

The sentence people forwarded was a different one. They called the post “SaaS is dead.” That wording is theirs. In the days after, he sized the claim himself: many or most SaaS products get replaced, because people pay an AI to build what they need, or to answer the question so the software never has to show up.

The two lines had over 1.5 million impressions at the time of writing. The argument was older than the post. The post was just short enough to forward.

He had already cancelled the stack

The months before are the receipt.

On 26 July he published the version with verbs: cancel the subscriptions, vibe-code the jobs. On 4 September he wrote that he would keep paying ScrapingBee $99 a month. On 9 September he wrote that a scraper of his own had replaced a $249 a month plan for about $1. Both figures are his, five days apart, and this page leaves them as two posts. The same day’s cancel list put the wider savings near $25,000 a month. The line-by-line ledger lives on its own page. What matters for October is the shape: a founder with a real stack, crossing vendors off it in public.

On 23 September he said he had passed $10 million a year. The figure is revenue plus investment gains, and he told people to take the gains with a grain of salt because most of them were unrealized. The margin he attached, about 94.5%, is the cost cut turned into a percentage. Also his. The shops, in that same telling, were still around $200,000 to $250,000 a month.

He had already argued, on 30 July, that execution was a dead moat and that a build stack could run near $20 a month. That essay is about indie hacking. The October post is about software as a thing you rent. The dossier keeps those years from becoming one slogan. Shipping fast is the method underneath both.

Two lines, and a shorter draft twenty minutes earlier

The viral post is a quote-post. Under it sits @iamgingertrash, from earlier the same day: “It is the beginnings of the end of software.” The rest of that post, in paraphrase, says the latest models got good enough at reverse-engineering and improving almost anything, and that the step-change flew under the radar.

About twenty minutes before his own quote-post, Levels had already answered the same person. “Yep, software is dead.” Then: “Personalized software remains.” The long version is the one that traveled. The short version is the cleaner statement of what he says he meant.

A mirror sits on his blog as the era of personal software.

Agreement was a supply story

Building gets cheap. Ordinary software gets cheap with it.

signüll, an investor and writer, set the adoption bar low. Paraphrasing that post: you do not need everyone to build software. If 1% can, supply floods and costs collapse. The personal version was one line. “i never pay for software because i just ask claude.”

Tony Dinh, who makes TypingMind, wrote “The case for personal.” A SaaS product, in that post, ships about 50 features. “I need about 6.” “Vibe coding my own tool will probably take 1 day.”

Ronan Berder took the same mechanism to zero. “there’s no moat, anybody can whip it out.” If any random developer on the internet can rebuild it, “prices go to 0.”

Read together, they are describing a product whose only defense was that it was annoying to make.

You cannot vibecode trust

The pushback was more specific than a vibe.

Raymmar, a community architect at Replit, posted the mockery in three beats: “tells everyone software is dead,” “keeps building software,” and “Please stop taking these people seriously.”

Arvid Kahl, founder of Podscan, answered the slogan directly. “software is very much not dead.” He pointed at enterprise customers who still want multi-year contracts and integrations. The next day he compressed the objection to three words: “Can’t vibecode trust.” The examples he reached for were SSO, SAML, and SCIM. His point, paraphrased, is that the second S in SaaS is the service.

Charles Burdett aimed at the crowd using the paraphrase. “the ‘SaaS is dead’ crowd is actually ‘my indie hacker SaaS for other indie hackers is dead’ crowd.” And: “SaaS is in a boom right now.” On 15 October Levels answered him. “None of my SaaS are for indie hackers though.” He named Photo AI, Interior AI, Remote OK, and Nomads, and said none of them are for indie hackers.

Steve Jones, CTO at Equalize Digital, wrote the ordinary-person objection in one line. “you’re grossly overestimating the average person’s technical knowledge.”

Brett MacDonald, a publisher, reached for a kitchen. “Software is no more dead than restaurants or fast food is because we all have well equipped kitchens.”

Garry Tan, who runs Y Combinator, did not write a line of his own on this. He reposted Jayme Hoffman. The words in that repost are Hoffman’s: “It’s like food: many people can cook, but not at a Michelin level.” Tan’s post is the repost.

A few people refused the binary

benja, who builds at Tembo, replied within minutes. “the average person doesn’t want to build their own software.” The shift he actually saw was the buyer. “who’s buying software is shifting more than software itself.”

David Cramer, who founded Sentry, posted on 15 October that “a ton of valuable software became more valuable to me since agents.” The kind he meant was “not software that someone vibed into existence.” Earlier that week he had written that “software eng isnt dead. we are building more software than ever.”

swyx was not answering Levels. He said “A lot of people talk about the death of SaaS but mostly in theoretical terms,” then pointed at a hackathon called “How to Kill Your SaaS,” where a team cloned a product priced at $43,000 a year. The nuance is the experiment. A funeral is a sentence. A replacement is a weekend with an invoice attached.

Marc Lou had posted on 10 October, the day before, so this is context from the same conversation the timeline was already having. “I’m not going to vibe code my $19/mo SaaS alternative.” The other half: “But $4000/year just to send a few emails is insane.” A cheap tool stays. An expensive tool that does a small job gets a second look. His portfolio is full of that kind of sorting.

Kisson separated the corpse from the cost. “software didn’t die, the cost of throwaway software dropped to zero.” The apps that die, in that post, are “the generic middle ones nobody loved.”

Jeremiah Weise, a product leader, dropped the death word. “the business model will change.” “We are going to have more software than we’ve ever, ever imagined.”

Then he said what he meant

The clarifications are more careful than the headline, and they arrived fast.

Within the hour he wrote “I’d call this a bubble in a bubble.” “Tech people somehow still think normies are not spinning up personal software already.” He said “their AI chat app does it without them even realizing”. “Everyone will mostly use personal ephemeral software inside AI chat apps I think.” The advice in that post, paraphrased: become a tool agents use, and charge the agents.

A little later the example got specific. “Normies won’t ask ‘build me software’.” Paraphrasing the rest of that post: they ask to edit a photo, and it “will spin up a Photoshop clone.” They ask about quarterly sales, and it builds an analytics dashboard that runs on a schedule. “They won’t know they’re building software.” “And then they’ll stop using most regular software!”

He also posted what his own card still clears. For the business: AI inference, Hetzner, Cloudflare storage, Backblaze backups, taxes. For himself: Claude Code, YouTube, Spotify, Starlink, an eSIM, cloud storage, utilities, taxes. The summary he wrote: “all my money now goes to: Content, Data centers, Power plants + water, Governments.” He expected “a tidal wave of cancellations once people find out how much money they can save.”

He quoted Chris Dixon’s 2013 line, “What the smartest people do on the weekend is what everyone else will do during the week in ten years”, and said the lag is more like one or two years now.

He offered Omarchy as a picture of the personal-software end. “personal software that removes reliance on paid SaaS subscriptions”, and “all free, open source and sponsor funded.”

On 12 October he tied the slogan to SaaS and gave it a size. Agents and chat apps, he wrote, will replace most knowledge workers and many or most SaaS, and the cost of everything will fall. Then a rough transition. In that post he also said a large part of his net worth is Google stock.

About half an hour later he rejected the doom reading. “People feel threatened and they think my point is doom and I’m a doomer.” “I’m not, there’s an explosion of creativity.” “It’ll just be a completely different paradigm.”

The moat list arrived as a quote of Nikita Bier, whose post had about 3.2 million impressions at the time of writing. Bier: “Technical moats are mostly over at this point.” What he left standing was proprietary data, network effects, and a physical component. Levels widened it to six: proprietary data and content, network effects and community, audience and distribution, a physical component, capital, and regulation.

Later that day he was blunter about the incentive. Half his net worth, he said, sits in AI stock, and he named Google, SpaceX, and Nvidia. He called it a bit dishonest when people at AI companies reply that this is the best time ever to build a startup. What they mean, in his wording, is “best time to pay my company money for tokens.” His own conclusion: “I don’t think it’s the best time to be in software ever.” Creativity, in that post, does not necessarily become money. He put the scale at about 40 million software developers and about 2 billion people regularly using AI tools. Those counts are his.

On 13 October, on the work of helping companies automate with AI, he called that work temporary, maybe a few years, “we’re vibe coding our own replacements already now.”

Then he sold the API

On 13 October he announced a paid API for @nomadscom, built for AI agents, with MCP in the pitch. The corrected prices were $2,490 to $9,990 a year. An earlier post the same hour had listed $249 to $999 a month. He reposted the yearly figures as the correction.

Read that against the list from the day before. Proprietary data was the first moat he said still holds. Charging agents was the business he had told other people to start. The API sells data he already has, to software that would rather call it than rebuild it.

Postiz had already spent 2026 turning a social scheduler into something an agent can call. Different product. Same new buyer. A customer that arrives with an API key is already a way this scene gets paid.

What is left on the invoice

Eight months earlier, the public market had a larger version of the same scare. Around 30 January 2026 Anthropic launched Claude Cowork plugins for work like legal, sales, marketing, finance, and data. Software shares sold off in the first days of February. Coverage nicknamed it the SaaSpocalypse. Reuters reported that software and services had lost about $830 billion across six sessions. Bloomberg’s explainer is the write-up under that nickname. The loss figure is from that coverage. This page is walking the October posts.

If the supply story is right, the invoice for “we built the software” gets thin. The invoices people actually named, in this week of posts, are older than vibe coding.

Levels’ six: content and data, a network, distribution, a physical thing, capital, a rulebook. Arvid’s addition is trust, and the service around things like SSO. Kisson’s addition is taste. The generic middle is what gets cloned on a Saturday. Cramer’s version almost reverses the slogan: agents made some existing software more valuable, the kind nobody had vibed into existence. Jeremiah’s version drops the death word and keeps the pile. More software. A different way to get paid for it.

Levels’ own card is a hint at which bills he thinks survive a cancellation wave. Content. Data centers. Power and water. Governments. And, as of 13 October, a data API with a four-figure yearly price, sold to agents.

Where do you think software goes from here, and what is left of SaaS that a person will still pay for on purpose?

A note on the numbers

Impressions in this piece are a snapshot from around 15 October 2026. They move. The $25,000 a month, the $99 and $249 scraper lines, the 94.5% margin, the $10 million mix, the $20 build stack, the Nomads API prices, Bier’s impressions, and the counts of developers and AI users are the posters’ own figures. Levels has said the investment gains were mostly unrealized. Treat them the way the September post asked to be treated: a public ledger, useful as a shape, worth a grain of salt where he said so.

FAQ