Thomas Ptacek is a security researcher best known for the Matasano consultancy, and this post leads with his own conflict of interest: he has just left Fly.io to work with Kurt on something new, and says so up front — “so you all know up front I’m talking my book.” What follows is an argument about AI’s second-order effects on computing rather than a product pitch; he says the product detail is being withheld on purpose.

His first-order claims are the ones he says he has already lived through. AI dissolves the boundaries between backend and frontend, between web and native, between systems and applications — each of which would be a career-defining shift on its own.

The second-order claim is that AI is also knocking down the boundary between programmers and users. His own menu bar is full of programs he made for himself, with English as the programming language, and his inference is short: “What I did on my computer, any power user can do on theirs.”

  • If that holds, the interesting questions change. What does computing look like when many applications have an audience of one or two people? What happens to distribution, and what do those programs run on?
  • Software stops arriving as finished, fixed-function products. Browsers and word processors survive, but he expects “thousands of times more applications that pull in 1/7th of the guts of a word processor” to solve somebody’s idiosyncratic work or home problem. Strangers ship building blocks; users ship the app.
  • His examples are deliberately mundane: an accurate forecast for Roscoe Village, where the nearest weather stations miss the lake effect; which of three routes home to take; whether a meeting is happening right now that he is supposed to be in. “These all sound trivial. That’s the point.”
  • The operating system argument: the core job of a modern OS is not hardware access but partitioning applications off from one another and controlling how they communicate. That makes sense “in a world where we’re importing all our software from strangers” — less sense when most software on the device has the same provenance, and almost none in a world where every application is malleable.
  • The product is a phone. Every mainstream 2026 phone was specified in 2023, he argues, for deterministic fixed-function software written by professionals; the result is beautifully machined hardware doing “the same job as a DEC PDP-11: running prefab applications.” Theirs is meant to build apps for you, on the device, on request.

He is aware of how it reads, and says so: he knows he will sound “like yet another startup nerd talking about ’the future of programming’ before announcing that I’ve joined, like, Sri Lanka’s answer to Uber or something.” His hedge is also the reason the post is worth reading: the useful content is the prediction about who writes software, not the hardware.

The 305-comment thread on Hacker News

What the thread adds

  • tptacek (the author, in the thread) — joins the pile-on against his own post’s genre: this kind of piece — “I’m leaving this company I’ve been very publicly associated with, and here’s the new thing I’m doing — is deeply cursed. There’s no way to say anything interesting without it just stinking like an ad for the new thing.” He notes he did not submit or upvote it, kept it abstract on purpose, and will write “something more fulfilling about this next week.”
  • chroma_zone — the security objection, and the sharpest one: “If I’m running software written by an LLM, even if I was the one who prompted the LLM, I would still want my OS to treat it as if written by a stranger.” The replies (cgio on giving devices their own identity, fennecbutt on sandboxing and users who grant every permission without reading) agree that the post never says what replaces the isolation model.
  • Xirdus (20 replies) — proposes skipping the generated app entirely: “why would you ever want to ask an AI to make an app for you to complete some task, if you can instead ask the AI to complete the task directly? The future isn’t a bunch of extremely personalized apps - the future is a single AI assistant doing all the work.” xg15 answers with the case for keeping local fixed software: your agent is somebody else’s computer, inference costs time and tokens, specialized graphical interfaces beat a chat box for many jobs, and a weekly repeated task should not re-roll its own results every week.
  • utopiah — the definitional objection: “most articles challenging OSes have no idea what an OS is… If your OS doesn’t change how a computer allocates resources, then you are talking about higher level.” Their note: chroot, jails, containers and VMs have been doing the partitioning job for decades.
  • II2II — the historical base rate, from BASIC, HyperCard and the early web. Those attracted droves of excited people, and most people “didn’t even explore the potential because they simply don’t think of computers as programmable machines.” wisemanwillhear tried to get non-technical family and friends to build their own tools and found “they just tell me that they can’t think of anything the AI can help with” — while rukuu001 reports the opposite: a non-technical partner now keeps half a dozen Claude-built tools and digs back through old chats to reuse them.
  • FiberBundle — the demand-side counterexample: “People could already vibe-code simple products today, but there’s hardly any non-programmers that do,” even for businesses, and the SaaS businesses that were supposed to be wiped out are largely still there.
  • linkregister — a practitioner who cannot relate at all. Agent-wrapped models made the authoring part of their job fast and left “the rest… virtually the same speed,” and the mobile examples read to them like features their phone’s voice commands have handled for a decade. Their conclusion: “I want good software that someone that I trust wrote.”
  • c-hendricks — tries to deflate two of the examples with existing products: “Waze and Calendars. Both have huge audiences.” Ptacek replies that this misses his actual complaint — Waze returns the fastest route, and he only ever wants one of the four routes that are reasonable.

The question the thread kept asking

For a post whose argument depends on applications the author built for himself, nobody could find out what they are. tim333 reads the thesis back as a consumer question (“So the product is a phone without the bits that allow you to run apps on the basis that AI will implement the stuff the apps would have done?”); hn_submit: “He talks about applications he generated without specifying what they are or do”; 32827, linking a news item about smartphones fading, asks flatly: “So, what are you really working on?” The author’s stated answer is that the vagueness was deliberate, and that the technical detail is coming in a later post.

On reading comments as evidence: HN handles are pseudonymous, the site publishes no per-comment scores, and the ordering here is HN’s own ranking rather than a vote. This is a slice of the thread, not a consensus, and the claims above are attributed to the people who made them.