Everyone’s software

We’re far from that.

Even in the richest countries, most adults can do little more with a computer than send an email or browse the web, and many can’t do that. The internet, built with no center, was supposed to belong to them: “Without asking anybody else, I could get a computer, put some software on it, plug it into the Internet, and I’d have a blog and a voice,” Tim Berners-Lee has said of the early web. People did run their own sites, and years later crypto promised everyone their own keys and their own nodes. Each time, centralized platforms won for the most understandable reason there is: they were simpler. “People don’t want to run their own servers, and never will,” Moxie Marlinspike wrote, and because we don’t, our documents and a decade of family photos live with companies whose business model we never check, on their terms rather than ours. That is the first gap: we use our software, but we don’t own it. That may be about to change, because if talking is the only interface you need, running your own computer stops being a skill, and the decentralized internet we were promised can finally be the simple option.

The gap is closing.

When talking is enough, a gap starts to close, the one between every product and what one person actually needs. We put up with that distance because making the exact thing always cost too much. That is the second gap: software is made for everyone, so it fits no one exactly. As that cost falls, so does the distance we’ll tolerate, toward zero, because when software fits us exactly we won’t settle for less, and every company that lives in that distance stops being necessary. Until now only developers could close it, by making software for themselves, and so far AI has mostly made us developers faster. Next comes the moment when needing software is enough to have it: software that asks the right questions before it builds,¹ listens to the answers and keeps adjusting as it’s used, so nobody has to be its support. When it is that easy, we’ll make software outside of work the way we plant a garden, to shape our small part of the world, and the apps we use every day will say who we are, the way our clothes do.

More and more of that software won’t have a screen, because people are starting to ask agents instead of opening apps. An agent, though, knows only what it has been given and can do only what its tools allow. Mine can see my room, knows what’s in my pantry and follows where my money goes, but only because I connected each of those myself. The abilities other people will want for their agents are just as personal, and rarely ones a company would ship: a way to see their own room, to reach their own private records, to use TikTok in a way nobody has imagined yet. It’s tempting to think we’ve already invented everything and AI only changes how we ask for it, but people’s imagination is wilder than that, and much of what they’ll make hasn’t been thought of yet. Each of those abilities still needs tools written by technical people. That is the third gap: even when you know exactly what you want, the distance between the idea and the thing is an engineer’s time.

Cut the tree.

Much of that cost is choosing. Before writing anything that matters, an engineer walks a tree of technologies and patterns: how people log in, where the data lives, whether work happens now or in the background, whether it ships native or on the web, who on the internet can reach it. Experienced engineers carry the tree in their heads, because most of it was settled long before the problem showed up and one good answer usually covers almost every branch (looking at you, Postgres). That is why the best platforms can cut branches for you, and each cut shortens the loop from idea to something real. Vercel keeps cutting them: Next.js settled how a web app is shaped and eve is doing the same for agents, the AI SDK gave every model one interface and AI Gateway put every provider behind one key, among others. Those cuts are how Vercel made the cloud simple for developers, and v0 is carrying that simplicity to everyone. I want to go one step further, to the computer itself, so that anyone can have one that builds for them.

A computer you talk to.

Nobody should have to administer that box, yet today an assistant fixes a computer with commands that work and are then forgotten by everyone, the machine included, so the next fix lands on top of something nobody can explain. A better way has sat in a corner of the Linux world for twenty years. On NixOS the whole system is described in files: change the files and, once the change passes its checks, the machine becomes what they say; change them back and it returns. That idea is leaving its corner: in September the Dutch government chose NixOS for its sovereign government desktop, to depend on no single vendor and because a fleet described in files is easier to manage and to audit.

Files happen to be what language models handle best: text is what they read and write natively, a diff is a change they can explain in a sentence, and a history of changes is a memory they can come back to next week. Files are also the easiest thing in the world to keep: back them up and you’ve backed up the computer; copy them to a new box and you have your old one back.

Because the model reads and writes the files, a person only has to speak. You say “my mom needs a page with the store’s hours” or “make the photos app work on my phone too,” and the system writes it down, builds it and shows you the result, without you ever seeing the files. The tree is already cut, down to how an app reaches a model and gets to a phone, a laptop or the web, so you never need to know whether you got a native app or a web app. My mom doesn’t know the difference, and she shouldn’t have to. All that’s left to talk about is what you want.

Wherever it runs.

Where your software runs shouldn’t be part of that conversation either, even though some of it will outgrow the home: the box goes down, a model won’t fit, strangers need to trust what they open, the data lives somewhere else. When that happens it should move to the cloud, or to a frontier lab or a data provider, in one click, with nothing rewritten. That works only if identity, AI, waitlists, invite codes, queues for when compute is scarce and notifications behave the same at home and in the cloud, so they have to come from one platform, whether it runs in a data center or in a living room. I picture Vercel as that platform: bring your own cloud today, bring your own box next.

Keeping all of that invisible is the job of people like me, who love the plumbing. We should carry the complexity and hand everyone else clean abstractions, the way electricity reaches the wall without anyone thinking about the grid, so they can spend their creativity on what they want to exist. Vercel’s mission is to enable the world to ship the best products, and I read “the world” in that sentence literally, a small store in rural Uruguay included.

I’m living in it already.

A computer in your home that you own, that builds whatever you need, and that you run just by talking to it: for me, that isn’t a forecast. I’ve been building it for myself for four years.

Afterwards I had meeting after meeting with people who wanted to make it a real product, and folks at Shopify were generous with their praise. Everyone was surprised by how finished it was, and nobody in those meetings could see why. Most of its technical decisions had been made before I opened the first agent chat, so security and completeness were already taken care of. The tree had been cut before I got on the bus. A few years ago, this would have taken a team months.

I didn’t have to build Lintel. I only had to ask.

Where I want to build it.

What I run at home works for one family, and that is its limit: a computer that builds for everyone can’t be finished alone in a spare room, because it needs infrastructure that works for millions of people. The best place I know to build it is Vercel, a company that has spent a decade making hard things simple, one cut branch at a time, and that is full of people who think about this problem every day. I’d love to work on the layer that powers this at scale, and to find the people there who want to carry it into every home, and into every small store like my mom’s.