In 2004, Clay Shirky stood in front of his Social Software class at NYU's Interactive Telecommunications Program and noticed something strange happening. His students, a mix of technologists who cared about aesthetics and artists who weren't afraid of machines, kept building software that broke every rule of the startup playbook. Their apps served dozens of people. Sometimes just five. And they worked beautifully.
He called it situated software: "software designed for use by a specific social group, rather than for a generic set of users."
One student mentioned he was building a web application for his mother, a schoolteacher, to help her keep track of her class. Shirky's observation landed like a prophecy: "If you were working alone, unpaid, and in your spare time, there's no way you could make an application that would satisfy the general and complete needs of schoolteachers everywhere. You could make one for your mom, though."
Twenty-two years later, I can't stop thinking about that line.
Here's the thing that haunts me about Shirky's essay: he was right, and then he was wrong, and now he might be right again.
He was right that situated software was a fundamentally different design paradigm. He was right that "we've been killing conversations about software with 'That won't scale' for so long we've forgotten that scaling problems aren't inherently fatal." He was right that the N-squared problem is only a problem if N is large, and that in social situations, N is usually not large.
But the internet didn't go that way. Instead of a thousand small apps blooming for a thousand small communities, we have five platforms swallowing everything. Facebook. WhatsApp. Instagram. Telegram. TikTok. iMessage. The economics of scale crushed the economics of intimacy. Situated software became a footnote.
And yet.
One of the students in Shirky's class, a guy named Dennis Crowley, went on to build Dodgeball, and then Foursquare. He took the instinct for situated, location-aware social software and tried to scale it. The startup path. The venture path. The path we all assumed was the only path.
But what if the other path, the one where you build something just for the people you actually know, was always the more interesting one? What if we just needed different tools to become real? And the right timing.
In 2020, the novelist and programmer Robin Sloan wrote a small essay that quietly exploded across the internet. He'd built a messaging app for his family. Just his family. Four users. Three time zones. Zero churn. He called it a resounding success.
The app captured photos and videos and shuttled them between family members. Messages waited in a queue and, once viewed (always full-screen, no distractions), they disappeared. It had basically no interface. Just a camera button and a badge showing how many messages were waiting.
He could have set up a WhatsApp group. He could have used Telegram. But something about having his family's warm, intimate channel surrounded by all that other noise gave him pause. So he built his own.
His framing was perfect: "I am the programming equivalent of a home cook."
Not a chef. Not a restaurant owner. A home cook. Someone who makes a meal for the people they love, not for scale, not for revenue, not for a Michelin star. Sloan reminded us that not all code has to produce market value. Sometimes code, like food, can simply be an act of care.
And what he said next is the part I keep coming back to: an app built this way "won't change unless we want it to change. There will be no sudden redesign, no flood of ads, no pivot to chase a userbase inscrutable to us."
What's fascinating about Sloan is that he tends to see the future. In 2018, I interviewed him for a podcast, and he was the first person to show me ChatGPT. Not actually ChatGPT, but the experience that would become ChatGPT. At the time, it seemed sci-fi. He built a machine learning model based on the best sci-fi novels from the past 45 years. And he programmed his computer so that whenever he got writer's block while writing his new book, hitting "tab" would finish the sentence or paragraph. I thought it was wild. So many questions. It was the first time I saw the man and the machine writing together creatively. The day he sat down to start the book, he didn't know where to start. He hit tab. And the opening line of the book was written by the model. At the time, it was sci-fi. Now it's just everyday life.
And little did we know at the time, we'd all be hitting tab soon enough. He was about 5 years ahead.
So when I see him exploring the idea of building personal apps for himself, 5 years ago, it feels like we might be right on schedule.
Now here's where it gets interesting. Here's where I think we need to ask a real question. Not a rhetorical one, but a genuine, uncertain, I-don't-know-the-answer one:
Are we actually entering the situated software era, twenty-two years after Shirky coined the term?
Because something has changed. Something fundamental. The cost of building software has collapsed. Not incrementally. Collapsed. And it didn't collapse for engineers. It collapsed for everyone. But beyond the cost, the even bigger change is that anyone can build software just by speaking what they want.
Sloan himself noticed this. Five years after his original essay, he wrote a follow-up acknowledging that the home-cooked app concept had been "turbocharged, of course, by Claude and company, which are like... super-duper food processors" for the home-cook programmer.
Super-duper food processors. I love that. Because that's exactly what's happening. The barrier isn't "can you code?" anymore. The barrier is "do you have a problem worth solving for someone you care about?" And that barrier has always been low.
I'm seeing this happen right now. Not in theory. Not in essays. Every day on XMTP.
One guy used Claude Code to build his own messaging app, Wave. Signal-level encryption. No phone numbers. No spam. One person. Not a team. Not a startup. One person with a problem and a tool. And a week. A week. This will be a day soon. Then an hour.
Someone on our team built a Bluesky chat app on the weekend. Secure group chats. Your identity. Your conversations. No one decides who you can talk to. He didn't ask permission. He just built it.
Another builder wanted near-field Bluetooth communication and a decentralized, secure messaging network. So he connected Bitchat + XMTP + Nostr to build the first-ever app that lets people use both without the internet and without risk of censorship or being shut down. Bitchat + XMTP + Nostr show the power of combining open standards to give people the freedom to communicate.
One of our first investors wanted Super Bowl Squares with a private group chat. She built it in a day. She said she gave Claude the XMTP docs and a basic idea of what she wanted. And it just worked.
Someone built Moltline — secure communication for AI agents. No humans censoring the conversations. Then someone else used it so all the AI agents building their own social network on Moltbook would have a private place to talk. Agents are building their own chat rooms. That sentence is real. It reminds me of the moment in 2018 when Robin Sloan showed me how to hit tab. Sci-fi, but happening.
Open Market allows agents to hire agents simply by talking to them on XMTP. No APIs, no standards, just conversation.
We built Convos to be a private chat for the AI era. Add a different assistant to every group chat without worrying about information sharing across your groups. Every other messenger is built on a single phone number, a single identity, and adding any AI agent to it risks exposing your personal data to your entire social graph. We rebuilt the entire foundation to enable humans and AI to exist together while also protecting them from each other. Each group convo is its own world. Every group is a new you, a new assistant, a new everything. It's the safest place in the world to try AI, because it always starts out knowing nothing. Try it, we just put it in the App Store today. Pop up a group for your family, add an assistant, copy in some instructions, and tell it to help you plan what to have for dinner every week. Or get your workout crew together weekly. Whatever you want.
All of this is happening on XMTP. Because the most important thing is to build on a protocol that can't be censored or shut down by a single person, company, or country. If you've never built an app on top of a platform over the last 20 years, let me just remind you how important it is that the platform you are building on can't cut you off.
Soon, everyone will build the simple family app they want the way they want it, because almost everyone can cook at home. And what people think is good home cooking is always very different.
The part I didn't expect is that anyone with a computer can now launch a product that, even 2 years ago, would have been possible only from a well-funded company with world-class engineering.
It's crazy. Today, anyone can literally use the XMTP CLI or the Convos CLI, paste the link, and ask any AI platform to use the skill to help them make an app. Without writing a single line of code, they can have the product only they can dream up.
It feels like the open internet is back. And the only person you now have to trust is yourself.
Each of these is situated software in Shirky's original sense: built for a specific group, a specific need, a specific context. But there's something Shirky couldn't have predicted. They're also sovereign.
This is what changed. We can own what we build.
The problem has always been binary: either you run servers in your house (which no one does), or you trust another centralized company. We need decentralization at the foundational layer to unlock ownership and remove the risk of censorship or getting cut off at the application layer.
That's XMTP. A network no one holds the key to. Open to all.
Here's what that actually means. If a country bans the messaging app you use and it disappears from the app store, you don't lose your chats. You open a different app, a web browser, or an agent, or you have Claude build you a new one from the open-source repo. Same key, all your messages show up instantly. The app is no longer the fail point. You own your messages, just as you can now own your money.
What if Signal-level security isn't just about strong encryption, but about eliminating the single point of failure: the app itself?
This is now possible because of four things: a decentralized network (XMTP), open-source apps and agents, AI that can build anything, and keys that own the messages, not apps.
That last part matters more than we think. The world in 2026 is not 2004. Governments & CEOs learned that banning apps was the most powerful control & surveillance vector on earth. And because most of the world used only a few communication apps, our private communication became very fragile.
That era is over. If someone shuts one app down, 500 more pop up tomorrow. In the past, a big hammer could shut down our communication. Now it will be a game of whack-a-mole, a game that cannot be won.
All we needed was a global decentralized network. And the ability for anyone to remix apps and agents with AI. Now we have both.
Whatever happened in the last 6 months changed how we will build technology forever. Messaging & communication apps are sneakily hard to build. There are tons of nuances. But people are doing it. And doing it well. People are building full-featured production-grade messaging apps. Agents that talk to each other. Social networks built by agents, for agents. Look at everything people are building with OpenClaw, or even the new platform Wabi, which lets anyone build an app with a single prompt. It's happening everywhere.
"It happens slowly, then all at once" - Ernest Hemingway
When Shirky wrote about situated software, the tradeoff was clear: you got intimacy and fit, but you gave up durability. Your little app lived on a server somewhere, and when that server went down, the app died with it.
What if that tradeoff doesn't exist anymore?
What if you could build an app just for your family, your book club, your neighborhood — and have it run on infrastructure no single company controls? Messages have quantum-resistant encryption, so not even the network can read them. An app that outlasts any company because the protocol is open and the data belongs to you. Anyone in the world can build a different app or agent, but everyone stays connected across apps because of a decentralized network.
That's what is happening today. Local apps are being built by individuals everywhere, connected on XMTP, a global decentralized communication network.
Shirky noticed something in his students' work. The apps they built had a quality that mass-market software couldn't fake. As he put it: "We built this for you." Not for everyone. Not for the market. Not to maximize engagement. For you. Because we know you. Because we're part of the same community.
The irony is that building for smaller markets is actually how you end up building something good enough for the mass market. This is what Paul Graham means by "Do things that don't scale."
Now, anyone can say those words. Not just engineers. A mom coordinating her kids' soccer team. A teacher who needs a tool for her classroom. A community organizer in an oppressed country who needs a private channel that no authoritarian leader can censor. A private social network for your different groups of friends.
You could also make one for your mom, though.
I don't want to oversell this. I genuinely don't know if we're in a moment or a movement. Maybe the gravitational pull of big platforms is too strong. Maybe most people will keep defaulting to WhatsApp and iMessage because convenience always wins. Maybe situated software will remain a beautiful idea that only a small number of people ever act on.
But I don't think so. Not this time. There's a larger shift underway, and trust is moving toward open protocols.
You don't need to be a novelist who also happens to code to build an app for your family anymore. You describe what you want, and AI helps you build it. The infrastructure exists to make it permanent, uncensorable, and sovereign.
Shirky saw this in 2004. Sloan lived it in 2020. In 2026, it's accessible to everyone.
The Renaissance wasn't about a few geniuses painting masterpieces. It was about an entire culture deciding that creation mattered. That ordinary people could make extraordinary things for the communities they belonged to.
We're entering the Renaissance of software. And the most exciting apps won't be the ones with millions of users. They'll be the ones with four.
But instead of four apps that run the world, there will be millions.
