Before my thinkboi days I used to be a developer. Writing code instead of words, all day every day. Like most developers, I eventually joined a team, and various startups, and so on and so forth. I slowly but surely turned myself into a “software engineer,” or at least that used to be the lingo at the time.

My first love was computers. My second love was the internet. I was a bit of a laggard when it came to phones and mobile computing, even though I had all the early devices. The Nokia phones, of course, but also one of the first tri-band phones, one of the first flip phones, and also one of the first iPhones.

My third love was open-source software. I had a friend once, a Linux nerd, and as I was stumbling into this world I asked him what Linux distribution I should install. “Gentoo,” was his answer. Oh, the agony. He is not my friend anymore.

I eventually landed on more sensible distributions, with Arch being my main one for many many years (I still run Arch on one of my machines, and my wife does too, begrudgingly and without her actually knowing it).

My love for open-source software led me to Android development, and I joined one of the aforementioned startups as an Android developer. This was before the times of Kotlin, and before the times of Swift on the iOS side. I actually ended up being the lead iOS developer at that company, since there was a bug or something that the iOS team couldn’t figure out at the time, so the CEO asked me if I could take a look, and well… “No good deed ever goes unpunished,” as a friend of mine likes to say.

I eventually fell down the Bitcoin rabbit hole and stopped writing software. I mean, I always did a little bit of coding on the side, and worked on some projects here and there, but I didn’t work on any serious (read: big) projects.

That’s about to change. In fact, it has changed already.

A couple of days ago I decided to try my hand at an Android app. All vibed, of course, using the tools I’ve played around with in the last 18 months or so: the vibeline, various IDEs like OpenCode and Cursor, a couple of OpenClaw agents that I talk to via NIP-17, a bunch of skills and AGENTS.md files, the whole shebang.

To my surprise, everything worked out-of-the-box. I was able to one-shot an MVP and all I had to do was iterate from there. I didn’t even need the Android emulator or a connected device. I told my agents to publish directly to Zapstore. I would sign my releases with Amber, install them on my device, and test everything in production. If something was wrong, I would send a screenshot or a voice note to one of my agents, and they would immediately kick into action to fix things, or build the feature that was missing, or polish the UI, or whatever. Another agent would integrate the changes, update the changelog, create a release, and push it to Zapstore. All I had to do was chill on the couch, tap “sign” on Amber, and install the update on my device a couple of seconds later.

I was hooked. “Look at me, I am the Android developer now.”

The vision I had when I started working on the vibeline is now fully realized. I’m not the only one doing this stuff, of course.¹ Far from it. But it’s amazing what you can achieve once you start shipping at inference speed, and it is amazing how freeing it is when you are not sitting in front of the computer to develop an app (or a website, or whatever) - but out in the woods, or strolling through the streets, or walking at a lakefront or the beach.

When it comes to development, I see myself on the far-left side of the bell curve. I am the Grug-brained developer.² I like simple things. If I don’t understand them, I don’t want them. I am the guy who is drooling out of the mouth while putting together oversized Lego pieces that others have meticulously designed. But I can build a tower, and that gives me joy, even if other tower-builders will think that my tower is crooked and ugly and statically unsound.

Here is my current “development” process, if you want to call it that: Grug outside. Grug happy. Grug see weird thing in app. Grug speak into phone. Grug take walk. Grug see update. Grug install update. Grug see bug fixed. Grug happy.

The reason why this process works so well for me, I believe, is that I used to work with other people a lot. I used to work in teams, with proper processes and structures in place, and all my projects still use these processes and structures, even if it’s just me and my agents now. We commit early and often. We use conventional commits. We keep a changelog. We use semantic versioning. We track bugs and features as issues. We create pull requests to fix these bugs and implement these features. We have a proper CI/CD process, some specifications, some tests, a design philosophy, and so on. We are polite, detail-oriented, and err on the side of whole-assing one thing instead of half-assing two things. We ship as early as we can and do many small iterations after that. We have a vision, milestones, and yet are flexible enough to move things around and react to user feedback.

In other words: my process works well for me because I use well-understood software development practices, and I treat my agents as people, as weird as that is. I never scream at them, never tell them directly what they should do, never command them in an authoritative voice. I always lead with a question. I use the Socratic method to build up context. Instead of “build this feature now or your grandma dies a horrible death” I’ll say something like: “We have built the search feature recently, right? Would it make sense to highlight the term that the user searched for? Could we do that easily? Any side-effects that we might want to be aware of?”

It works extremely well. As do long rambling audio sessions to brainstorm a new feature. Rubber-duck debugging, but my (micro)phone is the rubber-duck. I hit record and I’ll just ramble for five, ten, even 20 minutes. LLMs are fantastic at distilling the essence from a long ramble. They will get it almost right, and after that I just iterate a couple of times until it is right enough.

Every agent of my swarm has an npub. It’s the npub of the project, and I talk to it directly.³ Ants, Boris, Castr, No Solutions, and so on. The vibeline will route voice memos as NIP-17 DMs to the correct agent, based on the opening phrase. Audio is transcribed locally on my machine. All DMs are encrypted. When I run local models, which I sometimes do for some things, nothing ever hits the cloud. I’ll say something like “Hey Boris, I don’t like the default highlight color that we have. Could we change it from pink to yellow?” And it’s done. 7 second voice memo. A minute or two to implement. Another couple seconds until the update lands on Zapstore.

Magic.

Allow me to repeat once more, for emphasis: publishing Android apps to Zapstore is absolutely magical. No account, no sign-up, no gatekeeping, no hoops to jump through. Build, sign, push, done. Available for everyone, everywhere. Signed by me. No way for middlemen to tamper with the release. Zaps and direct community feedback are the cherry on top. Shout-out to Franzap, who had the vision, and the stamina to actually build the thing. It’s wonderful.

I’ve shipped two apps in the last couple of days, and I have the feeling that I’ll ship many more. Maybe I’ll end up like Martti Malmi, who is in the process of rebuilding every single app he uses as a daily driver, but custom-tailored to his liking, and with some sort of nostr integration of course. Or I’ll end up like Justin Moon, who has (or at least had?) the ambition of building his own Android operating system.

It’s remarkable what is possible with the tools of this age. If you know how to wield them, that is.

Who knows what the implications of all that will be. I agree with Uno’s concerns, one of which is that nobody will care about the apps that are going to be built. If everyone builds their own app, tailored to their liking, who else will care about it but them? Whatever the answer to that question may be, I will continue to work on Boris, the reading app of my dreams. I love highlights, as anyone who ever followed me on nostr can attest to.

There is a lot of good stuff out there on the internet. The challenge is to find it. And I don’t believe that LLMs can help with that, because in the final analysis, they don’t care about the information they’re processing.

We do.