Cathedral, Bazaar, Garage

February 13, 2026

I use QuickBooks Online to manage a couple of small businesses. It’s a good product, worth the monthly ~$30 fee. QuickBooks has hundreds of features, and most small businesses need about twenty. It constantly pushes customers to use features that we don’t want or don’t need, and then they seem to rewrite the UI every six months.

Google Docs has the same problem. Most of us need about twenty word processing features, and Google Docs has a thousand! When it’s time to learn a new feature, we dig through online docs, Reddit threads, YouTube videos, only to discover that the feature doesn’t actually do what we assumed. (Microsoft Word is already famous for this, but I haven’t used it in ages. I’ve heard that people still use LibreOffice/OpenOffice; the screenshots don’t do it any favors.)

This is the cost of software built for everyone – software not built for you. The features you need are buried under features you don’t need, wrapped in a UI designed by committee, and changing on someone else’s schedule.

To be clear, I’m not picking on commercial software; many open-source projects have the same problem. The new world presents new challenges for both the Cathedral and the Bazaar, and new opportunities for your own Garage.

Cathedral? Bazaar?

In 1997 Eric Raymond published his famous essay “The Cathedral and the Bazaar”, then he published it and other essays as book. If you haven’t read it then I’ll summarize.

The “Cathedral” is a team building behind closed doors, releasing carefully planned software versions. The “Bazaar” is open-source development in the wild – many contributors, frequent releases, users as co-developers.

The shift

The new-old paradigm is the “Garage”.

I recently read a blog post by Roberto Selbach about replacing subscription software with apps he built himself using AI tools. One of those is a dictation app called Jabber, to replace Wispr Flow. I use Wispr Flow, so I tried Jabber. Once installed, I immediately identified a missing feature: let me change the hotkey that triggers dictation.

So I forked Jabber and made the hotkey configurable. In. Thirty. Minutes. Most of my non-metaphorical Garage projects take much longer!

Jabber is written in Swift. I have never written a line of Swift, and I still haven’t. Code agents have crossed a threshold, enabling this humble backend software engineer to ship a native macOS feature in a language he couldn’t pick out in a lineup, delighting every single one of his users.

This isn’t about fun, novelty, or hobby projects. The world is changing, and I can now build and maintain tools that fit me exactly in my Garage.

Less software, fewer problems

As I’ve continued tweaking my Jabber fork, I’ve also removed features. I don’t need automatic updates because the source code, build environment, and end user all live on one computer, and because every one of Jabber’s missing features is bug free.

This is where personal tools have a structural advantage over both commercial and open-source software. A tool built for a million users carries the weight of a million use cases. A tool built for me is pretty boring – simple code, few dependencies. Such a tool is faster to implement, faster to deploy, uses less memory, and is easier for Claude, Codex, or even me to understand in the future.

Security benefits directly follow: fewer features, fewer dependencies, smaller attack surface. You might be thinking “given enough eyeballs, all bugs are shallow”. But these tools are so small that I can audit them myself.

I’m advocating for more software that is non-commercial and closed-source. Garage software.

Your data stays home

Wispr Flow sends every recording of my voice over the public internet to some backend service for inference. I’m sure the people at Wispr Flow are well-meaning and do their best to mitigate the security risks. Their dictation quality is that good precisely because (1) they can leverage larger models that wouldn’t fit on a laptop and (2) they certainly collect feedback from my interactions, if not from the recording content itself (don’t forget the value of metadata).

And that’s fine! Wispr Flow is a great product, everyone should totally check it out!

When I build a tool for myself, I get to design the tradeoffs myself. If I need dictation, I can decide to use a local model (low latency, good privacy, good security, lower quality) or a commercial online model (high latency, mid privacy, mid security, very high quality). That’s my call to make when it’s built in my Garage.

Every SaaS tool includes tradeoffs someone else made. Every Garage tool includes tradeoffs you made.

Where this leads

The economics of software have inverted. Building custom tools was expensive; using common tools was cheap. We don’t need the Cathedral or the Bazaar to get there – we can build them in our own Garage.