why i open source everything

The URL of this site is obedev.is-a.dev. I didn't buy that domain. It comes from is-a.dev, an open source project where you open a pull request with a JSON file and get a subdomain for free. My personal site starts with someone else's open source, which feels about right.

I'm Obed, a developer from Panama. During the day I build software for industrial operations: ESP32 boards reading Modbus sensors, data going out over Sigfox, dashboards for the people running the process. The rest of the time I build tools for coding agents.

I started programming during the pandemic, before AI could write any of it for me.

Almost everything I make ends up public. Right now that's:

• MendCode, the agent harness I use every day.
• Whittle, a 171-word skill for coding agents.
• NotchNull, a MacBook notch you rebuild by asking your agent.
• ukiyo, a game engine in C# where the agent changes the game while it runs.
• Cellium, mflow, nubli, a flight simulator that runs in the browser, and some smaller CLI tools.

Most of it is MIT. Here's why.

If you can't read it, you can't trust it. Whittle claims it delivers more of the product than other skills at a lower cost. In my published runs it delivered 94 to 95% of a hidden product checklist while costing less on average than running with no skill at all. That's an easy claim to make, so every run, every patch and every score is in the repo. If I'm wrong, anyone can check and show me where.

Agents work better on code they can see. NotchNull ships its settings and widgets as plain files, and the app itself is open. When files aren't enough, your agent reads the Swift source and changes it. A closed app would stop exactly where it gets interesting.

Somebody else's open code has helped me more than any course. A few days ago I had a pull request merged into Symbex-1, a project running 1-bit neural networks on microcontrollers: integer baselines, a kernel for the ESP32-S3 and a test suite. I could only do that because the code was there to read.

Open is better, and agents made it cheap. I started publishing for the same reason most open source developers do: code people can read, fork and fix ends up better than code they can't. What changed is the cost. Opening something used to mean cleaning it up and documenting it on your own time. Now an agent does a lot of that. REA is the obvious example.

Nothing ages faster than private code nobody uses. My first public tool was gitd, a small script for cloning repos, back in 2023. It has six stars. It's still there, it still works, and someone can still find it.

What comes next

This blog is going to have opinions, and each one will come with something I built or measured, so you can check it. The first one is about why minimal-code skills save money by shipping less product.

If you think I'm wrong about something here, open an issue on the repo in question. That's the point of doing it in public.