Claude and I built an app

Lobsters Hottest Tools

Summary

The article announces Roboterm, a minimal terminal emulator built with Claude AI, and reflects on using AI for coding personal projects, highlighting improvements in productivity and code quality.

<p><a href="https://lobste.rs/s/r8x9et/claude_i_built_app">Comments</a></p>
Original Article
View Cached Full Text

Cached at: 08/19/26, 02:25 AM

# Claude and I built an app Source: [https://folkwolf.net/claude-and-i-built-an-app/](https://folkwolf.net/claude-and-i-built-an-app/) Announcing[Roboterm](https://github.com/mattrose/roboterm)\. It’s a minmal terminal emulator that basically does what I want, and not much else\. It supports tabs, and split panes\. There were a few motivators for me to write it, if you can call it that\. The first was that I simply wanted to write a terminal emulator that I wanted to use\. I’ve maintained[Terminator](https://github.com/gnome-terminator/terminator)for over 5 years now, and I like it and still use it, but it’s macos integration isn’t great and I have stopped using Linux on the desktop completely in favour of MacOS\. Even though Terminator is still designed around systems administrators who log into dozens if not hundreds of servers a day, my workload doesn’t really fit that mold\. I wanted something different\. I also wanted to see working examples of python gtk4 code on macos\. The terrible macos integration on Terminator has always bothered me, and I couldn’t find any examples of python gtk code that ran well on a mac that I could follow\. Also, I wanted to update Terminator to GTK4, so I wanted to have a**good**example of Python GTK4 code that worked seamlessly on a mac\. Lastly, I wanted to test out the abilities of AI\. I have tried to have AI write this code for a few years now\. Every 6 months or so, I would pick up whatever the best, latest AI coding tool and tried to code, basically, this\. Up until last month, I gave up in frustration\. This was written using claude tui\. Just telling claude what I wanted, and had it write most of the code\. It was fairly slow going at first, until I figured out how to work with claude, and claude was able to build up a good CLAUDE\.md\. Now it’s fairly smooth to add new features\. This has been a real education in the ability of AI\. In some ways, it is a much better coder than I am, it writes far more thorough tests, and documents it’s decisions in CLAUDE\.md, leaving me with something that I can easily read and make sense of\. The entirety of the python code is less than 2000 lines for a fairly capable terminal emulator\. It has also allowed me to produce decent code without getting into a “flow state”\. I can sit in the evening and watch TV with my wife and direct claude at the same time\. When you have a full\-time job and a bunch of other hobbies and a family, this ability to multitask and still produce something worth using is nice\. Having said that, what I’m having it write is toy code\. It’s handy, but it’s light\-years away from the code that I deal with at work, where uptime is golden, and the stakes are much higher\. At work I use AI mainly as a research assistant, and still write my code myself where I know exactly what the code is doing\. I can’t imagine an AI that I would trust to write the code I need to write for work\. Anyway, if you want a simple, hackable terminal emulator where you can probably have claude add any feature you want, give it a try\.

Similar Articles

The Emacsification of Software

Hacker News Top

The author discusses the frustration of reading Markdown in the terminal and describes using Claude to quickly build a custom macOS Markdown viewer (MDV.app), illustrating how AI enables rapid creation of personal software tools.

Boris Cherny on Trying to Get Claude Code to Rewrite the Claude App

Hacker News Top

Boris Cherny, head of Claude Code at Anthropic, discusses an experiment where he directed Claude Tag to rewrite the Claude desktop app in Swift using a Mac VM and pixel-by-pixel comparison, with the task still running after two weeks. John Gruber critiques the approach as fixing technical issues but not the app's fundamental design problems.