| published by | Software Crafts |
|---|---|
| in blog | Software Crafts |
| original entry | django-deadcode: idea to release in under 2 hours |
A few weeks ago I noticed a toot from Jeff Triplett about Anthrophic releasing Claude Code for the Web. This was the final spark that coalesced a few different thoughts that had been lingering in the back of my head, some of I have written a bit about before. The first thought was the speed of prototyping that agentic AI enables. Essentially ideas or tasks can simply be executed rather than being written down and the time allocated to develop these ideas go from weeks to days or even hours. The second thought is related to the first in that tools like Agent OS allows for AI to build out products in a more reliable way for the most part. I have also been pondering how I can use my mobile more as an engineer, the Github app is ok for PR reviews, but to date building anything needs a larger screen.
The final thought goes back to the toot from Jeff and Claude Code on the being possibly the cloest thing so far to my post from day 282 about how our tooling doesn't yet fully leverage what AI can do for us.
Well this led to me creating two things this week. First was a template repo on Github which is only loaded with an install of my Django Agent OS profile. This enables me to quickly start a project in the browser without having to open a terminal or potentially even be at my laptop, I could start a project from my phone. Second was an experiment to see how much I could get Claude to build from the browser. I took my idea from Day 71 about analysing a Django codebase for dead code that could be removed. Over the course of about 2 hours of my time and letting Claude along with Agent OS, I have a package released on PyPI.
The unlock here is that I have yet to clone the repo to my laptop. In fact, the most time consuming part has been getting CI to work nicely to release new versions. Upon reflection this is something to go into the template repository, but then not every project needs to be uploaded to PyPI.
It's been a fun experiment to get a working proof of concept out the door so quickly, but it needs a bit more refinement, testing and review before I recommend anyone else use it! If you want to have a peak, the repo is here and package is here