Raahel Baig

Using Claude Code as a general agent

When Anthropic announced Claude Skills in October 2025, Simon Willison said this in his blog post:

Claude Code is, with hindsight, poorly named. It’s not purely a coding tool: it’s a tool for general computer automation. Anything you can achieve by typing commands into a computer is something that can now be automated by Claude Code. It’s best described as a general agent. Skills make this a whole lot more obvious and explicit.

While I had this in mind, using Claude Code for music is something that didn’t come cross my mind before. Josh Cohenzadeh got Claude Code to write an original song, an original EDM song, original rock song with vocals, and an original album — actually getting it to create an audio file for each using one-shot prompts. He didn’t mention the model he used in his blog post but I assume it is Opus 4.5 with extended thinking (the transcripts he added to the post had thinking blocks).

Reading the blog post made me realize that I am still blinded by Claude Code’s poor naming because I haven’t used it for anything other than building software. I use it everyday and I am aware that it is not “just a coding tool”, yet I haven’t really of thought of doing anything with it.

So, I thought why not get it to create a video. Specifically, a bar chart race video illustrating the most popular girl names in Telugu states from 1950s - 2020s (because why not?). So that’s what I did. I started a claude code session with --dangerously-skip-permissions and gave it the following prompt (heavily inspired from Josh’s prompts):

“Create a infographic kinda video showing the change in trends of most popular girl names in Andhra Pradesh and Telangana from 1950s to 2026. It is one of those videos that has horizantal bars with the most popular names and the bar sizes keep changing depending on the popularity of the corresponding name as the time passes. You can use the internet to fetch the data that is required to create this video file. If you can’t find the data from 1950s, start from the year from which you can find the data. Do not look in any other directories or do any exploration of any kind. Don’t ask me any questions. You should not complete until you’ve generated a video file in the current directory for me to watch. That is the only output that is acceptable.”

It gave me this:

It summarized the trend as follows:

  • 1950s-60s: Deity names dominate — Lakshmi leads massively, followed by Saraswati, Savitri, Sita, Subbamma, Parvathi
  • 1970s-80s: Padma and Vijaya surge into the top 3, Radha, Manjula, Sujatha climb rapidly
  • 1980s-90s: Modern names like Sunitha, Jyothi, Kavitha overtake the classical names
  • 2000s: Lavanya, Divya, Vaishnavi take over as naming modernizes
  • 2010s-2020s: Saanvi rockets to #1, Hanvika, Thanvi, Dhanvi surge — reflecting the current trend of shorter, modern Sanskrit-derived names

It gathered the data for the trends and wrote a python script which uses FFmpeg to generate the video. I don’t understand how the code works. You can read the full transcript of the session here.

All of the research, data gathering, code and the video generation was done within 8 minutes. I am mightily impressed with the output. This only makes me want to do more. I really do believe we’re still not really scratching the surface of its full potential. More such experiments to come!