March readings and discoverings

Clean codebase with Knip, harvesting energy for headset, agentic AI, my new site.

Hey, this is a smart way of assembling 18650 batteries together to build a power pack with desired voltage and capacity.

I’ve always loved projects that try to harvest energy from the environment in order to work without any battery, so I have to mention here the project by Alessandro Sgarzi, who made a headset that harvests energy from noisy environment, typical of some modern cities. This is possibile with piezometers.

One thing to try using LLMs, suggested by Andrea Belli Contarini in his Linkedin post, is to use ASCII art to plan things to do, such as DB diagram, UI wireframes, etc.. This will lead to a 60% of savings in tokens.

I’ve just discovered the Knip tool to clean up React/TypeScript projects that, after years of development, have grown too much. The size of codebases is a problem that can only get bigger with the use of coding agents that like to write a lot of code.

This month I decided to switch my AI budget from ChatGPT to Copilot. I’m slowly leaving my ChatGPT Pro account; I will keep using Openai on the free plan, with much more awareness of data. The key point was the agentic coding tool of Copilot inside VSCode, which is really awesome. I made my new site (this one), integrated with my existing WordPress installation, in about 8 hours of agentic coding, using both Lovable for design and Copilot agents to make changes and integrate it with WordPress (I worked on PHP for WordPress and React/TypeScript for the frontend). I’m really happy: my site is now headless, it’s a faster WordPress site and cleaner.

One thing that I want to test (when!?) is using Claude cowork for local agentic coding with Llama (also here).

I also switched to Umami Analytics, which seems slightly better for GDPR since it also has a self-hosted version, useful if you have privacy-first clients.

Currently trying Strawberry from Sweden. Its agentic crawler is really simple to use, and their characters are funny. I still need to test the results more.