Software Development

In my previous post, I talked about my experience writing software using LLMs. I realize, in shock, that for the past three months I haven’t written any code. Today (who knows in another three months) my current workflow is as follows: Divide the work in chunks, as I’d do if manually doing the code Prompt to the assistant Wait. If I foresee a waiting time of more than ~30 seconds, switch to another task Git diff all its changes Commit if it looks good. Otherwise back to step 2. The assistants are getting so good that I’m leaning into skipping steps 1. and 4. Of course, doing that means totally relinquishing control of the source code, a jump that I’ve already done, ocasionally, when under heavy time pressure. It hasn’t backfired, so far. I’m wondering if a time is coming when I just write prompts and never look at the code again. Unlike many developers out there I haven’t really tested the assistant ability to do a bigger task unsupervised, and that is the topic for this entry. ...

As I’ve written in my post about GitHub Copilot Costs, the only AI subscription I’m paying for right now is GitHub Copilot; a developer tool meant to be used alongside an IDE like Visual Studio 2026, Visual Studio Code, Jetbrains and so… However, I sometimes feel the need for a general-purpose LLM assistant. Of course, I can fire up Visual Studio and ask GitHub Copilot whatever non-coding question comes to mind because the tool itself is just an LLM, extended with conveniences for software development. Furthermore, many functions that one might think are not strictly related to coding, like analyzing images, are also supported. However, this approach is inconvenient because when not working I don’t feel like booting an IDE any time a question comes to mind. Regarding mobile usage, GitHub has an Android app I can use to chat with Copilot, but it is packed with repository utilities and other things I don’t care about when not working. All in all, I get by using GitHub Copilot for my real-world interactions, but I have to admit is not really convenient. ...

In my last post I spoke about how I’ve been gradually adopting AI-assisted coding. I haven’t thought much about costs because the tool had been experimenting major improvements over time and I felt every euro spent had been well deserved. Now, I realize I’ve gotten to a point where I rely on AI assistants to do my day job, and going back to manual coding would make my productivity take a major hit. Anyway, today I feel the need to check how much and where should I be spending. ...

My AI assisted coding history It seems ages have passed by, but it was only 2020 when ChatGPT-3 was released. We developers and technology enthusiasts had been aware of it, but in November 2022 ChatGPT-3.5 was released to the mass market and rocked the world. This is the timeline: Year Model Impact 2018 GPT-1 Research prototype. 2019 GPT-2 Significant attention in the AI community, but not widely used by the public. 2020 GPT-3 Impressive capabilities, mostly accessed through APIs and developer tools. Nov 2022 ChatGPT (GPT-3.5) First version to become a global phenomenon. Mar 2023 GPT-4 Major quality improvement, building on ChatGPT’s already massive user base. At first I used it like many other developers, I assume. Just copy and paste from the browser or ChatGPT app to the IDE. This was fine for scripts or very localized problems, and that small assistance to my daily coding was enough for a couple of years, until I subscribed to GitHub Copilot in August 2025. ...

At my day job we use a dual database setup, a standard relational database and Elasticsearch when the volume of the stored data gets too big, and it does get big. I don’t feel at ease with Elasticsearch. First, I’m already a veteran. I graduated from computer science in 2002. Back then, when you got out of college you had your standard SELECT...FROM...JOIN...WHERE...GROUPBY...HAVING in the playbook but NoSQL was not something most people know about. In fact, according to Wikipedia the term NoSQL was popularized in the early 2000s, MongoDB launched its first version in 2009 and Elasticsearch in 2010. ...

Since its first edition in 2018, I have been attending Cooltural Fest, a music festival that takes place yearly in my hometown of Almería. I use the months before the festival to learn about new bands. The organization keeps an official Spotify Playlist that I don’t find very interesting since they, rightfully, emphasize the bigger bands and their most popular songs. Every year I copy all entries to my own list, trim most of the headliners’ content, and keep adding entries from the smaller artists. ...