Luca - Dev <-> DevOps

@luca.lanziani.com

Hi, nice to meet you, follow me for Tech content, random jokes, food complains (I'm Italian). Remote Head of DevOps and Platform Engineering (8/5). Trying to make software teams run smoother, my money grow faster, and my lifts get heavier.

Yesterday, I started watching Due Spicci by Zerocalcare on Netflix. Today, I was in Rome for a celebration, and the church was right in front of the restaurant from the series. It honestly felt like a glitch in the Matrix.

Bild

TIL: People are saying it will be harder to learn Software Engineering if you delegate the implementation to Agents. Today, I actually learned about flock thanks to an agent that implemented a script for me. Check out `man flock` you might learn something new yourself!

The best AI users are the ones that understand what the model is good at and how to route around what it's not good at. Anthropic is reducing friction. But the gap between average and elite usage? That's on you. Start experimenting. Push the boundaries.

I've canceled my GitHub Copilot subscription and I'm missing it already. I'll use this occasion to experiment with other LLMs and harnesses and report back. For now, I've added some credits to OpenAI and I'm installing Pi (pi.dev).

Bild

I've been speaking English 8+ hours a day every working day for the past 11 years, and I'm at a point where I get annoyed when I'm not able to express a concept because I lack vocabulary. I'm thinking in English and I know what I want to say, but the word is not there 🤬.

I've been working out a lot more over the last couple of years! I'm lifting consistently and my pull-ups have improved, but one-arm hangs are still pretty hard for me. That's the next skill I'm tackling 🤞

The Anthropic trainer is bragging that they have a very high usage limit and can run at 50 concurrency, while suggesting that YOU run at 3... or more realistically, 2 or 1 or you'll be rate limited. 🤣

Bild

TIL: on the "Building with Claude API" course on Anthropic’s Skilljar I learned how to use assistant messages and stop_sequences to extract structured data. TL;DR: Inject an assistant message with "```python" to force the model down a path and set a stop sequence to "```" to extract the code.