Today was the last day of the YOW conference in Sydney. I’ll write some thoughts while the whole experience is still fresh in my mind.

My first conference

It was the first time I attended a conference. I had done one-day trainings previously through my company, but I hadn’t attended a conference until this week.

The “spirit” that is in the air at the conference is amazing. Lots of people who are normally coding for hours alone in front of a computer now are together talking, attending presentations, and eating together.

It is really inspiring and refreshing, I would say. It makes all those deadlines from the last few months feel more meaningful somehow (we clearly belong somewhere!).

Masterclass

I attended a one-day masterclass with Kent Beck called Tidy Together: A Software Design Masterclass.

Something interesting I learned was a graphical way of representing Design vs Features, and understanding how, in order to not end up in the “feature saw”, it is convenient to do a small refactoring before implementing each feature to provide a structure for future work.

I won’t try to describe the full day here. I just can say that sharing a day with Kent as a teacher was a real privilege. I learned and felt very inspired to continue learning.

Talks

I attended many interesting talks. I had to choose specific sessions to attend because there were always three running simultaneously.

I chose many AI-related sessions - I have been working with LLMs in the last few months, so I find the topic interesting. Some notes I took across different sessions:

  • Using LLMs for writing code is something now embraced by even the most experienced developers. There is a lot of excitement around how the models keep being improved.
  • I have the impression that fully vibe coding a project is something that is, at the moment, only working well when coding alone. The process is about designing the high-level way of building the project and advancing “until you can’t steer it in the right direction anymore”. Once you reach that point, it is necessary to restart the process in a new chat and apply the learnings in this new iteration for the architecture to be more correct from the get-go.
  • Techniques and libraries are available for developing agents in different languages. There are many different approaches and people who defend their specific way of designing these systems. I have the impression there is a tendency to make “too many calls to the LLM” to solve a problem that could be solved many times in a more deterministic way. It is not only important for the software to give the desired outputs.
  • There are ways of doing AI in a green way. Charles Humble gave a beautiful talk about this.
  • I have been working with vector databases, but it was refreshing to see a full introduction to the topic to clear doubts.

I did attend other talks (Crypto, Naming conventions, C4 Model, Sonic Pi, etc.). But AI, I think, was the predominant topic, and also a topic I needed to dive into on this occasion.

Conclusion

I loved the conference. I will definitely come back in the future.

This was a good opportunity to talk AI with people. Even though I’ve used AI to assist in my coding for a while, and even though I’ve already built RAG projects using vector stores and LLMs, I have never written anything in my blog about it - the space changes so rapidly that I didn’t want to think of writing, maybe? I’m not sure. But here it is, finally, some notes about AI in the blog :).