Highlights from "J on the Beach"

After hearing about what a great experience people have had at this conference over the years by speakers and attendees alike, I finally made it out to “J on the Beach” in Málaga, Spain last week.

This year’s edition was part of “Yay-Yay Conf”, which brought what were 3 separate conferences (J on the Beach, Lambda.World and WeyWeyWeb) under one roof, making the event multi-track.

It was incredible. It’s been a while since I attended a community conference and, besides learning about what cool shit speakers and attendees alike are doing in their work, I enjoyed hearing what people are looking at and experimenting with in terms of AI adoption such that they’re getting the benefits of AI while mitigating the risk that they’ll be dealing with a mountain of technical debt in a few years.

Some highlights in terms of learnings that I wanted to share:

The use of AI to instantly generate UI based on needs and context

I wasn’t able to make it out to this talk due to a scheduling conflict, but I got to chat with the speakers on a bus ride, and this sounded like such an interesting use case for AI in the area of accessibility.

It’s top on my list of videos that I want to watch when they are released by the conference organizers.

Formal verification is having a moment

The Lean programming language, which allows people to formally verify that their code matches the specifications that were defined for it, came up several times during the conference.

There were two dedicated talks on the topic, and was also called out in Erik Meijer’s opening keynote, in which he argued that learning formal verification was going to be an important skill in the coming years in software engineering.

Chatting with other speakers - Quint also came up as something that people have been experimenting with in this space with some success in their work.

Putting aside the discussion of the quality and maintainability of the software’s architecture when generated by AI for a moment, the argument for formal verification goes that, if programming languages become more of an implementation detail because AI is generating the code, then it becomes of paramount importance that the code that is generated complies with the project specification (i.e. the tests).

From these various discussions with folks at the conference, I plan on learning and experimenting with Quint in the coming months in my side project, as I think it will compliment my use of Linked Intent Development nicely.

Local development AI sandboxing

I’ve heard a lot about sandboxing AI agents in remote environments (Amp’s Orbs, Cloudflare Sandboxes, etc.) but not as much about sandboxing them in a local environment, although I know there’s been activity in this space for a few years now.

I ended up learning in more depth about what the local environment agent sandboxing landscape looks like through a workshop that took me through configuring foldyard.

It was pretty straightforward to set up, and it feels nice to know that if my agent ever goes haywire and thinks that nuking the / directory is a good idea, that it won’t completely bork my computer.