Stephan Janssen

#Devoxx Belgium 2025 : Java + AI + Agents

I'm convinced this is the biggest European AI conference using Java 🔥
#Devoxx #AI #Java #Agents

Already available on Ollama @ ollama.com/library/gpt-...

OpenAI releases "open source" models : GPT-OSS-20B & 120B models with impressive stats 🔥👀 Full details @ openai.com/index/introd...

Expect more details about our brand new Devoxx scheduling web app in the very near future 👀

We could have easily organised a 3 week conference with all the quality talks we received. We will accept a few more talks in the upcoming days and unfortunately also refuse the remaining talks before registration opens on August 18th 🔥

Bust day: Accepted the majority of the talks for #Devoxx Belgium 2025 & also released a brand new scheduling web app (while fixing remaining 🪲)
A BIG THANK YOU to the program committee and community reviewers for the many thousands of reviews!

Hello beautiful... 😍 #Ollama #UI

IMO I would say 70% of the rejected talks fall in the category: "great proposal, but we don't have any slots left"

Not when you’re reviewing 200 talks in a single track! 😅 But the best talks naturally rise to the top, thanks to many program and community ratings which is based on their judgement and experience 🧠✨

Seeing the level of quality in the content & speakers submitted for #Devoxx Belgium, I genuinely wish I could experience it all as an attendee instead of being the #HeadLessChicken organiser! 🐔

For example, we received nearly 200 proposals for the GenAI track, but only have 20 slots available. It is incredibly competitive, and while ratings help with the initial shortlist, there is often much more nuance behind the final selection.

The real value comes during our in-person selection meetings. That is where we debate, challenge each other, and really dig into what makes a talk stand out. Unfortunately, none of that is recorded or written down.

99% of the Devoxx Belgium talk reviews just get a star rating, with hardly any written comments. And when we do leave a note, it is often something vague like “Product pitch?” which is not very helpful.

When you say "comments" you mean the Ratings? Because we do have comments with the speaker which are ofc public to the involved speakers.

NotebookLM just transformed my AI Augmented Programming YouTube talk from earlier this year into a slick executive summary 🔥👀 (Original was 8' long, had to trim it to 3' for Bluesky post) #Gemini #LLM #MultiModal

Nope, because some ppl really leave talks in draft... but looks like your two talks did get ratings. #ThanksForSubmitting

Sub agents - Anthropic Create and use specialized AI sub agents in Claude Code for task-specific workflows and improved context management.

We now have custom sub-agents in #Claude Code 🤗
Just type /agents in the CLI to configure them!
docs.anthropic.com/en/docs/clau...

Deep into reviewing #Devoxx Belgium CFP proposals, some are absolute gems, others… not so obvious 🤩
BTW, if you're a Community Reviewer, now’s the perfect time to jump in! Our first selection meeting is mid next week! #ThankYou ❤️

TBH I'm not doing pure TDD. Because its a website, I check visually if everything works and then generate tests to make sure nothing gets broken by future LLM code changes.

That’s why I’m excited to see new products like Amazon’s Kiro pushing the 'spec-to-code' paradigm, along with many others that are sure to follow. I’ve invited several key players in this space to #Devoxx Belgium, more details coming soon!

This experiment actually gave me hope that we developers are far from being out of a job! We can become more productive, but we still need to keep our hands dirty with the (generated) code. Plus, design and feature documents (as well as UI mockups) will play an even more critical role going forward.

But also with the current state of LLM's, reviewing the generated code is as important as guiding the AI with different
(correct) approaches!

So I decided to restart but now use Angular (which I know a bit better) and use clear specs & UI mocks which were taken and destilled from the previous POC. Spec Driven Development (SDD + TDD) is clearly the way to go when using AI Coding Assistants like Claude Code, Gemini or Codex.

I basically needed to step in and look at the code. After reviewing, it turned out to be spaghetti layers of unnecessary caching and indirections 🫣

🧵 As an (Gen)AI experiment, I kicked off a new project in React (which I had zero experience with) using Vibe Coding. The first few days were incredibly productive, and within a week I had 90% of the project done. But for that last 10%, Claude Sonnet 4 just couldn’t help me anymore...

The #Devoxx CFP is now closed, THANK YOU for the overwhelming number of submissions! With over 810 proposals to review, our team will need some time to carefully go through each one. We’ll get back to you in early August 🌟❤️🥳

Had some fun this morning adding a #Devoxx speaker moderation tool for the CFP.DEV face recognition service. Now admins can easily remove false positives 😎📸
Excited to roll this out for Devoxx Belgium 2025!
On my todo list: Allow speakers to select their own favs 🚀

Any takers? #Docker #Devoxx #Belgium

TBH I hope we reach 800...

Current #Devoxx Belgium CFP status 🔥🔥
Still 1 day and 7 hours to go #procrastination