ElixirConf 2026

ElixirConf 2026 was a blast.

It was my first ElixirConf since 2023, because I try my very best to avoid going to Florida. For folks that don't already live in an airport hub city, going to Orlando can involve a layover and more expensive plane tickets. Also, everything in Orlando is more expensive than it should be for the quality level, because of the theme park tourism aspect of the city. If you aren't there to go to a theme park, you are paying too much for things that are just kind of meh. I even like theme parks, but the timing never worked out to add a family vacation before or after the conference. What am I going to do, leave my family at home and go to DisneyWorld without them?

Chicago was great. It was easy travel-- 2.5 hours direct flight from my airport. Hotels were cheap enough. The food was incredible. I ate at Mr. Beef, which is the place that the restaurant in The Bear was based on. I had deep dish pizza, which is not normally my thing, but when in Chicago, right? I ate great Japanese food that was not just sushi or ramen, at a very unassuming spot one block from my hotel. The transit was solid, which is nice because I hate renting cars and driving in unfamiliar cities. I made the mistake of booking a very tight itinerary. I really should've given myself a day or two around the conference to enjoy Chicago a bit more.

Anyway, it was great to catch up with people from the community who I had not seen in years. It sounds like a lot of people had a similar story of skipping the Florida years. I got to hear what everyone had been up to for the last few years, and what they were excited about right now. The talks were also great.

It gave me a much needed burst of inspiration. I was honestly feeling pretty burned-out and aimless in the months leading up to the conference. I knew I wanted to do something, but I could never quite identify the thing that was going to give me energy and keep me tinkering. This is unusual for me-- I usually have a whole list of experiments and side projects queued up, and I jump from thing to thing to keep my interest up. 90% of them never even become something I release, but I always learn something or at least keep myself entertained. I've been this way for over 30 years now. So not having anything I want to be working on is a big red flag. But after the conference, I've had a total turn around in energy. I left with a long list of techniques, packages, and side projects to try. I started sending messages to my favorite co-conspirators and sounding boards. I'm excited!

Favorite Talks

I wanted to list my short list of favorite talks here. I don't think the conference team has published the videos to the public yet, but they were available basically immediately on the conference page in Swapcard for those who had tickets. You can watch the ElixirConf YouTube Channel channel for when the video become public. I think the plan is for them to go up in about three months.

Disclaimer: I have not caught up on every session that I missed yet, so there might be future favorites not represented here simply because I have not made it to them yet.

Set-Theoretic Types from Scratch - José Valim

This one was all about explaining the Elixir gradual type system by implementing a minimal toy version of it in code. My formal CS training is twenty years behind me, so while I do read academic papers from time to time, that's not the best way for me to absorb information. I like reading code, because I read and write code every day. I finally understand what this gradual type system is about. The timing is perfect because the type system "officially" landed in Elixir 1.20 which was recently released.

A Dev Environment for Every Branch - Jason Axelson

A very impressive demo of Jason's fast-worktree tools, which is his setup for quickly bouncing around between different branches, features, and streams of work. The trouble with the basic built-in git worktrees is that they start from a "blank" state and you end up having to install dependencies, initialize your database, compile the project, etc. fast-worktree handles instead copies files and databases so that you can pick up development immmediately. It even handles handing out TCP ports so that you can have a server running for each worktree without conflicts. Each one of these features he went out of his way to make a fast as possible, using things like copy-on-write file operations and postgres template databases.

It sounds like Jason switches contexts a lot during his dev flow. I think doing that would drive me a little batty. I'm more of a "hyperfocus on one thing, never lose the context" person. Nonetheless, this level of tool-sharpening is inspirational.

Rethinking Modern Education: Building a Local-First AI "Forge" with Phoenix, Nx, and Bumblebee - Chris Bailey and Kathy Keppel

These two are working on a product that will address one of the big challenges in schooling in 2026. If you send homework home in 2026, a lot of students are going to use an LLM to do it for them. If they do that, they won't actually learn the material, and also the teacher loses the ability to check their comprehension and tailor their lessons around their students' needs. Take "homework" out of the home. So, they are building a box that makes it easier for teachers to serve up little comprehension check exercises in the classroom. And they are doing it with a very high level of respect for student privacy. No student data ever leaves the classroom. This approach (plus the usual very tight budgets of school systems) creates some very interest engineering requirements.

And they demoed their system-- including on-device AI model-- running on an Android gaming handheld. Very cool.

The Agentic Experience Crisis: How Juniors Become Seniors in 2026 - Bruce Tate

Bruce is just generally such a gift to our community. This is also something I've been thinking about a lot lately. I think anybody who cares about developing people in the software industry has been thinking about it. Bruce has been working with tons of junior engineers and he's landed on some great principles that early-career engineers can use to grow and learn, even at peak LLM. Actually, I think the principles apply at all levels of experience, it's just that a lot of more experienced engineers have experience which puts them a few steps ahead in terms of applying those principles. They might not even have to thing about it.

I will be doing my best to weave this into my own work with people who are new to our field and/or are having a hard time adjusting to how LLMs have changed things.

What Comes Next - Chris McCord

Speaking of LLMs, Chris is all in on LLMs, and that's mostly what this talk is about. I have a lot of reservations about LLMs for a lot of reasons. I won't detail that here, because well, you've certainly heard all about it on every corner of the internet for years now.

But Chris is smart and he has good instincts. That means I'm willing to hear him out, even if I disagree with him on a given subject. It seems like he's having really good results with an LLM-heavy dev workflow. Some of that is probably that if you have those good instincts, generating good code becomes much easier. It's interesting to see where he's putting tooling and safeguards and where he isn't.

One comment in particular has had me thinking lately. In the talk, he said something about how branching, creating commits, pushing to GitHub, etc is all starting to feel like a lot of ceremony. He made a half-joking, half-completely-serious comment about bringing back the old PHP days of just FTPing files to production (or maybe even editing them right there). I've been feeling it for a few years now too-- even pre-LLM. Our tools, languages, and frameworks have gotten so great that often I can fix a bug in five minutes. And then I spend ten or twenty minutes making a branch, and a commit, and pushing to GitHub, and waiting for CI/CD to run, waiting for containers to build, etc. It's so wasteful. I'm incubating a lot of ideas right now about how to fix that-- you know-- losing track of your changes because you make them directly on the server using ssh and nano.

Terminal UIs? That's a Breeze - Gary Rennie

This one feels like a magic show.

Breeze is a framework in Elixir for making TUIs. I think it has been around for a bit in a development state, but what Gary demonstrated in this talk is that it is ready for prime time. And very, very cool. He presented his slides in a TUI presentation tool built in Breeze. He showed off all the building blocks. He showed the application live-reload as he changed the code. He showed off TUI games. He showed off the tooling that lets you web-inspector-style debug the running application in another TUI!

I went home and wrote a TUI manager for todo.txt files in a couple of hours.