Thomas Bereknyei

[HSVSphere @HSVSphere]

I hate GHA so fucking much. Worst fucking build system ever. I will hack nix remote builders on all 4 runners in & do it, because its own dependency model sucks ass.

Also it can't cache anything, which is hilarious for something that should be a build system.

Hundred lines of yaml for something that should be expressed in 3-5 lines. Not joking, it's that bad. Whoever created this crap should be fired.

[Thomas Bereknyei @bereknyei]

@esrtweet A consequence of being fanatical about ensuring your system can support easy rollback, is that you end up using it very rarely. In practice, we see people updating the declarative specification and rolling forward, due to the confidence in moving in either direction. No fear.

[Thomas Bereknyei @bereknyei]

@esrtweet While there have been some issues, it is by no means a lost cause - still worth trying it out. I've spent a fair amount of effort to help grow the community and adoption, got elected twice to the Steering Committee, and I am optimistic about Nix/NixOS's future.

[Sam Rose @SamRose]

https://t.co/sLZe4HgQTG

[Farid Zakaria @fmzakari]

LFGO !
Sign up and come to the first North American Sprint!

We need sponsors too 🙏🤙🏄‍♂️

[Thomas Bereknyei @bereknyei]

@nzupan Any summary of interesting projects and ways others can get involved?

[Jonas Chevalier 🦙 @zimbatm]

Does Nix feel faster yet?

[Jonas Chevalier 🦙 @zimbatm]

Last week I built a little tool to find hot paths in Nix evaluations, classify and find simplifications trough code mutation.

Now nixpkgs eval is 30-50% faster on cross-compilation targets :-D

Let me know if you have slow code.

https://t.co/MzG7KqdGfr

[Thomas Bereknyei @bereknyei]

@leothrix For maximal dev joy? Htmx. Sprinkle in some hyperscript if really needed.

[adafruit industries @adafruit]

Aegis: a fully open-source FPGA, from the silicon up

https://t.co/8MiDBkSsy3

[Thomas Bereknyei @bereknyei]

@HSVSphere "install" is an overloaded term for software. Everyone means something a bit different by the word and this results in much confusion.

What can we do with Nix make trying out new stuff easier? I've been working on making binary-only distribution easier to alleviate eval issues.

[Thomas Bereknyei @bereknyei]

@joschelboschel You have a variant of this as well?

[Thomas Bereknyei @bereknyei]

@joschelboschel You explain the overall libnix goals and motivations. (Talk to Ericson, he has very similar goals.) Then we get to the cargo integration and it is hard to see what progress is, or what outcome people would or could help with. There isnt a clear call-to-action. What can they do?

[Rok Garbas @garbas]

What do chefs and Nix engineers have in common?

We both obsess over a clean kitchen.

A good chef doesn't prevent burnt steaks by controlling every second. They keep the kitchen clean. Ingredients fresh, knives sharp, everything in its place. The creative parts figure themselves out.

I've been doing reproducible builds for 20 years. Locked deps, pinned environments, deterministic tooling. Got into it because "works on my machine" was driving me crazy.

Then AI showed up and something funny happened. All that plumbing turned out to be exactly what AI needed too. You can't make AI deterministic. But if everything around it is locked down, you actually know when something goes wrong whether it's the AI or your environment playing tricks.

20 years of clean kitchens and I was accidentally building the best AI strategy the whole time.

https://t.co/yrJlaXqyyk

[Thomas Bereknyei @bereknyei]

@garbas Agents are comfortable with being explicit. Rather than relying on quiet/short commands that have the right default behavior, instead an agent can use long flag names and specify them explicitly, making the intent clear. We no longer optimize for fewer keystrokes. DX vs AX.

[Domen Kožar @domenkozar]

American friends, we're going to organize a Nix sprint in Mexico over the summer.

Rent a villa near a surf beach, do a bunch of brainstorming, hacking and enjoying tacos.

Who's in?

[Rok Garbas @garbas]

PlanetNix 2026 is behind us. Almost double the size from last year and I'm still catching up on sleep.

Helped organize and gave a workshop on getting Nix into teams where most people won't ever learn Nix. 40+ people in the room, which was great, but the thing I keep thinking about is something else.

Even the most hardcore Nix people got it. Like, the people who live and breathe flakes every day looked at the problem and said yeah, we can't expect everyone to learn this. And that's fine. Did not expect that from this crowd.

Anthropic, Nvidia, Spotify - all there, all using Nix. Not trying it out. Actually using it. With AI moving as fast as it is right now, reproducible infrastructure underneath stops being a nice-to-have pretty quickly.

Nix + Kubernetes was a hot topic. @lovesegfault from Anthropic showed rio-build (distributed Nix builds on K8s) and @devusb talked about Flox on Kubernetes. Different approaches, same direction. Talking to people about where this is going... can't wait to see what comes out of it.

Event is done but we're far from finished. Videos, interviews, all of it still coming.

Were you there?

@planetnix @floxdevelopment

[Thomas Bereknyei @bereknyei]

@domenkozar That...... is cool.