Roberto Empijei Clapis

Roberto Empijei Clapis

@empijei.bsky.social

Security Toolsmith, recovering xoogler. Posts mostly about Go, banter, web development, security and cooking. https://empijei.science

485 followers95 following144 posts

I've been saying some variation of this for the past year or so. Sadly, I will probably have to keep doing it for the next 5.

Stefano Zanero@raistolo.bsky.social

It’s gonna be fun to watch the AI tech bros relearn everything about cybersecurity ignoring all of history. And by fun I mean exhausting.

Big shout out to endeavour-os. I haven't used arch for almost 10 years, and the experience of installing, maintaining and using it has improved so much thanks to them that I was blown away. It's really the year of Linux on the desktop. endeavouros.com

endeavouros.comEndeavourOS — A terminal-centric Arch-based Linux distribution with a vibrant and friendly community at its core.A terminal-centric Arch-based Linux distribution with a vibrant and friendly community at its core.

Can't wait to watch this.

Alex Rios@alexrios.me

The world tour continues relentlessly. I keep submitting talks and somehow people keep accepting them. Next stop: GoLab 2026, Bologna 🇮🇹 Talking about Green Tea GC, Go’s new garbage collector, and why “scan spans, not objects” turns out to be a much bigger idea than it first appears.

I've been saying this since 2024. I'm waiting for a major, catastrophic vulnerability to shake the world due to this.

Tanya Janca | SheHacksPurple@shehackspurple.bsky.social

Basically: the robots are helping us write code now, which is great, except when the robots confidently hand us a security problem wearing a tiny little party hat. 🎉 If you build software, secure software, lead AppSec, or are currently side-eyeing your AI assistant, this episode is for you. 3/4

Roberto Empijei Clapis reposted

It's cringe to ask "what would Trump be doing differently if he were a Kremlin asset?" And yet we must.

David Roberts@volts.wtf

Putin is paying people to disrupt renewable energy projects & drive polarization around net-zero goals in Europe. Luckily, I'm sure he would never pull anything like that in the US, so we're probably fine.

Periodic reminder that modern computers are fast enough to do substantially anything without having to make the human wait, in any language. If your CLI is slow you don't need to change language (e.g. py->rs), but you should rethink how you design your code.

literally every person in the world will disagree with me on this but: codex is the fastest, most reliable agent CLI out there. Specifically the CLI itself. It's written in Rust and it shows. It doesn't have the 2-second stutter on startup and it's not the buggy, flashy mess that Claude Code is.

I'm officially looking around, if anybody needs a Go software engineer with a very strong background in web security lmk. I also know a fair share of frontend dev and was a SWE/SE for Google and Microsoft. I'm looking for security-sensitive dev projects and security reviews.

Roberto Empijei Clapis reposted

MacOS: I see you've set Finder to appear on 'all desktops' me: yes MacOS: So when you cmd-tab to Finder, which desktop should I switch to? me: well you shouldn't because… MacOS: So I can switch to a random desktop? YAY!

Roberto Empijei Clapis reposted

MacOS: When I restart, would you like me to re-open all these apps? me: no MacOS: hmm well that's a shame because I am going to do that. me: ok MacOS: Should I open them all maximised on the same desktop? me: no MacOS: hmm well…

Will LLMs become a fundamental tool in our box? Sure. Will they replace us? No. Linters may become very powerful Reviews may become more precise Autocomplete may be more useful Testing may become more thorough and widespread But only if we collectively take our heads out of our a###s soon

After about 12y of Vim and 1.5y of vscode I tried @zed.dev. I have to say, the speed is impressive, there still are some minimal graphic quirks, but I don't care about those. It feels amazing to use it, but I can't switch to it yet 🧵

View full thread

3) No way to run tests on save 4) No way to display test coverage gutters 5) Running a specific test requires two clicks. This means that I write code at insane speed to then hit a wall when I touch the tests.

I'll keep an eye on the project because, again, it's a very good IDE and addresses a lot of the issues I have with vim and vscode, but testing support is what really prevents me (and my company) from switching to it right now.

After about 12y of Vim and 1.5y of vscode I tried @zed.dev. I have to say, the speed is impressive, there still are some minimal graphic quirks, but I don't care about those. It feels amazing to use it, but I can't switch to it yet 🧵

The lack of testing support is what really is a dealbreaker for me, all the other minor inconveniences are stuff I can live with in order to get the speed. 1) No quick way to switch from a file to its test file. 2) No easy way to run a package or project tests

I am using a MacBook lately. It was a gift, so I took the chance to finally try it. I have to say: it's the best hardware AND the worst software I've ever had on a laptop. The operating system is barely usable and shaves the bar for "functional" VERY closely. But the machine is a beast. 🤔

The AI harness engineering hype is basically just people finding out that with leaner design, good encapsulation, simpler code, improved tools, working CI/CD and precise documentation one can be more effective at writing correct code.

View full thread

Not only this would have caused issues because some of the code was actually used (Claude didn't detect where because it was across an FFI), but this threw away weeks of work honing those libraries.

When we later needed those functionalities the code had evolved too much, so it was deemed better to rewrite instead of recover. Needless to say, the slop didn't quite replace the code that was originally there.

AIs made code very cheap, so people (especially inexperienced devs and execs) don't respect it. I've seen an engineer delete big chunks of libraries just because they didn't understand them, didn't think they were used, or rewrite them in a way that AI understood better.

Week one of sourdough, results are staggeringly similar to those of the bread maker I own. I accidentally started this experiment as a game for my daughter but it's already surpassed my expectations.

The value of the code isn't just the code itself, but it's also the mental model the developers build while working on it. Taking that away with AI creates maintainers that can't maintain the code, one little change at a time.