Norberto Lopes

There's so much doom and gloom about AI and software engineering, but honestly, I’ve never felt a bigger pull to be an IC again.

I kinda just want to run Rex server on compute engine (or whatever flavour you prefer) and then my laptop is basically a browser and Rex client.

Whoever at Google/Android that designed the new card picker in the Google Wallet: great work. Finally picking a different card is a great experience.

TIL: yaml parses tilde as null. I never quite realised this. This is madness. mounts: - location: ~ mounts: - ~: location All valid but that tilde is a null. And yes, guess who tried to use ~ for the home folder 😂

I ended up not resisting writing about this GitHub bug because it drives me slightly insane: nlopes.dev/writing/i-ne...

incident.io/blog/how-sav... wrote about Postgres internals for a long while

A prompt I've been enjoying using is: Go over the approach we wrote down for building an extensions system, and use category theory to review the plan. I think some of the concepts are immediately applicable. All of a sudden, the code gets better. 🤷 Wild times.

Ctrl-G is such an upgrade inside Codex or Claude Code. I use it occasionally when I need to write a bunch and I should use it way more.

How I *don't* like to spend my Sunday: sorting out security advisories.

"Remember to use the cli tooling instead of writing a python script"... This is getting tiresome. All newer models seem to reach for python a lot more than their counterparts from a few months ago. At least that's how it feels. Sol and Astra in particular are so intent on python scripting!

I love when this kind of technological convergence kicks off. Yes, we had internet on planes for a while but only in the past 5y did it become quite good (same goes for underground trains etc). And now, we also have high quality LLMs. Each one is a marvel, but together? Mind-blowing.

Book 4 of Saga just arrived. So excited about this.

One aspect that I'm proud of is giving feedback to candidates if they get rejected at the end of the interview process. I posted about this on LinkedIn and got a bunch of messages asking about how I approach these conversations so decided to write about it.

Wrote about how MCPs and skills are implemented in agent harnesses and how skills can help consolidate many specialised agents into one. This is a write-up of a “skills and MCPs 101” we’ve had several times at incident while building this into our harness. blog.lawrencejones.dev/demystifying...

LLMs are so good at writing fixtures. Makes a lot of the boring work a breeze.

Working with AI is either "Just do it, no need to always ask me" or "That is so stupid, please always ask me before making changes"

I keep finding stuff I hadn't clocked in cargo nextest. I said it before but I'll say it again: I wish all software was this good. Big fan. I can't remember the last project where I didn't use cargo nextest instead of cargo test. (except for --doc)

Who called it cargo clippy instead of rustycop?

This blog post from @incident.io absolutely rips. - Great technical detail - Solving a real problem - Leveraging open-source projects I wasn't familiar with but am now super curious Excellent writing and great to promote the rigour they put into system fundamentals ❤️ incident.io/blog/we-turn...

There are two things that I find are not great UX in jujutsu: jj bookmark move --to @- main # git just moving main makes UX overall better jj file search --pattern "<pattern>" # git grep just clicks more for me

Read receipts are "opened" receipts and they mean nothing at this point. I guarantee you I've "opened" plenty of emails because I'm navigating the inbox, read none of them, prioritised based on Subject and From and then a month later sent a "sorry, not sure how I missed this" because I did.

I would love 10 minutes with whoever leads the customer support team for the Google Store. I think 10 minutes of showing them how their system is actively anti customer would do more good than all the metrics/SLAs they might have internally.

Niiiice. Been looking out for this for a while.

This is cute but also... you ok github?

This visual bug has to be the most annoying bug in @github.com that I am faced with every single day. Unless this is on purpose but I don't quite get the point.

One of my favourite things about the Google keyboard/android keyboard is that if I have text selected and press shift repeatedly, it rotates through types of capitalisation. It makes it so easy to fix things. I wish other OSs had this by default.

Just passed the 2k test mark for the AsciiDoc toolkit test suite in github.com/nlopes/acdc. cargo-nextest is one of my favourite tools in rust right now. It just works. Every time. Doesn't get in the way, it's fast, and has great UX. For me, github.com/nextest-rs/n... resolved makes it perfect.

We're at that stage I have to politely say "you're holding it wrong" whenever someone has a strong take on "AI is useless and garbage and doesn't help me". It's no longer a toy. I will respect folks that have ethical objections to it but I object to the "it's useless".

GPT 5.6 sol is very very good. I'm enjoying a fair amount how it goes about things and the results. Opus 4.8 can still beat it in some stuff but I think this is the first time since late last year that I feel OpenAI truly catching up and perhaps overtaking.

Wrote about our experiences at @incident.io of migrating our codebase to routing most DB queries to the read replica and some useful patterns that came out of it, including LSN checking! incident.io/blog/dont-ad...

The things one finds doing a deep cleaning. Farewell dear hp-ux and Veritas. You won't be missed.

I've updated the AsciiDoc lsp github.com/nlopes/acdc/... to detect: - if a conditional is inactive (does not match the condition) - if a conditional doesn't close/is invalid - if a conditional is valid and active And you can also configure which "backend" it should use to parse your document as.

This was brewing in my mind for some time now. Finally put "pen to paper". nlopes.dev/writing/grow...

> I just wanted to express my deepest appreciation for the work you are doing with acdc. At the end of the day this is why I keep doing OSS.

Opus 4.8 on the left, my own on the right.

It still boggles my mind that I can write this at 30000 ft like it's nothing.

I'm adding asciicast support to acdc... which meant creating this today: github.com/nlopes/ascii... I'd rather have this from the asciinema folks but they don't expose the format as a library :(

OMG yes! I didn't realise jj allowed for "private" commits. That's awesome.

99.99% isn't a faster version of 99.95%. It's a different problem entirely. 🔥 Our VP of Engineering @nlopes.dev writes on what autonomous recovery, dependency ceilings, and CI/CD speed actually look like at four nines: incident.io/blog/humans-arent-fast-enough-for-4-nines

With a day out of work to relax, I never thought I'd end learning the spec for the various asciicast version formats. But here we are.

Very much worth a read.

Yeah, rust error messages are in a different category altogether.

Michael if it's any consolation, I've just bought Ouros and will be recommending it to everyone. I wish more devs made games like this one, great work!

Google customers: we really need good customer support. Google: here are our new icons. Rejoice.

You're giving too much credit to LLMs. You're not giving enough credit to LLMs. Both can be true. I don't understand what's so complicated about this. And the argument that using LLMs is *only* bad and destructive no longer flies. Even if there's a bubble. Even if said bubble bursts.

Wrote a bunch of words on building acdc: nlopes.dev/writing/buil...