Software Engineer by day, Software Engineer by night.
Berlin, Germany
Joined May 2020
- Tweets29
- Following66
- Followers5
- Likes290
João retweeted
This is absolutely true.
AI is incredible—like breathtakingly, jaw-droppingly incredible—at building apps.
But it's total ass at writing. And hasn't moved an inch here in the last 2 years.
Jesse Eisenberg says he sometimes "experiments with AI" to see if it's good at writing:
"I’m totally paranoid. I sometimes experiment with AI to see whether it’s good at writing. And it’s still not. I’m paranoid, so I check and it still isn’t. Thank God. I say, 'Write a scene like this.' I’m just testing it to make sure it’s bad, and it’s still bad. So I’m relieved, but, like, how far away is it from being good? How far are we from a kid putting on a pair of goggles and saying, 'I want to watch a movie about me where I rob a bank and then kiss the girl I’m in love with at school. Here are pictures of her, here are pictures of me' and then they press play? I’m paranoid about that."
variety.com/2026/film/awards…
João retweeted
I used to be a language lawyer. In the late 80s I _studied_ the ARM (Annotated Reference Manual for C++). I knew the language cold, inside and out. I could make it do backflips. I shipped a lot of code using it. A LOT.
I’ve attacked other languages with a similar intensity and depth. Until recently, that was simply the only way to become adept at building software systems.
I now consider the deep understanding of a language to be a supreme waste of time and effort. I don’t want to learn Rust, or Golang, or Typescript, or any of the thousands of other languages out there. I’ll leave that to my agents.
What I need now is to understand the high level gestalt of the system. I want to see the systems and subsystems. I don’t need to inspect the cogs. If I truly need to understand a detail at the cog level, I’ll interrogate my agents. I’m not going to pretend to be a language expert anymore.
João retweeted
Follow this guy. He’s about to spend an entire year without looking at a screen, while tracking his biomarkers and overall health to see what changes.
It feels like removing a dagger from my leg.
Headache. Dysphoria. Flatness. Hearing phantom iPhone dings in my sleep.
Days 1-3 were more of a rush. I felt light and free.
On day 4, the withdrawals began. I've never had to kick something like this before.
Screens are a legit addiction.
I'm curious to see if the month 6 PET scan shows enhanced metabolism in regions that improve in drug addicts who quit like the thalamus, frontal cortex, and striatum (reward circuit).
There are no PET studies of people quitting screens, so Dr. Eber and I are breaking new ground.
Image attached is baseline PET (F18 tracer). That's one of a few scans I'm tracking.
Day 6/365
Sent from my screenless phone
João retweeted
I love the innocence of Claude when estimates a task to take "roughly 2-3 days" and then proceeds to do it in 2 minutes.
João retweeted
The fact that Germany’s largest ISP considers 10 Mbps a speed worth discussing in 2026 is a fascinating glimpse into the technological ambitions of Europe's largest economy.
João retweeted
hot take: you don't need a software factory, and i don't love the term. it's just marketing
all you need is good software engineering and investing your tokens in the right places. the more deterministic the better (and token efficient)
João retweeted
The bug is like a joke.
On Alpine Linux, type 257 to 1024 characters of anything and you get root. 💀
‼️ BREAKING: OpenSSH has rushed out a fix for a critical flaw that lets attackers log into some Linux servers without the right password.
The team learned of the bug a day earlier. OpenSSH's Damien Miller says long embargoes now mostly help attackers, because AI agents rediscover bugs, sometimes within hours.
It hits Linux systems on the musl C library with PAM off, such as Alpine Linux. Miller told admins to disable password logins or SSH itself "IMMEDIATELY".
Version 10.6p2 works around it. Systems on glibc or PAM are not affected, and Miller says the root cause sits outside OpenSSH.
Readers added context they thought people might want to know
Exploitation requires specific conditions, not just typing a long string anywhere on Alpine: sshd built without PAM, password auth enabled, a target account with a '*' password field, and another local account with $5$/$6$ hashes. Root login also must be explicitly permitted. alpinelinux.org/posts/2026-11-… openssh.org/releasenotes.h…
João retweeted
Today we're launching Raylight 2.0.
Everyone's making motion graphics with Claude right now. Then they need to change one thing and it's a re-prompt and a re-render.
We gave Claude a real canvas. In 3D.
> Describe the video. It lands as layers and animation blocks, not a black box render
> Don't like a move? Drag it. Or ask again. Same timeline, same frame
> Screenshots become 3D product shots with real lights and shadows. Bring you own 3D models.
> Works in Claude, Cursor, or the app. No Team plan required
Every frame answers to you and your agents.
Comment "RAY" and I'll send free month of Pro 👇
João retweeted
Crazy how Large Language Models have gotten good at pretty much everything other than generating text
João retweeted
Holy sh*t, Opus 5.5 put 100 effects on one character in 30 seconds
Pause anywhere. Every frame is a different effect
Motion designers charge $500 per effect. That's $50,000 of work, from code
Drop the PDF below into Opus 5.5 and see what it makes
Then read the full guide on motion design in the article below ↓
João retweeted
You can learn a lot about which companies have decent engineering practices by how quickly they got their apps updated for the iPhone Duo
João retweeted
Introducing the fully open-source alternative to Grok Bot, Dots and Duet.
Use Rerun with your ChatGPT and Claude subscriptions, or any other model.
Unlimited seats for FREE!
The mobile app will be available this week ⏳
Use for free : rerun.build
João retweeted
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
João retweeted
This is f*king insane. SpaceXAI engineer Lauren Tan:
"You spend the tokens to set up a codebase so that even the dumbest agents can do a good job."
In this 60-minute workshop, she shows why the setup matters before you scale the team
Job → Rules → Skills → Verification → 6 Bots → 200+ Agents*
Most people add another bot when the first one fails. That just multiplies the mistakes
Spend the tokens on rules and checks once, instead of paying for the same failure every run
Worth more than most $1,000 agent engineering courses
Bookmark and watch it today
Then read the full setup: "Grok Bot: 6 Bots, 200+ Agents, a Smaller Bill" ↓
João retweeted
The coding interview is dead, at least for my team at Anthropic. I've seen enough to know that it no longer provides signal. I removed it from loops this month.
I have a spicy, optimistic take - I think there's an opportunity for junior engineers: felixrieseberg.com/hunger-be…
João retweeted
Guten Morgen Berlin und Brandenburg auch dieses Wochenende kommt ihr in den Genuss des Mehrwertwochenendes... Schaut hier bvg.de/endlichraus .
Weiterhin im Raum Berlin kann es aufgrund von Demonstrationen zu Einschränkungen kommen. Ich halte euch auf den laufenden.
