🟧Hugo H. Macedo 🟧 (@macedoh) header
🟧Hugo H. Macedo 🟧 (@macedoh) profile picture

AI amplifies judgment, doesn't replace it. 25 years building and growing tech business. Advisor, builder, pragmatic contrarian. Lisbon.

Portugal
Joined October 2007
Photo by 🟧Hugo H. Macedo 🟧 (@macedoh)Photo by 🟧Hugo H. Macedo 🟧 (@macedoh)Photo by 🟧Hugo H. Macedo 🟧 (@macedoh)Photo by 🟧Hugo H. Macedo 🟧 (@macedoh)Photo by 🟧Hugo H. Macedo 🟧 (@macedoh)Photo by 🟧Hugo H. Macedo 🟧 (@macedoh)Photo by 🟧Hugo H. Macedo 🟧 (@macedoh)Photo by 🟧Hugo H. Macedo 🟧 (@macedoh)Photo by 🟧Hugo H. Macedo 🟧 (@macedoh)
Two days ago: "I taught my OS to adapt to me." The real story is ~90 small fights. I'll post them one at a time: 🧨 Trigger: what annoyed me 🧠 Why: the real cause 🔧 Fix: what I changed ⏪ Undo: how to take it back Claude Code makes changes cheap. So everyone gets a way back.
For six months I adapted to a new OS. Then I taught it to adapt to me. Linux was always configurable. The barrier was knowing how. With Claude Code, it's one prompt away. 11 days: ~30 scripts, one git repo, a computer that works my way. What would you teach yours first?
21
The scariest AI failure isn't the crash. It's the failure that looks like success. A billing call times out and comes back empty. The agent writes "Payment cleared" and closes the ticket. No error. No alert. Just a confident lie in your system of record. Rule for anyone building agents: an empty answer is never a yes. Make failure loud. Where could your stack be failing quietly right now?
15
We obsess over how smart AI is, and not enough over how reliable it is. That's a killer for automation. Intelligence is the demo. Reliability is the product. That gap is the whole game here. The best agent in a CMU-led benchmark of a simulated software company completed about 30% of the tasks. It's what happens when "pretty good" has to survive a whole workflow. A step that's right 95% of the time, chained 10 times, ends up right about 60% of the time. Automation doesn't fail at the genius moment. It fails at the third handoff. Think goalkeeper, not striker. A striker can miss nine and win the match with the tenth. A keeper who makes one wonder save and lets in three soft goals loses it. Automation is the keeper's job. Credit where due: this clicked listening to Typeface CEO Diogo Almeida on @HarryStebbings' 20VC today. overcast.fm/+ABSzjpXHxu4 What would you refuse to fully automate until it's boringly reliable?
58
For six months I adapted to a new OS. Then I taught it to adapt to me. Linux was always configurable. The barrier was knowing how. With Claude Code, it's one prompt away. 11 days: ~30 scripts, one git repo, a computer that works my way. What would you teach yours first?
3
1
68
3/ ~30 small scripts, all in one git repo. That repo is the agent's memory and my undo button. The rule: inspect first, one change at a time, verify, log it with a way back.
1
5
4/ I keep the decisions. The agent proposes, tests and documents, and sometimes catches things I wouldn't have checked. Maybe that's where software is going: products you teach to fit how you work.
6
For six months I adapted to a new OS. Then I taught it to adapt to me. Linux was always configurable. The barrier was knowing how. With Claude Code, it's one prompt away. 11 days: ~30 scripts, one git repo, a computer that works my way. What would you teach yours first?
2
1
56
🟧Hugo H. Macedo 🟧 retweeted
This is one of the better litmus tests for whether you’re pushing your use of AI enough or not. Up until maybe six months ago probably the best you could do with AI was pull off something you already would produce, but just doing it a little faster. But now we’re entering a point where the output of working with AI can and should dramatically exceed what you would been able to do before, in a fraction of the time. The product prototype should not only be fully functional and interactive, but it should probably have meaningfully better features than it would have had before. The engineering architecture should be able to deliver performance gains that you wouldn’t have been able to go after otherwise. The research for a project should consume every single fact on the subject across every dimension of analysis possible. The business process you implement with AI should operate entirely differently and learn from all its past failures in a way no human could. We’re unquestionably entering a period where AI will be smarter than us on most axes. Perhaps not on every dimension that matters (intelligence is still very jagged), but at least on enough dimensions that when we combine it with what we’re good at, you get qualitatively different outcomes. So the simple test is: if you could have done what you’re doing before AI, you’re likely not pushing it enough.
If you’re going to make something with AI, make sure it’s 10,000x better than anything you would have made yourself. If not, you just come across as lazy not smart :)
172
78
30
1,032
113,465
Love the spot on analogy “Giving every employee AI — the best models, the most tokens — is not the same as becoming an AI-native company. It’s like giving every employee a Steinway and expecting a symphony. What you actually get is a lot of expensive noise. “
47
I’ve been building a record collection as if I were curating a painting exhibition. It’s called The Electric Line — 1967–2007. The idea is simple: a good exhibition isn’t a room containing the “best” paintings. It has a point of view. Someone decides where the story starts, what belongs, what doesn’t, what sits next to what — and what connections become visible. I wanted to do the same with records. The Electric Line is 40 albums tracing one particular path through popular music: how experimentation, technology and the studio changed what a band, a song and eventually an album could be. Velvet Underground → Bowie / Eno → Kraftwerk → punk → post-punk → New Order → acid house → electronic music → Radiohead → LCD Soundsystem. But it isn’t a canon. Important records are missing. Deliberately. The constraint is part of the curation. For every record I ask: Why does it belong? What came before it? What did it make possible afterwards? What was happening somewhere else? That has also changed how I collect vinyl. Instead of simply asking “Do I like this?” or “Is this important?”, I ask: What role does this record play in the story? I’m gradually collecting the 40 records on vinyl. And because a shelf can’t show all these relationships, I built a website for the collection. You can follow the chronological line. Or jump between records through their connections. Much like walking through an exhibition: there is a route suggested by the curator, but you’re free to construct your own. So The Electric Line is becoming three things at once: a collection, a curation project, and an exhibition about how ideas travel through music. electricline.whsrecords.com/
1
84
Se há algo que me irrita na discussão sobre o desenvolvimento de Portugal é a conversa da produtividade. Porque é sempre uma constatação de facto: é baixa, não cresce, tem que crescer. Mas o que raio é essa produtividade? Não trabalhamos? Fazemos ronha ou demasiadas pausas para café?! Nunca ninguém explica. Por isso fui ver. Fui pesquisar. eosistema.pt/artigos/porque-…
1
1
188
🟧Hugo H. Macedo 🟧 retweeted
Replying to @dhh
Evidence getting harder to ignore, but always hard to convert cult members. The below is PPP-adjusted, shows median and various income brackets, after in-kind transfers. Nowhere to hide.
2
13
865
🟧Hugo H. Macedo 🟧 retweeted
Coimbra to orbit. Neuraspace raising €15.6M to manage space traffic with AI is a proper Portuguese win, and the kind we should be shouting about more. Congrats to the team.
3
1
15
754
🟧Hugo H. Macedo 🟧 retweeted
Accurate.
Me pretending I understand everything Claude wrote💀
1
481
🟧Hugo H. Macedo 🟧 retweeted
Because we want our AI to thrive in a moment of exponential growth, we will impose some regulation so we are absolutely sure that only the best of the best can make it!! Thank you for protecting us!
Most of the AI act’s rules come into force today. The EU's #AI act is the world's first law on artificial intelligence. The act aims to ensure that AI systems are safe, ethical and trustworthy. Read more → link.europa.eu/rJWXFh
1
93
A - build a travel company app while waiting for boarding B - disconnect and open a book 😅
Family Vacations with city hopping, trains and road trips. Slightly planned. Great way to immerse yourself in the present and divert the monkey-mind. Family wise is also a challenge. It is the time of the year we spend most time together, we decide together, we experience everything together. Negotiation and tolerance are key for thriving and getting the most out of our time together.
44
Family Vacations with city hopping, trains and road trips. Slightly planned. Great way to immerse yourself in the present and divert the monkey-mind. Family wise is also a challenge. It is the time of the year we spend most time together, we decide together, we experience everything together. Negotiation and tolerance are key for thriving and getting the most out of our time together.
1
79
Interesting that, contrary to the initial beliefs, AI is evolving to become a system of models instead of a model to rule them all. Somehow we thought that even if we’re mimicking human thinking/language and even if human intelligence is spread across different individuals with different skills, the AI model would be a model to rule them all in some kind of superintelligence or Avengers-in-one.
25