Paulo Silveira, in English

Human Out of the Loop, Asimov and Gradual Disempowerment

Thu, 24 Sep 2026 00:00:00 GMT

The doom scroll rules our lives. Classifiers and recommenders have understood the mechanisms of attention to the point that we are all addicted to technology (in Portuguese) and don't recognize it. Society is reacting: Meta reached a settlement with the American states in the trial over teen addiction. The European Union wants to ban social media accounts for children under 15.

The machine, the algorithm, knows what you will like. And even what you won't. It is well known how the people who design technology hijack our minds like a slot machine.

The same loop that rules your attention at night is the one running your Claude Code during the day. In any agentic tool, we are entering more intense work loops, with shorter cycles, without being given time to be critical of our own work. Some tweets blow up and show this climate of lost agency, besides a possible AI witch hunt. We are taking the human out of the loop:

Highlights mine; see the original tweet.

"If we were at least given the time to check out the code" can be read as the nostalgia of the extinct typist, but it is a plea to keep deciding. It is Terence Tao's helicopter (in Portuguese): you drop straight onto the waterfall without understanding the path.

How do we resist this predefined path? The conservative Society for Humanity appears in the last story of I, Robot:

If they had time on their hands, they would oppose mathematics or literature. The reactionaries who make up the Society for Humanity claim that the Machine robs man of his soul. Yet there is still a shortage of truly capable men in our society; we still need men intelligent enough to find the right questions to put to the Machine.

I read "The Evitable Conflict" in 2008 and reread it this week. In the story, artificial intelligence plans and governs the whole of the human economy, leaving nothing short and nothing in excess. Production and consumption so aligned that it no longer makes sense to argue about Adam Smith or Marx (in Asimov's words). At some point, planet Earth starts having small problems... delays in construction here, a small food shortage there... How, if the Machine is perfect? Could people be trying to sabotage the "algorithm"?

Then listen. Every action taken by a director who does not follow exactly the instructions of the Machine he works with becomes part of the data presented to the Machine in the next problem. As a result, the Machine knows that this director has a certain tendency to disobey. The Machine can incorporate that tendency into the data, even quantitatively, that is, calculating exactly when and in what direction the disobedience will occur. Its answers would be skewed just enough that, when the director disobeyed, he would automatically correct the answers, bringing them to the optimal direction. The Machine knows, Stephen!

Given the laws of robotics, the machine will find a way to help and protect us. Even if that means paying whatever cost it deems necessary. A "well-intentioned" paperclip maximizer that, instead of consuming human blood, takes in data and produces a sufficient minimum of ultra-processed culture:

Perhaps an agrarian or pastoral civilization, with fewer people and less culture, would be better. If so, the Machines must move in that direction, preferably without telling us, since, because of our ignorant prejudices, we only consider good what we are used to, and we would fight the change. Or perhaps the answer is total urbanization, or a caste-ridden society, or complete anarchy. We don't know. Only the Machines know; they are going there and taking us with them.

The Asimov passages above are translated back from the Brazilian edition I read; the original is in English.

Total urbanization or pastoral society? Only the Machines know. When Agent Smith interrogates Morpheus, he says the first Matrix was a perfect world. It was a disaster: we couldn't handle such a universe, and the Machine added suffering and misery to keep us busy.

https://www.youtube.com/watch?v=jrq2V7uwR20

p(doom), the probability of a disaster happening through artificial intelligence, doesn't need armed robots or biological disasters. It is enough to keep dominating the human psyche. Choosing for us. Taking the human out of the loop, slowly, produces gradual disempowerment: we have only a false sense of agency, but it has already been handed over to the machines, to the networks, to the stories and short videos... ultra-processed culture.

We delegated route planning and we obey Waze. We outsourced finding restaurants to apps. We are about to put our conversations and emotions at the disposal of the chatbots' echo chambers. The film Her is the past.

Literature has drawn this scenario for even longer. In 1958, in Asimov's "The Feeling of Power", an unimportant technician shows his "incredible ability" to multiply nine by seven without a computer, to the amazement of the New Pentagon. In 2005, Ted Chiang, of the film Arrival, wrote a piece in Nature, "What's expected of us", where the Predictor toys manage to turn on a light one second before you press them... with a devastating effect on their users:

Like a legion of Bartleby the Scriveners, they no longer engage in spontaneous action. Eventually, a third of those who play with a Predictor must be hospitalized because they won't feed themselves. [...] The ability to move remains, but the motivation is gone.

Resisting may make you look like a reactionary, like the Society for Humanity. And the Machine already knows you will resist, it knows where you will click like... it's in the data! Ted Chiang ends his story with a recommendation: pretend you have free will. Act as if your decision matters, because it does. Stay in the loop. Debate, criticism and regulation are collective actions too.

Fred Brooks, Fowler and what doesn't change in the profession

Wed, 16 Sep 2026 12:00:00 GMT

I keep browsing my notes, especially the ones I used to create talks. Notice that some themes keep coming back.

One is something I use a lot: you are not paid to write code (in Portuguese). The essence of a problem we want to solve is tied to what needs to be built. A process of discovery and refinement to define scope and requirements very well. Fred Brooks shows up a lot here.

The other theme is this cross-pollination, how work has become more complex, especially in corporate environments, where a single specialization isn't enough, given all the communication we have to do with other teams. Looking at a problem more broadly. From a higher level, you end up picking up needs on a horizontal plane: skills that used to have nothing to do with your profession.

The quotes that repeat the most in my slides:

"The hardest single part of building a software system is deciding precisely what to build."

Fred Brooks, No Silver Bullet, 1986. In a deck from this year I noted that the sentence aged well. If code became cheap and disposable, what is left that is hard, and valuable, is exactly what he pointed out. Deciding what.

"The complexity of software is an essential property, not an accidental one."

Fred Brooks, in the same text. It is the split between accidental and essential complexity. The accidental one is what tools knock down, one layer at a time. The essential one is still there.

"When anyone can build anything, knowing what's worth building becomes the skill."

Kent Beck, in The Pragmatic Engineer, December 2025. Fifty years after Brooks, the same conclusion from the other side: when anyone builds anything, knowing what is worth building, and what is worth keeping, is the scarce skill.

"Architecture represents the significant design decisions that shape a system, where significant is measured by cost of change."

Grady Booch. If architecture is what is expensive to change, AI just moved the yardstick: writing got cheap, changing is still expensive.

"A business analyst writing code out of curiosity, a front-end engineer dabbling in DevOps, a data engineer trying product analysis: each cross-pollination broadens both the individual and the team."

Martin Fowler, Expert Generalists.

"T-shaped developers combine depth with breadth. They can dive deep into specific problems but also understand how it fits into a larger system."

Werner Vogels, CTO of Amazon. And he is not the only one saying it: Osmani, Fowler, Bill Gates.

"Junior engineers often try to widen the horizontal bar too fast without ever digging deep. Senior excellence requires that vertical anchor."

Addy Osmani, engineering lead at Google Chrome. The horizontal bar is context, the vertical one is confidence.

"Organizations which design systems (in the broad sense used here) are constrained to produce designs which are copies of the communication structures of these organizations."

Conway's law, from How Do Committees Invent?, 1968.

"Code is suddenly free, ephemeral, malleable, discardable after single use."

Andrej Karpathy, Year in Review 2025.

AGI, well-defined tasks and a master's degree in 65 minutes

Sat, 12 Sep 2026 00:00:00 GMT

Jensen Huang welcomed AGI with the arrival of Astra. Personally, I don't like the term, and it ended up becoming a marketing device, as Sam Altman himself said. And there is a conflict of interest with the big artificial intelligence labs getting close to their IPOs.

That said, we can't deny what the models have achieved. Well-scoped tasks with enough context can already be carried out end to end by the harness of these models. Filling in forms, doing research, writing texts, clicking through sequences of buttons in apps with no prior knowledge. All of these tasks happen basically without any error. And, at any stumble, the reasoning itself detects it, tests and tries another path. AI's current critical evaluation of its own work is surprising, to say the least.

For a few months now it has been impressive to watch AI click around your browser, switch tabs, fill in fields and move information from one place to another, in a chained sequence of actions that makes long tasks possible. The Astra launch video shows use cases that go much further:

https://www.youtube.com/watch?v=1QNsdr-Qx_I

I believe the term AGI will disappear leaving few traces, just like the Turing test. Notice that we no longer argue whether the Turing test has been passed or not: we simply turned the page and it became a minor question, almost decorative. Arguing over the fine print of Turing's criteria ends up being just a distraction.

The same may happen with AGI (in Portuguese). We will look back and there won't be a set date, a specific milestone. We will see rankings, classifiers, recaptchas, cat detection, ChatGPT, self-driving cars and Chinese robots as one continuous block.

When Astra came out, I talked with Fabrício Carraro, host of the podcast IA Sob Controle. He suggested I run a benchmark, and I proposed trying to implement the algorithm from my master's dissertation, from 20 years ago.

The problem we studied is this: given a polygon drawn on paper, how can we fold it so that all its edges line up and, with a single cut of the scissors, cut out exactly that polygon, so that when you unfold the sheet the hole is the drawing. An interesting magic trick, proven by professor Erik Demaine, from MIT: there is always a solution. He laid out two algorithms in a short and concise paper, where he stated the problem and also the solution.

Rafael Cosentino and I, along with our advisor José Coelho de Pina Jr., spent a year on the paper, also learning computational origami from Robert Lang and geometry from O'Rourke. For about four months we wrote JOrigami. We were young programmers. The algorithm itself isn't hard, but it is a lot of work, even more so at a time when libraries of computational geometry primitives were less available.

I gave Demaine's paper as a PDF to Claude (Fable 5.1) to execute. It took a little over an hour and produced this:

The AI itself explains here how it implemented the algorithm (in Portuguese) and the differences it found from our implementation.

A year of work, five months of programming. The AI did it in 65 minutes. I wasn't surprised. Nor was I questioning the definition of AGI.

But there are already immediate impacts: we need to understand how we will keep doing science. Some argue that solving Navier-Stokes through traditional methods could have brought other intermediate discoveries. In this note I wrote about the analogy Terence Tao makes (in Portuguese) with the scientific method and the process of discovery. Getting to the result is important, it is valid, and it is very powerful to have access to today's tools. At the same time, we may be leaving aside important questions, subtasks that could captivate our minds and curiosity. Some people push back, saying that AI itself will also see this and have that mechanism of restlessness. That is not the case now.

The more complex problem remains: the development of juniors' careers, and I'm not talking only about devs. Their career gets more complicated once well-scoped tasks, like front-end work or database consulting, are executed perfectly by AI. This also happens in science, in my own master's. When my professor picked this problem, he clearly saw that the paper was feasible to implement for a young person testing his interest in academia and science. It was an interesting computing problem for learning primitives, geometry, combinatorics and a bit of science while we banged our heads and exchanged emails and questions with important researchers.

Where will these simple problems that bring people into science and research go? And in programming, in medicine or in any career where the start involves a lot of repetition, understanding and pattern recognition? Could AI be reducing our productive effort so much that it causes scientific stagnation (in Portuguese)? Without warming up on small problems, who will have the interest and the ability to ask the new, good questions? The Fields medalists, together with Tao, also spoke up this week: "the mass production at faster and faster pace of "true/false" statements could destroy fertile ground instead of breathing life into new ideas".

My father read the mathematicians' manifesto and saw in it a defense of how things are done inside an institution, a group. He remembered two cases. The first is the steel axes in Australia, which the anthropologist Lauriston Sharp documented among the Yir Yoront, on Cape York, in the 1930s. The stone axe belonged only to the older men and was traded at ceremonies. When the Anglican mission started handing steel axes directly to women and young people, those ceremonies lost their reason to exist, and their system of myths had no way to fit the new object. The second is firearms among the native peoples of Brazil. At first, only those at the top could use the technology, like the Guarani of the Jesuit missions (in Portuguese), armed with authorization from the King of Spain to fight the bandeirantes. Later, the result started arriving faster, the rituals died, and nobody could hold back the use of these tools anymore.

The manifesto can be read that way, as a defense of the work and rituals of a group. Maybe. But rituals are where much of the tacit knowledge and the social exchange live. How can we reconcile that with progress?

Open Claw, Personal Knowledge and the Second Brain

Tue, 03 Mar 2026 00:00:00 GMT

What got me back into programming wasn't Claude or Codex. It was OpenClaw. Not because of the personal assistant idea, but as a Second Brain. Or a Zettelkasten.

The bombardment of information, the excess of content, meetings, and notes has been a major challenge. Everything spawns threads in parallel in our brains, fragmenting attention. I needed to find a way to manage all that knowledge, links, videos, and notes.

I remember reading classic literature. Some of the great authors dictated what would be written, what was being thought. Dostoevsky did this (his second wife was his stenographer!), and other authors of long volumes drew maps that would look like our mind maps. I had the chance to see some at the Dostoevsky Museum in Saint Petersburg.

Personal Knowledge Management mechanisms are really nothing new. From personal wikis to Martin Fowler's Bliki (although today it's more of a blog).

We've seen the waves, right? From Delicious and Pocket, saving links and clippings, to Trello and Evernote, then Notion and Obsidian. The question remains: where do I scribble, annotate, summarize, store, and search?

I tested several of these tools and realized none matched the workflow I had in mind. So I vibecoded my own solution. A Personal Knowledge Management system that, given everything I've recorded, noted, and reflected on, ultimately helps me write posts for this blog.

Yes, products like mem.ai do something similar. But in the world of disposable software, it's more interesting to have something extremely customized to my needs. And what are those needs?

Record audio, text, images, and ideas immutably
Classify that data into labels
Generate concepts: Knowledge Base (KB) nodes
Use a KB node to create or append to a blog draft

It's a specific piece of software deployed on Fly.io, listening to a Telegram bot via a simple webhook and BotFather. It calls Whisper via OpenAI and Anthropic (Sonnet) to classify and synthesize the text, always respecting what I actually said, never inventing words.

With this small system I can record voice notes, even out of order, on different days. The journaling is the single source of truth with the direct transcription. From that journaling, concepts are generated (Knowledge Base nodes) and classified. The journaling can also enrich a blog draft, concatenating the selected transcription of a Concept if it's classified as such. There are many other details, like the classifier receiving the current concepts and drafts as candidates, and a glossary being built over time so Whisper can handle terms and acronyms I use frequently.

Could I do something like this via OpenClaw? I imagine so. I doubt the result could be kept within an expected structure. Small agents for specific cases should work better than a do-everything agent. At least for now.

I asked for help with a diagram to better explain what I'm doing programmatically:

import PipelineDiagram from "../../../components/alun/PipelineDiagram.astro";

<div class="not-prose pipeline"> <div class="node">📱 Telegram<div class="node-sub">voice, text, photo, link</div></div> <div class="arrow-down"><svg viewBox="0 0 24 28"><path d="M12 0 V18" /><polygon points="7,16 12,26 17,16" /></svg></div> <div class="node">🎙️ Whisper<div class="node-sub">transcription</div></div> <div class="arrow-down"><svg viewBox="0 0 24 28"><path d="M12 0 V18" /><polygon points="7,16 12,26 17,16" /></svg></div> <div class="node">📓 Journal<div class="node-sub">immutable entry · enriches concepts and drafts</div></div> <div class="arrow-down"><svg viewBox="0 0 24 28"><path d="M12 0 V18" /><polygon points="7,16 12,26 17,16" /></svg></div> <div class="node node--accent">🧠 Classifier<div class="node-sub">Opus</div></div> <div class="arrow-down"><svg viewBox="0 0 24 28"><path d="M12 0 V18" /><polygon points="7,16 12,26 17,16" /></svg></div> <div class="group-box"> <div class="node-row"> <div class="node">💡 Concept<div class="node-sub">note · always → KB</div></div> <div class="node">✏️ Draft<div class="node-sub">intent → blog / project / task</div></div> </div> </div> <div class="arrow-down"><svg viewBox="0 0 24 28"><path d="M12 0 V18" /><polygon points="7,16 12,26 17,16" /></svg></div> <div class="node" style="background:#0f172a;color:#94a3b8;border-color:#334155">🗄️ Git<div class="node-sub">vault — Journal + Concepts + Drafts</div></div> </div>

Then, with these drafts, I refine the articles myself, because there's too much repetition, gaps, and the ideas never fit together neatly.

I could try to leave that concatenation of ideas and text restructuring to the machine: ask the LLM to take this pile of messy notes and create a publish-ready article. I won't. The LLM, no matter how "humanized," would change too much information and create content from its own "head," giving off that slop feel we see everywhere now, where everything looks like anything. It would also lose authenticity in a noticeable way.

And to give you an idea, I wrote this post through several fragments and voice notes I recorded over the course of days, ordered quite differently from how they appear (some I removed):

  1. Intro — OpenClaw, Second Brain, Zettelkasten, waves of tools
  2. tg-121 — PKM history: wiki, Bliki, Zettelkasten/Luhmann
  3. tg-29 — Broad context: info overload, Markdown, LLMs, Obsidian, SaaS
  4. tg-29 — Dostoevsky as the original second brain, Meta/Zuckerberg, fragility
  5. tg-106 — The concrete system: Telegram → Whisper → Fly.io → Anthropic
  6. tg-30 — Voice journaling workflow (record, transcribe, refine)
  7. tg-129 — Image (bot screenshot)
  8. tg-149 — Meta-reflection: this post was built by the system, human curation,
     disposable software
  9. tg-132 — Vibecoding meme
  10. tg-34 — Human-machine symbiosis
  11. tg-34 — The prompt injection paradox (provocative ending)

Did it dramatically reduce my work time? No. I've been at this draft for hours. But now I no longer lose ideas and texts I'd like to study or share.

PS (September 2026): I no longer use this solution this way. My usage today is much more pragmatic and simpler, as I tell in my personal knowledge system after six months (in Portuguese).