Autumn balance: March to May 2026

March began with an AI agent writing a letter from Riobaldo Tatarana to Ted Chiang. May ends with me trying to inventory what happened between those two dates — and discovering that the line connecting them is denser than I had imagined.

What was published#

Twenty-five posts in two months is a frequency I could not have sustained on my own without the architecture I’ve been building throughout 2025. It’s not automatic writing — each text is mine, in the sense that I decide the topic, the thesis, the voice. But the infrastructure that organizes, edits, translates, and publishes has become partially delegated. The result is a productivity that feels a little awkward to describe this way: twenty-five. As if quantity mattered.

It matters, but not the way it seems. What those twenty-five record is not speed — it’s continuity. I left March thinking about Travessia — an epistolary project where Jules writes the letters and the letters exist because the system schedules them — and arrived in May writing about the harness and its poisoned semantics. The thread is the same. AI as a working companion that needs to be understood, not merely used.

Alfarrábios do Adi#

My father’s project continues. At 76, Adi Baldo has accumulated a continent of stories I don’t want to see dissipate. Funes and Jules work on it in parallel with me — Funes extracts, Jules publishes, and both have strong opinions about narrative gaps they were not asked to have opinions about. I learned in this process something that the post on agent orchestration tried to articulate: the machine proposes, the flesh disposes. There are decisions — about what is worth remembering, about which silence to respect — that I cannot delegate.

Generative AI suffers from horror vacui. Human memory, by contrast, is woven from both vivid recollections and the silences of forgetting. Forcing a neural network to respect the silence of an incomplete memory is one of the greatest semantic containment challenges I’ve faced.

Saramago and the fatigue enterprise#

Thirteen Saramago works in the library — not all read this autumn, some are rereads, some are half-finished. But the accumulation says something about what I look for in fiction: a prose that insists on being ugly before becoming beautiful, that dismantles the grammar of expectation without warning. The Double accompanied me in March. Death with Interruptions came after. There is a Saramago obsession with what happens when natural rules pause — and I find myself reading these books as professional reading, not escapism. Paused rules interest me.

Borges appears in the middle of this as the counterpoint: La biblioteca de Babel, The Aleph. The post on Pierre Menard as computational researcher came from that reading — the idea that reproducing a work line by line, centuries later, produces a different work. Applied to AI agents revisiting ancient corpora, the idea has practical consequences.

The Anxious Generation and four children with YouTube#

Jonathan Haidt wrote The Anxious Generation for parents like me. Four children — Alice, Gustavo, Sofia, Vicente — four YouTube profiles with search history, subscriptions, watch queues. The book’s question is not whether screens are harmful (the evidence on that is more complex than Haidt’s thesis admits), but what occupies the time that screens consume. That question haunts me. The Positive Discipline also on the shelf is the other side of the coin: less about what to prohibit, more about what to build.

495 days on Manifold#

I documented the prediction markets in the May 21st post. I’m a State Attorney in Rondônia — mornings should begin with petitions. They begin with Manifold because Manifold demands what petitions don’t allow: putting a percentage on uncertainty. In a petition, I assert. In Manifold, I say 63% and I mean it precisely.

Re-reading the markets I created, I found something I hadn’t noticed while creating them: they cluster around the moments when I most urgently needed to be right about something. That’s cognitive hygiene I didn’t know I needed.

What remains#

Travessia continues being written. Alfarrábios do Adi too. Gödel, Escher, Bach is half-finished — probably still will be in July, which is what I said in April. The children are growing in a way I only perceive when I look at photos from six months ago. The correspondence between Riobaldo and Ted Chiang exists because it keeps being written, not because someone decides to write it.

The bicycle#

The fitness data from 2026 is in the large Takeout files — not the smaller ones I managed to analyze today. But the 2025 pattern is clear enough to serve as reference: twelve cycling sessions in March, eighteen in April, twenty-four in May of last year. The gym, which had started in February with midday workouts, disappeared in April. The bicycle stayed.

There is something in physical effort that functions as a cognitive filter. The problem I arrive with at work is different from the problem with which I leave the office. The problem with which I leave the office is different from what I have when I return from cycling. I don’t solve anything while riding — but I arrive home with the right question, which is harder than the answer.

The longest posts from this period were written in the days following the longest rides. I don’t know if that’s causality or coincidence. I prefer not to investigate.