Getting Fried Part 2c: Sacred Deceleration

In which a year passes and the founder, for the wrong reasons, does some of the right things.

(This is one of three branches continuing from Part 1: The Week I Lost the Plot at Cogentiv.ai. The other two: 2a: Apotheosis and 2b: Feierabend.)

A year had passed. I’d bought a notebook and filled it — one honest sentence a day, starting with the coffee is cold now. Ines, my manager, had taken me to Currywurst at a place in Moabit with no wifi, told me she’d been crying in the third-floor bathroom, added me to a Signal group called Wir werden hier leise verrückt where I learned I wasn’t uniquely broken, just commonly broken. By July I’d negotiated agent-free Fridays by lying to Malte — our founder — about prototyping a deep-work methodology. Life drawing Wednesdays. A novel finished in bed with crying at the end, which I think is what novels are for. [cognitive debt, first installment toward repayment]

It saved me. It would not, on its own, have saved anyone else.

By March, two or three new faces on my team had names climbing the Index in the old pattern — overnight sessions, single-sentence Slack replies at any hour, lunch at the desk, no lifeboat. I thought about saying something. I didn’t — because my whole quiet insurgency worked partly because nobody else was doing it. If everyone did it, it would have become visible; visible things at Cogentiv get measured; measured things get optimized; and optimized things end up as a “Cognitive Downshift Protocol” deck with a weekly adherence KPI. So I kept it private, and keeping it private meant keeping it from them.

This is the part where, in a different version of this story, I’d tell you that individual resistance was enough.

It wasn’t. Here’s what was.

April. Lisbon.

Here’s what you need to know about what happened to Malte in April: he fainted. Specifically, he fainted during a fireside chat at a tier-two AI conference in Lisbon, mid-sentence, in front of about four hundred people who were, at the exact moment of his collapse, hearing him say the phrase “post-cognitive leadership.” He fell sideways off a Herman Miller stool. There’s video. I’ve seen it. Malte has seen it. Malte has seen it many, many times, because the video went mildly viral on German tech Twitter, and because Malte (this is the thing about Malte) metabolizes. He doesn’t have experiences. He has content.

The diagnosis, as relayed to us in a company-wide Loom filmed from his hospital bed in the Charité: stress-induced syncope, compounded by chronic sleep deficit, compounded by what his cardiologist apparently called, with some understatement, “an unsustainable autonomic load.” [brain fry, hitting the founder tier] He was wearing the Arc’teryx gilet over his hospital gown. He looked, for the first time in the twenty months I’ve known him, like a person.

“Founders,” he said (because we’re all founders), “I have been running my nervous system like a seed-stage startup. Burn rate too high. No runway.” He paused. He was reading off something. “The body,” he said, “is the original product, and I have not been maintaining it.”

Ines, sitting next to me in the Kreuzberg collision space, turned and mouthed oh no. Not oh no as in poor Malte. Oh no as in: he’s about to turn this into a framework.

She was right.

May. The Pivot.

The all-hands deck was fifty-seven slides. It was titled HUMAN SYSTEMS OPTIMIZATION: A FOUNDER’S FRAMEWORK FOR SUSTAINABLE COGNITION. It contained, in order: a photo of Malte in the Charité (gilet, hospital gown, IV drip, caption: “Day Zero”), a scatter plot correlating Throughput Index rank with self-reported sleep hours (no axis labels), a quote from a neuroscientist Malte had cold-DM’d on LinkedIn, a section titled “The Neurofeierabend Thesis,” and a roadmap.

The roadmap was (and this is where I need you to hold two things in your head at once) both genuinely good and almost physically painful to read.

“We have been treating cognition as an infinite resource,” Malte said, standing at the front of the room in his Salomons, pointing at a slide that read COGNITIVE SUSTAINABILITY IS THE NEW GROWTH LEVER. “We have not been designing the workplace for these tools. We foisted it on people.” He actually said this. He said foisted. Ines grabbed my arm under the table. On Slide 34 he also name-dropped the EU AI Act’s August enforcement deadline, which he’d previously called “regulatory theater” but which now served as a convenient branding garnish (“we’re not just sustainable, we’re compliant”), though everyone in the room understood that the real reason for the pivot was lying in a Charité bed three weeks ago wearing an Arc’teryx gilet over a hospital gown. I don’t know if he’d read the same CNBC article I’d read in the Signal group or if he’d arrived at foisted independently or if Claude had suggested it, but he said it, and for about four seconds the room was very quiet, and I thought: you absolute creature, you have stumbled face-first off a Herman Miller stool into something true. [“we never designed the workplace for these kinds of tools”; arrived at via cardiac event, garnished with regulatory branding]

June through August. The Measures.

They rolled out over six weeks, and every single one had a name that made me want to lie down on the floor.

I should tell you what it feels like, watching the man who caused the problem rebrand himself as the solution. It feels like watching someone set a building on fire and then accept an award for redesigning the exit signs. It feels like Ines and I spent months in a Signal group, sharing papers, naming the damage, doing the slow invisible work of figuring out what was wrong, and then Malte fainted off a stool and got a standing ovation for discovering gravity. I want to be clear: I’m not being fair. I know I’m not being fair. The measures are good. The measures work. The fact that they came wrapped in a fifty-seven-slide deck and a podcast tour doesn’t make them less real. But there’s something in me (the part that was crying in the second-floor bathroom while Malte was in the sauna dictating specs to ChatGPT) that needs you to know: the people who see the damage first are never the people who get credit for the repair. That’s not a Cogentiv problem. That’s just how it works.

Cognitive Load Ceilings. Maximum four hours per day of active agent oversight per person. The remaining hours are designated “Sovereign Processing Time” (Malte’s phrase, god help us) during which you’re expected to do non-agentic work: thinking, sketching, reading, talking to a colleague face to face. The Throughput Index still exists but now has a cap; exceed the cap and your manager gets a flag. Leander exceeds the cap on day one and looks at me with the bewildered expression of a golden retriever who’s been told the ball is not for throwing. [automation complacency, structurally interrupted; the leaderboard now has a ceiling]

Neurofeierabend. I’m not making this up. This is what Malte calls mandatory end-of-day. No Slack, no Claude, no Flora after 7pm. The policy is enforced by an API integration that throttles token access after hours. Malte announced it with the phrase “we are building a culture of sacred deceleration.” Ines texted me during the announcement: sacred deceleration ich kotze gleich. But the thing is, the thing is: it works. Not because the language is good, but because the API throttle is real. You can’t, after 7pm, talk to Claude. You can’t, after 7pm, talk to Flora. You can, after 7pm, sit on your balcony and stare at the sky and notice that the sky is a sky. Some people hate it. Leander hates it; he’s started going to a co-working space in Friedrichshain after hours to use his personal API key. But most people, quietly, have started sleeping. [Feierabend, industrialized; cringe name, real mechanism]

Pair Prompting. No agentic task over a certain complexity threshold to be done solo. You sit with a colleague; one prompts, the other reads; then you switch. Malte calls this “Cognitive Cross-Pollination” and has had posters made. The posters feature two brains with a lightning bolt between them and they’re hideous. But the practice itself is (reluctantly, irritatingly) useful, because it turns out that agents produce output that feels correct the way a well-designed phishing email feels legitimate: the confidence is the problem, not the solution. Having a second human in the loop catches the moment when the output is beautiful and fluent and nobody knows what it means. It’s pair programming, in essence, reinvented by a man who has never heard of pair programming, branded with a lightning-bolt poster, and it works. [metacognitive laziness, structurally addressed; fluency illusion, caught by the second pair of eyes; you can’t skip the monitoring if someone is watching you skip it]

Mandatory Analog Wednesdays. One meeting per week, no screens, whiteboard only. Malte calls this “Paleolithic Strategy” and has sourced a full Neuland facilitator set — the markers consultants use at design-thinking retreats — for the occasion. The first session is a disaster; people stare at the whiteboard with the hollow gaze of tourists at a museum without an audio guide. By the fourth, Priya draws a system diagram from memory. It’s wrong in several places, and it’s the most useful artifact the team has produced in months, because the wrongness is visible and arguable. Someone points at a box and says, “What does this do?” and Priya says, “I don’t know,” and the room doesn’t nod. The room discusses. [intent debt, partially addressed; shared understanding, rebuilt by hand, literally]

Flora, restructured. Flora isn’t killed (Malte still loves Flora in the way a man loves his most expensive mistake) but she’s moved from “emotional co-founder” to “admin assistant.” Flora can book your meeting rooms, summarize your calendar, and remind you to drink water. She can’t, anymore, ask you how you’re feeling, validate your dreams, or tell you you’re doing brave work just by existing. This change appears quietly in a changelog. Jarek, who’s come back from the Engadin in July (thirty pounds heavier, newly capable of sustained eye contact, talking about his therapist with the reverence he used to reserve for Cursor), notices the change on day one and says, to nobody in particular, “Oh thank god.” [digital therapeutic alliance, surgically reduced; the tool remains, the parasocial surface is removed]

September. What the combination does.

Here’s the thing I couldn’t have predicted from the balcony: my notebook and Malte’s framework aren’t the same thing, but they rhyme.

The notebook taught me that I could think a thought. The Cognitive Load Ceiling gave me four hours a day in which thinking a thought was permitted. The life drawing taught me that my hand still knew how to see. The Paleolithic Strategy Wednesday gave other people’s hands the same chance. The Signal group told me I wasn’t broken. The Pair Prompting showed Leander that he also wasn’t broken; he was just, like all of us, trying to review code at a speed the human visual cortex was not designed for.

None of this would have worked alone. My notebook, without the structural support, would have remained what it was in March: a private lifeboat on a company-wide Titanic, and if you think buying yourself a notebook is a solution, you haven’t understood how shit rolls downhill. And Malte’s framework, without people like me and Ines who’d already started the work from below, who’d already learned what it felt like to think again and could model it for the people who hadn’t, would have been another fifty-seven-slide deck that everybody nodded at and nobody absorbed.

The Index has a cap now. The people who’d been climbing it the way I used to aren’t climbing past that cap. Ines invites them for Friday coffee. They come, some of them. They start arguing with Priya at team drinks about what the orchestrator does, and nobody is right, and everybody is thinking, and the thinking is loud and slow and human and wrong, and it’s the most productive sound I’ve heard in this building in two years. [System 0, de-occupied; the difference between coping and structure: one saves you, the other saves the next person]

Saturday morning.

Saturday morning. Cold, the way November in Berlin is cold: not dramatic, just persistent. I’m on the balcony in a jacket that isn’t warm enough, holding a Kaffee that is, notebook open on the railing.

Malte is, right now, on a podcast. I know this because Leander posted the link in Slack at 7am with three fire emojis, which is Leander’s way of saying he’s awake and performing enthusiasm. Malte is explaining “sustainable cognition” to a host who keeps saying “wow.” He will, by January, have a LinkedIn post about it containing the word “journey.” He will file a trademark application for “Neurofeierabend.” He will always be Malte.

But the API throttle is real, the cap on the Index is real, and on Wednesdays the room argues about wrong diagrams. And somebody is sleeping tonight who wouldn’t have been. And I didn’t have to choose between my notebook and my job.

A thought arrives. Mine. Something about the light on the Altbau across the street, warm on cold stone at this hour. I write it down.

Neurofeierabend, I write at the bottom of the page, because Malte has ruined the real word and I might as well let him have it.

I close the notebook. I roll my eyes. I go inside, and it’s warm. [losing the plot, structurally interrupted; not by the narrator alone, and not prettily, but interrupted]

What if nobody had fainted?

• Apotheosis — nobody brakes.
• Feierabend — one person brakes, privately.

For the vocabulary behind the margin notes, see The Vocabulary of Getting Fried.