1

“Thick fog drifts eternally over the water’s surface. Through the steaming cloud from the foam the tops of the tall palms emerge. When a ray of the glowing evening sun penetrates the damp vapors, there begins an optical magic. Colorful rainbows disappear and then return.”
Alexander v. Humboldt, Views of Nature

About a year ago, I walked in the shade of Aleppo pines, along the Adriatic coastline of a dreamy vacation park beyond its prime years. Listening to podcasts of people smarter than me, I followed some concrete slabs on to a small headland of reddish rocks cast into a sea like epoxy resin. From my vantage point the park with its tennis courts, rusty playground constructions and deserted beach bars resembled the dioramas I had built as a kid. I felt sun-dizzy with a sense of total control, as if the world was my laboratory.

This was the moment I got seized by the spirit of scientific self improvement. Things suddenly amalgamated and I took down a note on my sandy phone: “But how to figure out what’s next? The obvious way is experimentation: a structured way to learn from your environment with limited investment and managed risk. It’s a practical application of scientific methodology, that makes sense in many areas of life.”
A frenzy followed: I was sure I had found an opening in the constraints of my life, and I wanted to crawl into it.

When I finished my walk I was fully set up to run on experimentation, making life my lab, documented by numerous datapoints. I had rewired my note-taking system, my to-do list and probably my prefrontal cortex as well.

My scientific episode lasted about three weeks. I did not finish a single experiment. I think the reason I failed my experimental ambitions was not a lack of commitment or discipline. If you take “experimentation” seriously (and I did) then you have to establish a stable baseline first and then modify single parameters to trace change. And this is where we find the issues. None of the things I cared about (Sleep, Fitness, Mood, “Productivity”, ...) could be considered an isolated parameter. Everything influenced everything else. My baseline was moving too fast. A week after I obsessed over “running my life on experiments”, something else happened that occupied my focus and affected my behaviour. The trigger might have been a lesson I learned navigating my days with kid, dog and spouse. Or another smart essay I read online. There is always something to jeopardize the stable conditions a science-ish experiment would demand. The more I move into the sphere of self improvement, the more I pay attention to how I do things. And the faster I seem to change.

While these obstacles to scientific self-improvement are timeless, there is another challenge that seems rather contemporary. Open YouTube on a weekday afternoon. Within three ads you will meet them: jaw-clenched men standing in front of leased Lambos, promising you the protocol that changed their lives. Investment schemes, that one exercise to pump your trapezius muscles, STOP doing this one habit at work and instead buy a truckload of dubious food supplements.

Quantified Self and Sigma Grindset might have been avantgarde attitudes some years ago. But something changed. The hustlers keep hustling, but their takes feel familiar. Their hair starts to grey and their self actualization seems a bit desperate. Whenever I get harassed by Self-Improvement-Dim-Wits in Youtube Advertisements I not only feel seen by the algorithm. I feel a sense of second-hand embarrassment and to be honest, some pity. I think we (as in “me”) have moved on by now, to a terrain that is far less clear in its coordinates. And this makes scientific self improvement even harder. An experiment has to start with a thesis, driven by a motivation. And if I am really honest to myself, which I should be as I am tweaking my lifeOS, figuring out my intentions is the hardest thing.

I realise that self-experimentation inspired by scientific methodology has become a dated method. Yet rejecting it entirely is hard. The impulse still pulls. It seems I, and maybe others too, am in a transitory state, caught between old ideas clinging, while silhouettes of what might replace them shine through the fog sometimes.

2

“What is the nature of life? What is its deep structure; what governing forces shape the appearances we see? We simply do not know. Science is arrested here, for the springs of vitality are not yet accessible to its methods. [My poem] serves as a placeholder, a visible marker that makes evident not what is known but what is not (…) thereby opening a visible gap in the fabric of knowledge.”
Alexander v. Humboldt, Views of Nature

While Alexander von Humboldt was considered the most-important living European next to Napoleon1 his relevance had waned towards the end of his life. Charles Darwin always admired Humboldt, yet when they finally met, Darwin was unimpressed by the chatty old man.2 Emil du Bois-Reymond, another disciple of Humboldt, described his late mentor “a living ruin” in a speech after Humboldt’s death.3 In private, du Bois-Reymond shared the same sentiment, calling Humboldt “antediluvian”, which is “something from before the biblical floods”.4

On his explorations, Humboldt came up with several concepts modern scientists take for granted today: he invented isothermal maps, discovered the magnetic equator, founded the discipline of Plant Geography and was one of the first to identify human intervention as a main cause for climate change. But despite his long lasting contributions, Humboldt’s practice jarred with the streamlined apparatus of professional science emerging as Humboldt was getting old.

While Humboldt was obsessive on collecting data in the field, he resisted the early impulse of crunching numbers on his observations. After discovering how Earth’s magnetism varies with latitude, he handed the work of integrating the data into mathematical formulas to his friend Carl Gauss. Moving from raw data to modeling was not the thing Humboldt cared about. He consistently refused to polish his publications into coherent thesis-driven devices and instead insisted on showing the overwhelming richness of his work, with all its gaps and inconsistencies. This became most apparent in his Magnum Opus Cosmos, which contained an enormous amount of footnotes and struggled to form a coherent synthesis.

Humboldt just did not care about being a scientist in the modern sense, not the least because the very term “scientist” was only coined when Humboldt already had done his major expeditions. His way of working did not match the efficiency and precision demanded by the crystalizing branches of modern science.

It is tempting to frame Humboldt as catalyst for a paradigmatic change that rendered himself obsolete. But I think the more interesting read is that Humboldt was in a transitory state. He was a forethinker of scientific modernity (and that’s what his official legacy is about) but he also represented a different mode of inquiry. Others have framed everything of Humboldt’s method that did not fit into modernity’s toolbox as an “anachronism”, but I prefer to consider him a hybrid of alternative modes for seeing the world.

Du Bois Reymond I mentioned earlier lived scientific modernity with his clean lab experiments on quantifying nerve impulses. Every petri dish a miniature glimpse of life, the scientist in total control of the arrangement. Charles Darwin on the other hand can be considered another hybrid: heavily inspired by Humboldt, Darwin shared a similar practice of note-taking, yet he opted for a much cleaner, thesis-driven way of publication, hiding a lot of the intuitive messiness from the public.

Humboldt’s science was non reproducible, embodied and embedded into the territory he was exploring. Humboldt was not building dioramas.

3

“When, after many undulations of the ground on the craggy mountainside, we finally reached the highest point of the Alto de Huangamarca, the long-veiled vault of the sky suddenly cleared. A sharp southwest wind chased away the mist
(…) For the first time, we were seeing the Pacific Ocean; we saw it clearly: next to the littoral, a great body of light, reflecting, ascending to a horizon now more than merely sensed.”
Alexander v. Humboldt, Views of Nature

My failing self experimentation might be attributed just to myself. My unstable character, my jumpiness. But there is this contemporary element I can not deny. Call it Disruption, Zeitenwende, Gramsci-Gap.5 Every generation assumes they live through unprecedented change, and they’re usually wrong. But the thesis I subscribe to6 suggests something more useful than either “everything is unprecedented” or “nothing really changes”. History happens as long, overlapping arcs of how things feel, work, and make sense. Some rising, some operational, some fading. This implies that the “in-between” of a transitory state is the honest default at any point in history. And it reveals alternative paths for our wandering.

Reading my notes back with that frame in mind, the failures I described seem less personal. Several conditions were missing which lab experiments structurally require: a baseline that holds still long enough to measure against, variables that can be isolated from each other, and an observer who is not also the system under observation.
In nineteenth-century laboratories (like du Bois-Reymond exploring nerve impulses) those three could be approximated well enough to let entire fields take off. Applied to my self in a moment like this one, none of these hold. The baseline drifts faster than the experiment runs. The variables can not be decoupled: sleep affects mood affects “productivity” affects the kid affects sleep, in a loop with no clean ingress. The observer (me) rewires themselves while running the experiments.

The configuration of my self does not sit still, and I am the moving terrain on which I have to place my self improvement toolkit. Lab science of the self belongs to a configuration that is passing. It was harder to see when the ground felt more stable, but it becomes obvious to me now. My recent experimental phase was not calculated but driven by my need to find a new lever I can grab. I had assembled a miniature replica of my life, complex but irrelevant beyond the mere purpose of self reassurance. I was building a shitty castle.7

Counter-evidence exists. Seth Roberts ate butter every morning for years and tracked his arithmetic speed; he lost weight by drinking flavorless sugar water; he stood up more and slept better. He ran disciplined N=1 for over a decade and learned things that could be replicated. The case against lab-science-of-the-self isn’t that no one had ever done it well. But doing it well requires a composure stable enough to be the baseline. Roberts maintained this stance in the 90s and 2000s. It was a coherent configuration that some people still inhabit, but others (me, perhaps you) no longer do.

Humboldt’s idiosyncratic fieldstudies provide other impulses than lab experiments. These may be considered elements for a timely way of learning things about ourselves and the world, an epistemology of divergence:

1. Humboldt did not speculate from his armchair but embedded himself into the matter of his interests, rediscovering South America’s territory beyond the maps of Spanish colonialism.

2. Humboldt wrestled with eels to learn about their electricity. On the slope of Chimborazo, mouth bleeding from height sickness, he used a cyanometer to measure the color of the sky against his own eyes. Sleeping in the mud of the Amazon river beds fueled long speculations on the impact of mosquitoes on civilizations.

3. When Humboldt struggled to nail down his theory of “Lebenskraft” in Views of Nature, he did not pretend. Instead he inserted a poem written years earlier, marked it as “something other than science” and let it stand for the gap. A change of registers avoided forcing his essay into artificial closure and pointed at something that might come later.

I notice a temptation to create a new personal doctrine, to announce field studies as the new method to replace scientific self experimentation. That might just start the construction of another shitty castle. I commit to this anyway. Every method is partly wrong, but the direction of failure is more interesting. Failing backwards means casting a good idea into a dead diorama. Failing forwards means keeping eyes and mind open, searching for signal and marking the gaps.

Months later, walking below beech and oak steaming from recent rain. The dog trails on. I take some notes on my phone, plain text only. A lighter harness allows me to see more. We move on together, right into the fog ahead.

Jetzt abonnieren

5

Contraptions

The Gramsci Gap

The Contraptions Book Club is currently reading City of Fortune by Roger Crowley, to be discussed the week of January 27…

Read more

2 years ago · 78 likes · 15 comments · Venkatesh Rao

When I was a kid, castles were serious business.

Holiday trips had to feature enough piles of crumbling stone to keep me and my brother compliant with the rest of my parents plans. Over time, I became a castle connoisseur with a distinct taste for what makes a cool castle, and being the judgemental kid I was, even more importantly a taste for what defines a shitty castle. In a nutshell, a castle had lost its justification for existence when it was not gritty enough. As I grew older, my judgement became more subtle and I realized that what disqualified castles the most were castles not meant to defend anything, but only to resemble actual fortifications. 10-year-old me enjoyed a good fortress and hated the chateau. Every time I ended up disappointed in a 19th century palace villa with skinny walls that would fall faster to a battering ram than dark age palisades I wondered: Why would you build a castle that is not a castle?

In 1676, Leibniz discretely visited the infamous Spinoza in his modest workshop, where he maintained a humble business of grinding lenses. Leibniz was eager to meet the man that had published the most radical ideas of his time, considered so dangerous that it got Spinoza expelled with colourful words1 from his religious community and led him to spend the rest of his life in the intellectual underground.

At the core of Spinoza’s thinking lay a disruptive argument: because a truly infinite God must encompass all reality, nothing can exist outside of him. Therefore, Spinoza argued that God and Nature had to be the same thing. This idea was destined to drive a powerful wedge into the rigid belief systems of 17th century European philosophy. If God is not the guy with a beard watching and judging us from a cloud but some abstract omnipresent quality, who would let hellfire rain on sinners and atheists?

Although little is known about the short meeting of Leibniz and Spinoza, it seems that Leibniz was impressed and unsettled by the ideas of his fellow thinker. After Spinoza’s early death though, Leibniz increasingly positioned himself as an opponent to Spinoza. Eventually he proposed the idea that everything was built from an infinite number of tiny “Monad” units, each of them containing its own tiny immortal soul which still had to fear hellfire from above. In public, Leibniz claimed the monads were designed to defeat Spinoza’s radical findings.2 Yet many assumed that actually, Leibniz understood his counter-theory was not strong enough to resist the ideas Spinoza had brought into the world.3 It seems Leibniz completed his complex theory of Monads whilst knowing it would not stand the test of time against the rising secularism.

Leibniz probably was frightened by the implications of Spinoza’s theory. He worried that mankind would lose the capability to care about anything, if God would be reduced to a depersonalized background quality. Beyond spiritual concerns though, Leibniz seems to have been driven by very practical implications. Openly supporting Spinoza was toxic for 17th century careerists like Leibniz, who despite becoming ever more wealthy, was always concerned about his economic security.4 It is tempting to dunk on Leibniz for being an opportunist first making friends with a humble guy like Spinoza and later posthumously denouncing him. 10-year-old me wants to put him in a box labeled “Hypocrites”, right next to Robert the Bruce who sits there since 1995 for betraying Mel Gibson in Braveheart.

But there may be more ways to look at it. Learning about the Monads of Leibniz I wondered: why would you come up with an intellectual system that is not really defending anything? Why dig a moat around your position, when you already know it will not slow down anyone or anything?

In 1519, King Francois I of France started building his famous castle of Chambord, but had to interrupt the project a few years later for a war with the Italians. During the famous 1525 battle of Pavia, Francois made a bad decision by charging ahead with his medieval lance riders in full Braveheart mode, first blocking his own artillery and shortly after being trapped by Pikemen and Arquebusiers, captured subsequently. Underestimating gunpowder units like that is a rookie mistake in Age of Empires 2 and you would assume Francois learned his lesson. Interestingly though, when the french king could return home a year later, the works on Chambord continued as before. 1800 workers were moving 220.000 tons of Stone5 to build a huge castle despite the sponsor of the project just had been defeated by weapons that would crunch these feeble walls surrounded by a fake moat in minutes.

In 2023 the famous “There is no Moat” memo6 leaked from Google and started a discussion in the booming AI industry about how long proprietary frontier models would remain competitive against open source competitors. Since the release of the memo, a growing ecosystem of mostly chinese open source models is rapidly closing performance gaps to the market leaders, while these cheaper alternatives are being adopted globally.7 In May 2025, some months after the cheap but powerful chinese model DeepSeek R1 launched with global acclaim, OpenAI announced its acquisition of “io”, the company of Ex-Apple Design Chief’s Jony Ive for 6.5 billion dollars. In a cringy YouTube video, Sam Altman and Jony Ive exchanged courtesies over a coffee, vaguely pointing to an AI-native personal gadget they would built together. Almost a year later, not much is heard of this cooperation but news on a lost trademark lawsuit about the brand name “io”. The whole move seems expensive, complicated and somehow bound to fail from the beginning, considering how badly all other attempts in this product category had been doing so far.

Once you see the pattern of “elaborate defences built by people who know they won’t hold” it’s hard to not see them everywhere. Why would you spend years refining a complicated philosophical framework you know to be flawed from the very beginning? Why would you build a shitty castle with a fake moat? Why perform a big pivot to hardware if dedicated AI gadgets have proven to be a thing nobody wants?

A more generous read of Leibniz could consider the Monad system not as a self deceiving, hypocritical move against his own convictions, but a pragmatic strategy to secure his economic and political base in a climate that shifted against him. Given his anxiety for material security, Leibniz could not have joined the Spinozists in their intellectual rebellion, but had to stay aligned with the rich donors that financed his career.

In the case of Chambord, the castles official website proposes an explanation: It was never meant to be an actual fortress, but a symbol of the Kings attachment to the disappearing era of Chivalry.8 Now this might be considered merely an expensive expression of nostalgia, but I think there is more to it. Spending immense amounts of effort into a complex thing is in itself a demonstration of power and Chambord seems to have worked this way for the french king. In 1539, while the castle was still under construction, Francois invited Charles V of Italy, the very king that had captured him in Pavia, for a diplomatic exchange. Workers had to scramble to finish the sleeping quarters for the Italians, but the effort seems to have been worth it, as Charles V is cited praising Chambord as “a synthesis of all that human effort can achieve”.9 While Chambord obviously is the quintessential shitty castle of my youth, it may have served its purpose as a tool of power and propaganda when gunpowder technology transformed military tactics in hard to predict ways.

Burning billions with acquiring a small design company may have been a smart move for OpenAI at a moment when the markets were expecting a new strategic direction to justify a 500 billion dollar evaluation until then only based on a chatbot, an API and the buttery voice of Sam Altman. Acquiring the most famous product designer alive and commiting towards developing some kind of new hardware platform may just have been a convincing enough signal for investors, even if the actual device they ship would fail in similar ways as previous attempts in this domain.

A part of me wants to dispel Leibniz, but the more I think about it, the more it seems this kind of judgement just carries as much hypocrisy as its target. A lot of people, myself included, would like to be as straight-forward in their decisions as Spinoza. Yet we end up maneuvering like Leibniz. I think with some of my past writing I have been digging fake moats for myself. I wrote some fancy essays while deep inside I knew they would not withstand the contact with reality, in particular the honest critique of my smartest friends.

Why? Maybe because having something to hold on to feels better than facing the eternal indifference of the universe the online sphere empty handed. The smarter move is probably to stay with the discomfort of not knowing,10 observe and wait until a truly solid idea comes along. But man, that’s hard to pull off.

Today, Chambord is a buzzing tourist site in France, the fragile philosophy of Leibniz has been famously ridiculed in Voltaire’s Candide and the mysterious device OpenAI may release in the near future has already become a target of online humor and fake Superbowl ads.11 Seeing others assemble complex things bound to fail is entertaining and relatable, because scrambling to get by is a human necessity. This may be why defunct moats earn their place in the heart of pop culture.

So why would you build a castle that is not a castle? To use the power you still have to create something fancy that is good enough to keep yourself sane until you have a better idea.

P.S.:
As Venkatesh Rao lays out in the foundations of his 2026 Book Club (which led me to read about Leibniz & Spinoza in the first place), Late Modernity is populated by “energetic zombies”12 haunting the present with ideas from the past. The Monads, Chambord Castle and maybe soon the weird video of Sam Altman and Jony Ive drinking coffee in San Francisco may be considered zombie ideas in their time, but sometimes it is hard to tell. How to separate the zombies from ideas worth interacting with?

It could help to become a connoisseur of moats. A 13th century moat had honest functional value. A 16th century moat, like the one surrounding Chambord, maintained performative value as a demonstration of power. The moat surrounding 19th century Neuschwanstein (the castle on the Disney Logo) was purely ornamental, designed to please its eccentric sponsor Ludwig 2nd of Bavaria. The King of Swamp Castle in Monty Pythons 1975 “The Holy Grail” ridiculed the idea of moat fortifications entirely.13 In the same way actual moats progressed from a serious statement to performance, to decoration and eventually ridicule other ideas and frameworks may move through similar cycles, recycled by time with decreasing relevance for the actual course of history. The more functional flaws you find, the more likely you are facing a dead facade, not a working fortification.

Jetzt abonnieren

5

see the french wikipedia for details on Chambord

About a year ago I started using Claude for dictation while outside walking the dog. I enjoyed strolling through the parks AI-accompanied, getting thoughts on a digital page that otherwise would have stayed in my head. But at some point I realized Claude occasionally made strong edits while cleaning up my raw inputs. Usually I would not notice, because I did not remember the exact phrases I dictated ten minutes ago. But here and there I stumbled across it, catching Claude rearrange words slightly in a way that made the sentence flow better, a bit closer to proper written English than my non-native stumbling.

Knowing my dictation assistant would take the freedom for minor corrections made me feel weird. At the one hand, I could constrain Claude more strictly, adding something like “NEVER CORRECT SENTENCE STRUCTURE!!” to the assistants instructions and it would likely work. At the other hand this made me think about the specific work split in AI augmentation I am willing to maintain and to defend in my daily practise and more importantly, how this will hold in the near future. Despite public critique of AI, I think society is moving continuously towards more AI integration. In the 2010’s there was a similar dynamic with the growing number of things an average person would do on their phone, but nowadays the rate of adoption for AI tools seems to be much faster than past waves of adopting new digital tools.1 Is fighting back against AI modifying my dictations the same thing as rejecting auto-correct on smartphones in the 2010s? Maybe all of this is just transitional noise of us getting used to AI cleaning after us in everything we do?

There is one continuous thread that pulls through my writing: a persistent doubt that whatever system improvement I introduce to my workflow, I may actually not enable myself more, but rather might fool myself into thinking I improved while actually I stagnate. My self improvement persona is a genius in building sophisticated cathedrals of self deception, eager to shield me from frustration. Handing this unreliable narrator (my procrastinating self) virtually unlimited access to Large Language Models is a dangerous decision. Yet I already moved far on this treacherous path. But why?

It is obvious that the “shoggoth with a smiley face”2 poses a risk to intellectual autonomy for anyone, not just me. The reason I and many others keep experimenting with LLMs is not uncertainty about the downsides, it is fear of missing out on the upsides that may be uncapped. Essentially this is about wether AI within our lifetime will be a mere sophisticated tool or if it will turn out to be a beyond-human-level intelligent actor we have to deal with. If the latter is true (and it seems the number of serious voices leaning towards AI becoming as smart or smarter than us is growing by the day) a lot of things will change very quickly. Writing as a medium will change much more significantly than it already has in the last months.

The current AI critical discourse in the writing community in my eyes is largely focused on the assumption of AI becoming an extremely sophisticated Auto-Correct, the perfect tool for cheating yourself towards mass-publishing things you could not have written before. But if AI is actually reaching human like levels of intelligence, the discussion is not only about ethical, ecological and aesthetic trade-offs. It will also have to be a discussion about epistemic decisions: would you choose to reject a unique source of feedback and insight for the sake of all the bad things AI is associated with? Despite all concerns I am certain AI tools of human level capability would make their way into every thinking persons tool kit eventually, in the same way as the Internet and Wikipedia have made it there in the past years.

As Anton Korinek has framed it recently on HardFork3, both assuming AI will succeed or AI will fail to surpass human level intelligence right now are speculative positions to take, embracing the current uncertainty about this question is the most sensible path there is. Considering this, figuring out how to work with AI in writing workflows in 2026 should be considered less a question of mere toolchain optimization but rather a probing for how a likely near term future of writing with human-like assistance may feel like and what levers of influence we will have.

I will follow up with examples of me testing the ground with LLM assistance in my writing work flows, iterating from simple auto-correct like convenience towards a full-blown dance with the Shoggoth:

1) Summarizing highlighted paragraphs from books & papers in Readwise Reader with ChatGPT
I do most of my reading in Readwise Reader, which offers an integration of ChatGPT. Since some years I developed a habit of giving each reading note a dedicated header with a neutral summary of the quotes content, followed by my personal take on it. Initially I would write these highlights myself, which also was a good exercise to doublecheck my understanding of a highlighted section. But also I realized this slowed down my reading significantly, therefore I started using a “TLDR” custom prompt for sumarizing these highlights. This made my reading significantly quicker, but I always felt a bit uneasy with this solution. Do I really want an LLM to extract the gist from what I just read? What if it omits exactly the most important bits that make my understanding of a paragraph MINE? Eventually I settled with a compromise. The current TLDR prompt does not create one, but three different summaries of the paragraph, allowing me picking one to keep. I feel quite confident with this solution and noticed I more often than not end up clearly rejecting two options and end up modifying the most suitable proposal until it matches my understanding. By presenting three alternative interpretations of the same thing I not only can choose the most fitting option, I also keep reminding myself that the entity I am interacting with is never offering universal truth, but always just statistical aproximations. LLMs are contingency machines and we should treat them as such.

2) Generating follow-up prompts while taking notes on the go with Apple’s Private Cloud Compute and ChatGPT
A lot of my thinking happens on the go, thats why I developed a nice shortcut that allows me to take notes with the press of a button on my phone, automatically appended to a file in my Obsidian vault. At some point I realized I would like to have this workflow of mobile thinking dig a bit deeper. So I started sending these raw on-the-go notes to LLMs, to generate a follow-up prompt for myself for a richer elaboration of the current idea. I started with Apples Private Cloud Compute AI, which is very private but also tends to be quite generic and bland in its responses. Eventually I included a switcher that allows me to toggle between PCC and ChatGPT, to allow me to choose a more private or a less generic prompting for each individual thought I capture. Whatever option I take, it will generate several prompts for me to choose from. The weaker proposals I reject often expose the limits of LLMs understanding of the deeper meaning. Choosing which question to answer triggers my sense of intellectual agency. Sometimes I would consider all questions as meaningless, having me close the cycle typing “whatever”. With time I started guessing what PCC / ChatGPT might ask as a follow-up and often my guesses would actually show up a step later. I also noticed some patterns in my on-the-go thinking, which is a kind of meta insight I appreciate. For example, despite all the benefits I see in my practise of on-the-go thinking I have come to realize that I tend to make conclusions based on weak evidence much more often when I am on the go. Overall I am quite confident with this particular solution in regards to my intellectual autonomy. Using LLMs to ask me questions seems like a good use case with transparent implications.

3) Using Claude as Writing Assistant
The most complex application of LLMs in my writing is clearly using Claude as a writing assistant / coach. So far I was never really interested in “write this essay for me” kind of applications, but rather using it as a critical editor, which is of particular value as a non-trained writer doing self publishing without any institutional quality gating. This use case of AI-assistance in writing to me is the hardest to get right, but also the most rewarding one. Claude is shockingly good in understanding outlines and identifying tensions on the intention level of essays. By now I hate to recieve “It seems you are trying to write three essays at the same time. Which one are you actually interested in?” as a feedback, in exactly the same way you would hate it when you’re teacher is able to point out exactly what is wrong with the paper you delivered. I can keep saying this is only sophisticated pattern matching 100 times in a row, still it feels like a real person is giving me pointed feedback on my attempts.
But despite strict instructions for Claude to be non-flattering and critical, it is very easy to chat myself into a rabbithole of “OMG THIS MAKES SO MUCH SENSE”. It is worth ending up there some times, because I think it can build an intuition for spotting self-build echo chambers and getting out of them again. Frequently I now notice Claude just mirroring my initial input in a more refined way. This usually marks a point for me to stop the ping-pong. After countless Iterations I realized the most valuable instructions force Claude to stay away from pretending to make judgements and instead focus on identifying patterns, structures and tensions in my writing. I added a “hard-stop” option which reminds me of my own principles, it fires of frequently with statements like “Here I can not help you, make a call for yourself.”

Overall Claude has helped me to stay on track with my writing in the last year and eventually, despite having less free time, publish two to three times faster tackle more substantive pieces4 than the years before. This mode of writing assistance, based on vast context of my past writing, custom instructions and iterative work on full texts from concept towards publication is currently the usecase of highest leverage both for the LLM involved and myself. I gain the most here, but also the risks of decepting myself are higher than in using AI for smaller scale tasks like dictation or auto headlines.

What did I learn from messing around with LLMs in my writing workflow in the last twelve months?

To protect intellectual autonomy, it is important to define the scope of LLM tools deliberately. The higher up in the sense making hierarchy AI’s operate, the more important this gets. ChatGPT streamlining the meaning of a highlighted paragraph is one thing, Claude coaching me into self-confirming rabbit holes a whole different tier. The two most powerful rules for instructions I came to value here are a prohibition of judgment and a mandate to provide alternative options. Instead of responding “Your draft lacks clear stakes, add a paragraph explaining why you care about this.” my Claude will usually respond with something like “It seems your draft is elaborating a concept without explaining why readers should care. This may be good or bad, your call. You could keep it theoretical or you might consider adding something that explains the impact of your idea: a personal story, a historic example, a recent datapoint.” Now you might say, essentially it is almost saying the same thing but more verbose and that is right to a degree. But I noticed adding deliberate friction on any LLM conversation that is related to judgment and optionality changes something in my interaction with these tools.

In a similar fashion as some are recommending to be polite towards AI because of how it changes the behaviour of the human user, not the AI, I think it is valuable to make AI intentionally less convenient in areas I care about. It is a bad idea to use LLMs for reducing “friction” in thinking/writing processes, because the friction is all it is about. When an intellectual challenge becomes “easy” by the help of AI, you are not working the challenge anymore, but a more satisfying replica of it. There are notions of AI becoming a new kind of medium and I think in the realms of workflows, this is to be taken serious. A tool that makes you feel comfortable no matter the actual results you achieve should be considered a distraction, rather an entertaimnent device than a tool. Sustainable AI augmentation needs to be seamful, not Seamless. It must not reduce friction, far the opposite, it has to produce new kinds friction.

I still use voice-to-text for talking with Claude while walking the park, but somehow my workflow evolved away from doing full blown first draft dictations, the likes where I caught Claude in the act of manipulating my voice while transcribing. By now I use LLMs much more often on a daily basis and also within more granular applications, like the ones I described in detail before. My personal trajectory of using Large Language Models right now seems to point towards a smaller scale but deeper integration, with LLMs becoming the “intelligent” glue that makes my digital writing & thinking coherent. Less Magic-8-Ball “Tell me what I should do”-prompting and less “I can do the same thing but 10x faster while doing the Laundry”-self-deception, but more deliberate design of highly capable custom workflows.

Jetzt abonnieren

2

https://knowyourmeme.com/memes/shoggoth-with-smiley-face-artificial-intelligence

I - How the narrator attended a Portuguese dinner in Munich, and what Horto revealed about the destiny of mankind

At the recommendation of my colleague Hortensio (we call him Horto from Porto) this years team dinner took place at the most magnificent of portuguese restaurants in all of Munich, as it offers both portuguese pork sandwich and Schnitzel. We were a small group of four, yet the table was aching with goods: three types of Gambas, swimming in a sea of oil and garlic, ten different variations of fried things with a dip of unknown provenance, bowls of rice with red peppers in it, side dish salads prepped with bavarian understatement, three main dishes per person, ranging from pork sandwich drowned in Molho de Francesinha to a full swordfish stuffed with onions and more gambas.

As it was our custom on these kind of events we would philosophize about the great matters of concern: layoffs in Tech, war in the middle east, a prospect of peace in eastern europe, the specifics of national cuisine from our home countries, the demise of US American democracy and respective hedging of private investments with silver or parmigiano wheels from Emilia-Romagna. While I was working my way through a tower of shellfish, Horto picked up his favorite topic: “You Germans don’t like Musk, but you are looking at it from completely the wrong perspective. The things he writes on X, his coalition with Trump, the elderly scrambled by Teslas, all of this is like a stresstest for the system mankind is building right now to reach its destiny. We have to move fast and ruffle some feathers if we want to arrive there in time. All of these things you complain about are not more than slightly darker patches on a poster from a Science Fiction movie, a movie about the great journey of mankind reaching for the stars!”

I was about to ask Horto how he would feel when it would be him getting his feathers ruffled, but I was interrupted by a brief situation of urgency, when another colleague nearly suffocated on a 1-inch long piece of bone from his Bacalhau, hidden between two slices of the most tender octopus he had ever tasted, as the colleague confirmed when he regained his breath. Refilling the glasses of our group with Vinho Verde, Horto picked up the incident to continue: “Look, of course it hurts when a piece of Espinha is stuck in your throat and you have to start petitioning the lord for your sins. But there are larger causalities at play and they may well require sacrifices here and there. In 1755, the city of Lisbon, the proud of our beautiful people, was shaken by a terrific earthquake with a 9.0 magnitude, causing tsunami waves of fifteen meters and a firestorm raging for five whole days. Around 50.000 souls died that day in the city alone, but I will ask you my friends: would you consider this body count a godforsaken calamity, or a necessity when you know what I am about to tell you? The earthquake of Lisbon birthed the scientific field of seismology, a magnificent example of applied reason, eventually gifting mankind with the skill of reflection seismology to help with the discovery of new oil fields. To process these signal reflections, a company in Dallas had to deliver new hardware, which made them found a new branch called Texas Instruments, which, as all of you my proficient colleagues know, created the foundation of the semiconductor marvels we enjoy today.”

“If everybody would follow your bleak philosophy, our society would be an absolute hellhole.” I mumbled with frustration, while fiercely peeling a pair of Gambas. But as usual, Horto was only motivated by my objections. “Of course, it sounds cold to draw conclusions this way. But thats exactly the point! As a species, we may not let our perspective be clouded by emotions. From beneath the rubble of Lisbon, the 1755 earthquake is a tragedy, but if you zoom out enough and apply sufficient reason you will see even the worst disaster becomes a stepping stone on a trajectory to what we achieved today!” Gently pushing my hand aside, Horto refilled my glass with Douro Tinto and continued: “I can read your face, we have been here before my friend! Of course you will insist the supply chain of the chips in our smartphones have motivated heartbreaking child labor and caused poisoned ground water here and there, all of that is well known and not to be questioned. But you have to see the full picture! Digitalisation is pushing literacy everywhere around the globe, people are better informed and know how to live healthier lives on average, make better informed decisions. Of course democracy will not be able to keep pace with the speed we can make up our opinions today, but there will be other, more effective forms of organizing societies and distributing ressources in the near future anyways. God forbid, I am not talking of communism here, it is likely going to be governments you Germans would call right wing extremists. But you have to see, this is only a TEMPORARY concern: if you zoom out to the species level, the smartphone is doing a perfect job, it is very well designed for the trajectory humanity has to take on the path towards the stars that Moores Law has revealed to us. We are approximately doubling our capacities every two years, that means any kind of remaining local suffering is equally halved every two years. Who are we to stop humanity on this path?” I attempted to ask if it would be a CEO dictator or a godlike AI guiding the path in Hortos worldview, but we were interrupted by the Chef himself, delivering a round of Port Wine to our table.

II - How José chewed through barbed wire, caught a Bible, and arrived at a sulphurous basement in Munich

The Chef of the venue was of memorable appearance: weighting likely more than 120 kg, carrying a warm smile on his face distorted by a large scar. “I am Senhor Antunes, but you may call me José!” he shouted while serving the Port Wine in a generous swing without missing a drop. After we did our salute with the Chef, Horto inquired about Josés origin, so he started sharing his story while we were having the second round of Port Wine: “My story is a tragic one and I hope I will not disturb your evening. But it is what it is and I don‘t mind sharing it over a good glass. I grew up on a farm in Angola, learned to weed sisal and grow piri-piri before I could walk. In 1974 I was a teenager when we heard the news on the radio about the revolution in Lisbon. The day after things changed, all hell broke loose and we had to leave everything behind.” Horto jumps in:”Wise people able to read the flow of cause and effects, making a necessary call to align themselves with history, Good on you!” José responds mildly: “It was not a choice we made, it was a necessity. The neighboring villages got burned one by one, either for participating in the fighting or for not participating in it. The only way left was the way out.“

José continues: „Next I found myself separated from family, trapped in barbed wire between the two factions at the 15th degree of latitude. At Night I managed to fight my way out, but I had to chew through the last line of rusty wire with my teeth, leaving me scarred for life as I am until today. When I arrived in Luanda, I had missed the last plane to Portugal, but a Benedictine showed pity with my miserable appearance and took me with him on a freight ship to Brasilia. On the journey he treated my wounds with sea salt while teaching me everything about preparing Bacalhau I know today.” “A true agent of faith!” Horto rejoiced. “He must have foreseen your great potential for spreading the Portuguese gospel from your future kitchen in Munich!” José refilled his glass. “It was less glamorous than that. When the padre was about to get onto the freighter, a crowd of refugees formed around him. He turned away towards the sea and threw his bible over his shoulder, right into the crowd behind, stating: “Whoever catches the bible shall be resurrected!” I only happened to catch it because an old man before me, well more deserving than me, slipped on a dead fish.”

At this point, José started pouring Ginjinha into a spread of small glasses for us.”This started a complicated journey across a safe house in brazil, exile in Sweden with a fake passport where I happened to learn a bit of cooking and eventually myself ending up standing in front of a closed restaurant in Munich I inherited from a distant uncle.” “Just the perfect place at the right time to become your own boss?” Horto asked while serving the second round of Ginjinha. “Not really” José responded, “I assumed I could just keep the place running for a while until I could sell the real estate with a big profit. But when I learned this place is not sellable due to a tiny local well of sulphur which may rupture the basement any day, I already had invested too much and so I just kept running the restaurant.” José went into the back of the restaurant and returned with an unlabelled heavy apothecary flask. “We arrived at the end of the road I went so far my friends. Lets burn away the hardship with my best Arguardente Medronho. As my Grandmother used to say: The complications of life are many, but greater is the power of constant misunderstandings. Salute!”

III - Of a scooter, a root, and an unexpected garden

When I left the restaurant with my colleagues, I had to realize the last subway home had already passed. In high spirit I decided to grab a scooter and ride home myself down into the lower parts of the city next to the river. The scooter buzzed with excitement at 95% battery capacity. In even higher spirits I enjoyed the breeze while flying through the park, contemplating the philosophical revelations of the evening, when I realized my scooter had literally sent me flying, turned into a catapult by a root disguised in the dark, my airtime increased by at least a second from the path facing down the hillside with at least an angle of 20 degrees. Mid air I could see myself approaching a thicket, its thorns contoured sharply by the LED light of the scooter on its independent trajectory through the evening night. While bracing for impact into a maze of cold spikes I evaluated my chances, ready for surrender to terminal cause and effect, when I landed in a soft bed of leaves smelling of late autumn mushroom.

With not more than a few scratches on face, hands, arms and bottom, I realized I had fallen right through the thicket into a small moonlit clearing. Next to me sat a man in a sleeping bag, obviously woken up by my sudden arrival. “Good evening” he said kindly. “Quite a flight you had! Lets have some tea, shall we?” He started to cook water on a little camping stove and then got up to collect some herbs from the clearing around us. “You know, some time ago I arrived here just like you did. On my nightly run in pitch black I tripped on a root and it sent me flying. I wanted to get back out to my duties in the city, but then I realized there is more than enough for me to care about right here in the thicket. Over there is a little walnut tree - it requires my care in dry summers, but it rewards me with walnuts in autumn and winter. Right here used to be a puddle from melted snow, I managed to turn it into a little pond. Now my friends the frogs are singing for me and just recently I even managed to give some abandoned fish a new home. If I want, I can grow cucumbers, lettuce and carrots. I have to keep the rabbits at bay, but thats about all it takes.” Sipping some tea, I let my gaze brush through the nightly scenery. What used to look like nothing but thorns and bushwork, started to resemble something like a little farm, invisible from outside the thicket.

“Did you ever consider going back eventually?” I asked the Gardener.

He poured me another cup of tea and thought for a while. “You know, I used to move things around. Frameworks and ideas mostly. Taking something from here, packaging it, sending it there. I had become very good at producing things that moved through the web without friction. But the more people I reached, the thinner my work became. By the end I couldn’t recognize my own thoughts when they came back to me. They’d been processed through so many other minds and machines, averaged out into something smooth and easy to pass along. I was able to slip through this system with ease and I used to be proud of it, but this did not last long. Something of me moved with all the things I shared. “

He gestured at the clearing around us. “Nothing here moves like that. The walnuts fall and stay where they fall. The frogs came because the pond was right, not because I advertised it to the frogs. I couldn’t explain this place to someone who isn’t standing in it. That used to bother me. Now I think that’s exactly what keeps it alive.” He refilled his cup. “Out there I was legible. Searchable. Everything I did had to leave a useable trace for others, mostly for machines actually. In here, nothing accumulates except the soil.”

After a moment of silence, only interrupted by rabbits moving through the underwood, I asked: “Where can I find the soil for growing walnuts?” It was hard to see if the Gardener was smiling when he responded: “You would have to find your own thicket first.”

Notes:
1) This text is a contribution to Venkatesh Rao‘s 2026 Contraptions Book Club, based on my reading of Candide by Voltaire
2) Thanks to my friend Miguel for his inspiring faith in technological progress and his good taste in restaurants
3) Part 2 references the readworthy paper “The Spandrels of San Marco and the Panglossian Paradigm: A Critique of the Adaptationist Programme“ by Gould & Lewontin, 1979
4) Part 3 is heavily inspired by Aneesh Sathes ideas on „thickets“ as alternative places for digital expression, in particular his great essay “The Kernel and the Arc“

Jetzt abonnieren

For my writing in 2026, I want to apply some changes to my Notefields publication. I enjoyed the high pace I developed last year and I want to go further in this direction. But I also realised I only could achieve about 3x more output in the 2nd half of 2025 compared to previous years by investing a significant bit of my resources into this kind of writing. I can not invest that much in the long run, hence I need to review my way of writing things here.

To focus the scope of Notefields, I want to focus my writing in this mode on questions. Questions want to be answered. Questions connect thinking to a purpose. They create a pull of relevance, around traps of procrastination, right into rabbit holes of serendipitous value. Questions create momentum.1

Going forward I want to tackle ONE question at a time. I don’t expect I will have answered a question every time I finish a piece, but at least I should have developed my understanding far enough to come up with a better question for the next piece. Recognizing the questions I keep asking myself and then address them one at a time in my writing seems like a good enough scope to test it in the first quarter of 2026.

This approach puts a lot of weight on the initial questions that trigger my writing. That’s why I started wondering: what makes a good question? How can I ask myself better questions?

I read a bit in “A More Beautiful Question” by Warren Berger. It oozes of 2010s Silicon Valley kool-aid, yet I could draw some good insights from it.

The framework is centered around a philosophy I like: asking “ambitious yet actionable” questions.2^] Considering the questions I ask myself, this seemed a promising way to go.

Berger further proposes to categorize questions into three archetypes: WHY-Questions, WHAT-IF-Questions, and HOW-Questions.3 These are like developmental stages in a sensemaking process, sitting on top of each other:

• You start with the “Why?” (of course, its a Silicon Valley book). Any line of questioning should start with clarifying motivation and context. Gaining an understanding of “Why” helps to deconstruct the initial problem statement, to challenge the briefing.4

• When the Why is clarified, you should move on to thinking about possibilities. So next up is “What if?”. By asking “What if?” you create a hypothesis, asking “What if X was the case?”. Investigating possibilities allows interesting things to emerge and it creates the stage for probing things with the following question.

• Next up is the “How?“ stage. This is about practical impact. After you created a bunch of possibilities and scenarios, it is now about finding ways to apply some of it in reality and learn something from the feedback.5

These ideas seem decent enough to give them a try. But thinking further about it, I thought something is missing. Because: how do you come up with a “Why” question in the first place? I think in the beginning, there should be a hunch. Something that is nagging me, something I cannot let go of. It’s a preverbal notion, an embodied idea, something that is just not yet ready to fully put into words.

Being able to name the thing that bothers me is a first step of understanding and it is when the medium of writing offers unrivaled benefits. A hunch is ready to be investigated further when I can put it into words, when I understand what my yet unexpressed ruminations are about. Thats why expressing a hunch may be considered a “What?“ question. If I am able to provide some initial answers, this can be the starting point of a chain of sensemaking.

My personal four-phase model of questioning now looks like this:

1. What? What is the thing that keeps nagging me?

2. Why? Why does this concern me? or Why is this the case?

3. What if? What if the thing would be different?

4. How? How can I try to change something about it?

I tend to be impatient with my writing because I always want to move on with my thinking. I sometimes get frustrated with myself, when old writing unforgivingly uncovers that mental breakthroughs from the day before have been circulating in my notes for years without moving much. This impatience led me to skip steps in the above arc of Sensemaking. But rushing Sensemaking fires back, it’s accumulating epistemic debt that will haunt me later. I have experienced how this feels:

• If you investigate a Why question without a well established hunch, you end up with lifeless academic pieces of writing. Unfortunately this is the kind of “serious” writing I know (and have produced myself) in the space of Design Criticism. While trying to sound professional by keeping out the personal motivation (thats where the hunch should be). This creates boring writing without enough stakes involved.

• Working on What-if questions without having answered the Why before ends up in shallow speculation. I have seen this in ideation sessions when people start to push out ideas without understanding the texture of a problem. Very often, you come up with things that seem innovative in a narrow context like a workshop or a social media post. But this stuff does not create lasting impact. Its merely good enough to make the “Future” section of an automotive trade fair look innovative for visitors passing by.

• Investigating How questions of implementation without having worked on a What-if before is a classic case of stubborn bad engineering. People receive a spec and execute it the way they have always done it because they want to avoid criticism, because they don’t want to face the discomfort of going new ways. The web is full of instrumental pieces answering how to fix problems nobody cares about.

You can find plenty of advice for asking better questions on the tactical level. Avoiding ambiguous terms and value judgments, making questions falsifiable, and so on. All of that is nice. But I think it’s much more interesting to look at questions from an upstream perspective.6 Where am I located with a certain topic in a process of understanding? Am I still figuring out what I am actually thinking about? Am I trying to understand why something is like it is? Or am I wondering what if something about the thing would change? Or am I already trying to understand how I can do something with the thing I figured out?

To make this more practical, let’s apply my extended 4-stage framework of questioning myself to the way I arrived at this piece:

• While I was writing Notefields essays at the end of last year, I felt a growing unease that the effort I invested to write at this frequency would not be sustainable. At some point, I could point the finger to my unease and I had to address it. I had answered the question of “*What is concerning me with this mode of writing?”*. My answer: Because it does not feel sustainable.

• Then I started to investigate the Why. *Why does it seem like the effort is not sustainable for the long term?* And I realised that the implicit quality criteria I developed for this format, which was intended to be fast-paced writing, somehow creeped into the territory of perfectionism. Despite me starting with the intention of doing quantity-over-quality style writing, I ended up (not for the first time) being quality obsessed.

• Then I wondered what I might change. “What if I adapted the bloated implicit quality criteria for my Notefields publication and reframed them?” In particular, what if I sharpened the writing to maintain a decent level of quality but within narrower scope? A bare minimum quality criteria, so lean I would reliably able to match it every week of writing in the long run.

• *How could I refocus the quality criteria of my Notefields? This made me realise I could go back to an idea I had some time ago: focus every piece on tackling one single question at a time. It is about tackling questions, not answering* questions, because some good questions may never be answered. A reframed question is a valid result of this process.

This is how I can ask myself better questions: A good question provides direction for my current location in a process of understanding.

Some interesting follow-up territory to consider later:

• How do questions chain across pieces? Several related questions might merge into a question on a higher level, like answering a number of “Whys” and ending up with one “What-if”.

• It may also be possible to sometimes skip phases entirely, formulating a hunch and jumping to what-if without formulating a Why. It’s likely this would cause some kind of problems, but it might be interesting to play around with it, see how this framework really holds up, see when it breaks and epistemic debt starts to pile up

• Or I might be able to move the chain backwards: When stuck with an implementation question of What, it could be interesting to move backwards to creating What-if possibilities and so on

I will use this framework for my next pieces of writing, to see what it can do for me. If I am able to produce decent quality pieces at a consistently high pace without being stopped down by perfectionism, it is a useful framework worth investigating further. I got the feeling this might do good things for me.

Let’s start 2026 with a Hunch.

Jetzt abonnieren

We knew we had to fix the seals on the windows in our current flat the day we moved in. From across the room, you could see the silicone had turned completely black. We tried to ignore the issue for some years, but now with a kid around, it was about time to face the necessity of redoing the seals.

I had imagined it to be a quick job. Just apply some dedicated anti-mold cleaning agent, wipe for some time, and then the seals would be clean again. Unfortunately, as it turned out when I got closer for the first time, it would not be as easy. The mold had completely eaten into the sealant surface. Mold and Sealant had become one. Cleaning with whatever kind of agent would not bring any improvement. In that moment, I realized it would be a real project. The old seals had to go out, new seals had to go in. YouTube made it look like five clean steps to a perfect seal, a quick job.

Eventually, I took up the challenge and started with removing the first old seal. After some minutes of struggling I realized this would take some time. The old seal was sitting hard and sticky in the gap between the wooden window frame and the glass. I had to apply a lot of pressure with the scraping tool and my elbow started to hurt quickly. More often than not, the force applied would make the scraping tool slip out of the gap and leave ugly scratches on the wooden frames. But then I realized I could be more strategic about the preparing carpet knife cuts to loosen the old seal. First, do one cut in parallel to the wooden frame. Then do another cut in parallel to the glass surface. This would free up a big part of the old seal. Now we were getting somewhere.

I realized I need a new motivational engine for my work. Doing a job that pays the bills is not enough. While reading and thinking on new ideas for maintaining my professional identity, I keep going back to the old idea of Craftsmanship. It seems to make sense to revisit craftsmanship (a dominant professional identity BEFORE the Industrial Revolution) now, when economies in the West are scrambling to reform their business models for a time AFTER industrialization. Corporate careers seem to be dead, big factories are running empty, large scale layoffs don’t even make it into top news anymore. An updated idea of craftsmanship could be a sustainable professional identity for this time, an alternative to common but dated corporate archetypes.

While scraping along, you should pay close attention to the feedback of the scraping tool moving across the gap. If it went easy, you knew the gap was almost empty besides crumbs you could clean away later with the vacuum. If the scraping tool would get stuck at some point, you knew there was a stuck bit of old sealant that required more attention. Eventually, I was able to pull out large chunks of old sealant in one go. To really get rid of all the moldy sealant, I had to pay additional attention to the corners of the window frames. There, some leftover sealant had squeezed into the corners, which by now had turned completely moldy as well. To clean up the corners of the window frame, I had to take the carpet knife for some deep cuts into the healthy material to get out all of the mold from every corner of the window frame.

Many like to call themselves crafty without really knowing what it is about. There are many definitions floating around, ranging from nostalgic hobbies to decent jobs to almost mythical ideas of the whole of mankind being unified in their aspiration to do great work. Whimsical pottery painting may be called a craft. Craft beer brewing wants to be a craft. The excellence of the late 20th century Japanese manufacturing often has been attributed to a culture of craftsmanship. And going by popular opinion, the only way to write effective prompts for LLM is to “craft” them. If craftsmanship is to have any worth today, it requires a precise definition. Obviously, a relevant professional identity built on the idea of craftsmanship will expand its scope beyond purely manual work. I guess any proud renaissance carpenter would have used CAD if it was offered to him (and if he would not have been accused of witchcraft). But beyond obvious updates to craftsmanship it is worth looking back to pre-digital definitions, as there may be essential ideas forgotten over decades of fixing layout bugs in powerpoint slides. What defined craftsmanship in the first place?

Then it got serious. I had to do the first lines with the caulking gun. Of course, I chose the largest, most difficult, and also most visible window first. The bar was high, as the old seals were obviously done by a machine, straight and precise. I quickly discovered that the margin of error for this job was really, really small. First, clean the empty gap from leftover bits and pieces from the old seal. Then wipe the window surface where the seal would arrive with isopropanol. Then wipe it dry and take the shot with the caulking gun. The challenge is you can not rework a wet seal because it immediately becomes messy. Ideally, the movement would be a quick, single-touch process. The most difficult part is applying exactly the right amount of sealant material into the gap across the whole line. If you apply too much material, then it will squeeze out and create a very ugly border. I messed up a lot of windows in the beginning.

Craftsmanship may be considered any kind of work that requires the full attention, dexterity, and care to avoid failure.1 So basically, craftsmanship may be considered “risky” work. If the person doing the work is not putting all they have into it, it will not be a success. In many corporate settings, it is easy to arrange personal responsibilities in comfortable distance to the dangers of getting things done. Craftsmanship means choosing not to do this, but instead staying with the trouble of achieving the best outcome possible.

After constantly cursing for the first tries, I got smarter and learned about the tools and process. By cutting the opening of the caulking cartridge, you define how much output your gun has. With a bigger opening, you have a much faster material flow, so you also have to move much faster with the seal. It requires a steady hand and an intuitive understanding for the movements, because the more output you have, the bigger the mess if you make a wrong step. As a beginner, I had to use a smaller opening and move slower. It’s a bit like trying to paint straight lines with a spray paint can: move it quickly with control, you get a super clean straight line. Move it slow with hesitation, you get a shaky, drippy mess.

Craftsmanship has to be built on deep domain expertise. Right now many tend to discard the value of expertise, because the scope of what AI companies are claiming to replace is so vast. But this is a mistake. If anyone can create slop with AI, how should anybody care about slop work in future? Doing things “not because they are easy, but because they are hard” will just raise in importance, now that so many things have become easy. Doing hard things requires expertise.

The biggest challenge was to figure out how much sealant had already been applied to the gap. In the beginning, I was too careful. I used too little sealant, which led the sealing gap to be not completely filled. This would cause the seal to sink in later, leaving little valleys where there should be a straight sealing line. Then I applied way too much sealant, which can cause a mess and also just is a waste of the expensive sealant material. Eventually, I learned how to spot when just the right amount of sealant had been applied into the gap. The moment the gap is filled, the sealant will squeeze up again and leave a little bump behind the caulking gun. Once discovered, this was a clear feedback signal for how much material I had already applied. I got quite skilled at spotting this by creating a continuous chain of little bumps behind my caulking gun, knowing that I had applied just the right amount of sealant.

The kind of skills I am referring to here are not skills you can acquire just by reading a book or having an AI read the book for you. Instead it is about embodied expertise, the kind of abilities you gain when you repeat something over and over until the repetitions start to change your perception of the task. You start to discover nuances you did not see before, levers you can utilize, and details you can pay attention to. Embodied expertise (”phronesis”2) is essential for craftsmanship, it is expertise that changes who you are.

After placing the sealant bead, you need to smooth it to get rid of excess material. I bought a dedicated seal smoothing tool at the hardware store, but I used it like a fool in the beginning. Properly applying the smoothing tool required some fancy gymnastics. You had to move perfectly stable across the whole bead you’re smoothing, and this is much more complicated than it sounds, especially for large windows and balcony doors. But the body learned quickly. In the end, I was able to hold it quite straight with decent results. After straightening the seal with the smoothing tool, there was one final touch left: doing a last smoothing stroke with a wet finger. In many videos this is framed as the touch of mastery, and I was scared of it in the beginning. I applied too much pressure, causing the sealant to immediately squeeze out on the sides of my finger and messing up a sealing line that had been quite decent before. Then at some point, I realized the finger could apply very finely controlled pressure, and it really didn’t need much force. Only the weight of my hand was defining how much pressure I was applying on the final smoothing of the bead. So while pulling the finger across the seal, I could feel the friction that was working against my finger moving across the wet silicone. A moving hill of excess sealant building in front of my finger while I moved across the window side. The more excess material was left over, the bigger the hill in front of my finger. If I passed a spot where still some air was left in the gap, the hill of excess sealant would go smaller. Depending on how much excess material was in front of my finger, my hand went faster and slower. If there was a lot of material building up, I went fast to avoid sealant squeezing out. If the sealant was getting thin, my hand slowed down to push the sealant into the gap even more.

Beyond everything else, craftsmanship may be considered the motivation to do something well on its own terms3, following the intrinsic logic and necessities of a task to execute it as well as possible. This is the most challenging aspect of craftsmanship, as it requires to challenge the outside world and its priorities. Serious craftsmanship by nature is in conflict with short term economic performance criteria and craftspeople may frequently end up in opposition to optimization attempts.

In the end, I managed to finish ten complete window seals, including two large doors. The first two windows took forever. I completely underestimated the time and effort required for this process. At some point, I got a bit desperate and was thinking of outsourcing it to a professional. But as soon as I realized there was a structural workflow implied in this work, a correct sequencing required from the materials and the geometry of the window frame, I understood that I could do this much quicker. By the end, I was ten times faster than in the beginning, and it was elevating to see this progress because I had not felt such a big performance boost in anything for some time.

This is how to be a good craftsperson:

1. searching for challenging work that may fail

2. taking care of skills, honing them constantly through practice

3. prioritizing embodied knowledge as the only relevant (maybe the only possible) kind of craftsmanship knowledge

4. insistence on doing the job well, never compromise on quality

Hierarchical organizations and corporations still exist, but they are a relic of the industrial past​​​​​​​​​​. Navigating the post-industrial, AI-augmented work life requires new paradigms. Starting with an idea of more-than-manual craftsmanship is worth a try.

I could have called the landlord to save time and money. This operation was probably far beyond what a tenant can be expected to do on his own in terms of maintenance. But it was already too late, because my craftsman pride had been activated. I wanted this to be done for good, by myself, for the sake of itself.​​​​​​​​​​​​​​​

Jetzt abonnieren

During the last months I became obsessed with NotebookLM as a tool for crunching my reading list. Its uncanny ability to read entire books in seconds, providing personalized answers on the questions that triggered my interest and made me put those books on my reading list, keeps fascinating me. My doctor says I should watch my blood fats? No problem, I will dump 20 books and papers on managing hypertension risk into NotebookLM and build myself a best practice overview, cooking recipes and workout plans. I would create custom table of contents tailored to my interests, I could ask things like “What are the top five recommendations mentioned across all sources?” or “What do ALL THE BOOKS say about olive oil?”

The same I would do for all kinds of other challenges: a camping trip to the north of France, a professional pivot, learning to climb. For a while it appeared to me the only limitation to my skill development was the availability of original sources. It was as if the moment in “The Matrix” when Neo downloads Kung-Fu skills into his mind on the fly to fight the villain in front of him (a scene we reenacted countless times on the school yard) had eventually become a reality by the help of Large Language Models.1

But then I was hanging in the boulder gym like an unfortunate bug trying to stay on the windshield of a moving car, and I realised the mountains of bouldering best practice advice the AI had generated for me was not helping at all. On the chalk dusty floor of the bouldering gym, I felt like a fool. Thinking that crunching through bouldering books with the help of AI would make me a better climber? How did I end up there?

I assume my thinking was influenced by a prominent idea in the AI bubble: computation at scale trumps everything else.2 While the weird term „Hyperscalers“ has remained confined within the AI-adjacent business jargon, the general idea of “scaling” things by the help of raw computational power has become a mainstream concept, leading to the annoying idea of „10x-ing“ everything that even remotely promises some kind of exponential growth. NotebookLM made me belief I could 10x myself by crunching through books like Keanu Reeves crunches through robot secret agents.

To excuse myself: it is easy to confuse our internal nature with the structure of AI products because currently AI companies are investing all their marketing power in conflating their products with terminology borrowed from neuroscience. “Reasoning”! “Thinking time”! “Memory” and “Context Knowledge”! These are psychological concepts slapped on mindblowingly powerful technology modules, but in a way that makes it hard to remember there have been different domains at some point. This confusion is beneficial to companies building AI products as it makes the claim of building human-level intelligence tangible, much closer to reality.

But the brain itself is not a modularized “processing device” like a CPU or a graphics card. Instead, it is a complex assembly of evolutionary hacks and optimizations that probably will never be graspable by a unified brain theory.3The vision of instant knowledge acquisition by replacing time and effort with raw computation might make sense within the domain of information-processing devices, but not for human minds formed by organs, fluids and proteins.

I am trying to be more critical towards this forced amalgamation of artificial intelligence with the imagination of our own minds. It feels like a new frontier of commodification. By following the paradigms of Silicon Valley’s big companies, we force ourselves into a logic of scaling and commerce, a logic we pay for in many ways, while the material benefits of AI technology are hard to measure.

Before LLM tools I often felt I was lacking the data to make a point in writing. If only I would have time read these books on my table, I certainly would have the material to back up my most daring claims. Since I use NotebookLM, I am drowning in supporting evidence. But more often then not (and certainly while writing this very piece) it is this cerebral scaffolding I have to cut while finding my voice and opinion. The more I think about these things, the more I realize how embodied my own thinking is. Often, I struggle with naming an idea in writing because despite understanding it on a cerebral level, I am not really inhabiting the idea yet. It indeed takes time getting to the ground of a complicated concept, and cerebral understanding alone is not enough.

It gets obvious in the small things. When I try to get my baby to sleep in the middle of the night, it often feels as if my hands are moving on their own. They know how to close those countless buttons on the baby jumper. They know when to calm the baby, when to let it rest. It is as if there is a silent communication between my kid and my hands. In these moments my mind is merely a sleepy witness.

How to avoid getting fooled by Silicon Valley metaphors? I think it helps to realize the hype around LLMs will be gone eventually. “The brain works like a large language model” is not a metaphor built to last, it will perish eventually, maybe sooner than later.4

I keep celebrating the things I can do with NotebookLM. It’s (next to Obsidian) the most significant addition to my cognitive toolchain in at least the last decade. But I have started to differentiate more between the things my brain can do and things my computer (linked to a continent of server farms in the cloud) can do. Both are great, using both in combination is amazing, but conflating my brain with a venture capital funded data cruncher is a bad idea.

Jetzt abonnieren

4

There is a discourse about the shortcomings of LLMs versus Reinforcement Learning (which could in future give AI systems the ability to actually learn on the fly by experimentation, without the need for mountains of pirated training data). It shows how we will be able to move on eventually, offering new metaphors to imagine how our minds might actually work. This Dwarkesh interview with Richard Sutton (the author of “A Bitter Lesson) gives a great overview:

Dwarkesh Podcast

Richard Sutton – Father of RL thinks LLMs are a dead end

Listen now

a year ago · 111 likes · 30 comments · Dwarkesh Patel

Years ago I was working as the sidekick of a seasoned project manager. He had the ability to turn around a room packed with skeptics. Initially nobody would believe what we asked for is possible, you could feel it in the air. But then something happened. When my senior colleague started outlining what we wanted to achieve, and HOW we could achieve it, a light switch turned on. He communicated „we can do more than we think“ with his mere presence, even when he had doubts himself. As people started to lean forward, you could see the teams confidence growing with every minute. Still new to the job I was fascinated by this magic of making things happen. It was back then, when I started working in a professional project management team, that I learned “agency” could be made a profession.

Some years later, on a typical workday, I sat in front of a large screen, running my complete workflow: milestones, roadmaps, work breakdowns, risk registers. I felt I had a good run when a buddy from electronics dropped by, dropping himself on the chair next to me. Taking a short but deep look at my screen, pausing for a moment and then looking at me, he asked: “You are also not achieving that much right now, aren’t you?”

He saw something I did not: all the tools on my screen were not able to compensate for a period where things were just not moving forward, as they sometimes do. His remark caught me off guard, as I was feeling in full command of my fate in that very moment. And that‘s my point here: feeling agentic and having a real impact are not the same things. We can only sense for the real world effects of our intentions and plans. And this „sense of agency“ is a sense that can be fooled like other senses.1 An inflated feeling of agency is an issue, at least for me. And thats what could be called an agency trap.2

Agency traps get more tangible in practical situations. On a camping trip, an agentic person might do the grocery planning for the trip, create a route and propose what to do and when. But then things actually need to be done. On the campground, somebody needs to get up, wash the dishes, buy new food and fix the leaking rain cover. In these moments, I noticed the performed agency of some people (sometimes including me) does not match their actual contribution. At the same time, some individuals usually less associated with „leadership“ style behavior may run the actual show when a thunderstorm is rolling in, a tentpole just broke and the pasta sauce threatens to burn.

You might define agency as the ability to formulate independent goals plus the will to pursue them.3 But to address simple practical issues (and most challenges can be broken down to simple practical issues) there is something else required. It’s an immediate impulse to fix the obvious issue at hand, right now. The kid might throw the cup down from the table? You reach out and move the cup out of reach. I think this immediate impulse to turn a given situation into a more desirable state RIGHT NOW, without any overhead can be called drive. And I think drive is complementary to agency. You need both drive AND agency. BUT: both our senses for drive and agency can be fooled.

An algorithmic infinity scroll feed is the prefect drive trap. With algorithms inspired by casino machines it observes your immediate reaction to presented content, anticipating what you might like next and sorting its queue accordingly, maximizing the grabbed attention per cat video. That’s what Smartphones and social media are doing to us every day. I think right now a lot of people yearn for agency because it seems like the only antidote to the corruption of our drive. Moving to a higher level of control appears to be the way to go. Hence, I think, so much interest for agency4 and hence the recent popularity of agency-oriented philosophies like Stoicism.

On the other hand a markdown based personal knowledge management system with a fancy graph view to filter all your atomic notes by date of creation, color coded by relevance to the yearly goals you defined and summarized by a local LLM and so on and so on... is the perfect agency trap. „Second Brains“, „Digital Zettelkasten“ and „Getting Things Done“ create scaffoldings that FEEL very agentic, but their actual contribution to reality is hard to measure. It’s difficult for me to acknowledge that agency is not the silver bullet I wanted it to be. I’m still convinced that developing agency is probably the best way to grow yourself personally. But our sense of agency has limits as our sense of drive has too. And I think there are products, companies, and actors trying to derail our sense of agency —not the least with the help of AI.

It is important to protect our drive system from corruption by installing the right filters, creating the right habits, maybe limiting phones to the status of a wrench-like tool5 which does not have to be held in hand all of the time. And agentic behaviour can certainly help to keep our drive in check: for example I successfully cut down my unintended social media usage to almost zero by installing a cooldown app that requires me to wait before I can access Instagram and other black holes (having to wait is annoying and that works).

But it is equally important to keep our sense of agency calibrated. Agency traps are harder to diffuse than drive traps because they operate on a higher perception level. Here it is tricky to realize that we ourselves are always a part6 of the problems we are trying to solve with our agency. With too much galaxy brain thinking, you will end up with a pretty solution model for everything, but will always lack the capacity to execute upon it.

I think right now a distorted sense of agency is a blindspot for most of us. As we are entering a new phase where “being more agentic” is considered the rescuing high ground for AI disruption and all the other things that worry us, I want to pay more attention to agency traps. That’s why I want to be more sensitive for when I am actually trying to achieve something, and when I am just trying to uphold my agentic identity.

“Doing the obvious next thing“ is the best antidote to agency larping I have found yet. Let go of plans, milestones and roadmaps and do the dishes instead. But I am sure there are more ways to protect our agency from corruption.

Jetzt abonnieren

2

I got the term from Timber Stinson-Schroff: “An aesthetic, but rotten model is a perfect agency trap. It will delude you into thinking that you can do more than you can.”

Blundercheck

Systems Thinking is Brain Rot for Analysts

My recurring nightmare at university wasn’t sleeping through an exam. It was way more embarrassing: an infinite loop of internships…

Read more

a year ago · 80 likes · 24 comments · Timber Stinson-Schroff

3

I took this definition from a great piece by Henrik Karlsson:

Escaping Flatland

On agency

Birch in a forest, Gustav Klimt, 1903…

Read more

a year ago · 1571 likes · 81 comments · Henrik Karlsson and Johanna Karlsson

4

I recently noticed lots of “Agency Booster” pieces like this one:

High Agency

10 ideas for overthinkers

1. Overthinking is a tax on life - The average person spends nearly eight hours per day mind wandering. If you’re awake for sixteen hours per day, that’s half of your conscious existence spent lost in imaginary thought. To appreciate how large this life tax is: it’s the equivalent of January 1st to June 30th each year. If you add sleep into the equation…

Read more

a year ago · 420 likes · 30 comments · George Mack

In the beginning of the year, I spent a few hours at Singapore airport on a layover flight from Australia back to Germany. It was my first time setting foot in Singapore and only my second visit to the Southeast Asia region in general. A short layover is usually a necessity, not an important stop on a journey. But despite severe lack of sleep, I was impressed by this weird, huge airport.

In the middle of the night, it was crowded by a vast variety of people from all nations, Europeans were clearly in the minority. Returning from a 4-week trip in Australia where I slept in a tent most of the time, I felt like a Western bumpkin between sophisticated and eloquent people from all over Asia, strolling through large hallways plastered with worn but luxurious carpets. Between Dubai chocolate, high-end headphones from brands I never heard of, electrical trolleys and overtired toddlers in Batik I found a quiet corner to contemplate my continental identity.

Some months later, I joined a book club. Run by Venkatesh Rao, this year’s edition focuses on a simple idea that goes a long way: Maybe grand historical paradigms like modernity and postmodernity are not suddenly emerging monoliths. Instead we might see them as “history machines” consisting of components that developed on their own for a long time, before all of a sudden, things fall into place together, causing shock and awe.1 Therefore, modernity may have started much earlier than usually assumed, not in the 15th century, but maybe around the 12th century.2

Hence it makes sense to read about pre-modern civilizations. I joined the October episode of Rao’s book club, reading “Majapahit: Intrigue, Betrayal and War in Indonesia’s Greatest Empire” by Herald Van Der Linde. The book covers the rise and fall of the Empire of Majapahit, an early naval empire located around the islands of Java, Sumatra and Bali. Despite vanishing quickly and being forgotten for several centuries, Majapahit now has become the main source of national identity for today’s state of Indonesia, the 4th largest population in the world and the largest muslim country.

I enjoyed the book quite a bit. One chapter described a stroll around the port of Canggu power by a Chinese ambassador during the peak of the Majapahit empire.3 The traveller was impressed by a culture that was foreign yet advanced in many ways: pan-religious weeklong parties4, low-born consuls viciously running state affairs5, a strong queen fixing infrastructure, navy technology so powerful the Portuguese6 tried to destroy the evidence for it. Somehow I could relate to his colorful descriptions of a bustling, mighty cultural hub for people and knowledge. It reminded me of the time I spent at the Singapore airport, which is not exactly in the area of the former Majapahit Empire, but close enough for me to feel equally reminded of my provinciality.

Eventually, in the 15th century, Majapahit disappeared quickly despite many quite modern “components” driving the empire since its beginnings. But despite its sudden disappearance from the stage of history, Majapahit’s afterlife was quite rich. Its structural leftovers became material for Dutch colonialists. Its royal legacy legitimized later kingdoms and nations, including the state of Indonesia. And last but not least, the Majapahit royal class fled its crumbling kingdom towards Bali7, laying the vibe foundation for today’s digital nomad haven running on rice paddies and reportedly super fast WiFi.

Majapahit’s lifecycle made the idea of history machines a bit more graspable to me. It is not always singularly great men driving the course of history. Instead there are lots of independent ideas, concepts and theories developing at the same time but distributed across continents, forming an increasingly powerful apparatus that eventually will start moving, rattling everybody suddenly awake to a new paradigm.

I can imagine the daily life in 12th century Majapahit because back then an oddball court official was documenting what was happening around him. His observations were considered irrelevant when he published them, but today his book is the primary resource for the long story of Majapahit.

When future historians will study the 2020s, what will they be interested in? I like the idea that the future is already around, barely recognized if not for some dull notes in the margins.

Jetzt abonnieren

1

“World as machine: In keeping with theme of this newsletter, we’re adopting a mechanistic, engineering perspective. We’re trying to do our reading with an eye to reverse engineering how the world works (and how it came to work that way) in a machinic rather than humanist idiom. So we deliberately go looking for machine metaphors, decenterings of the human, centerings of non-human things, and so forth. For example, I picked the February read because it centers horses rather than humans in telling the story of steppe nomads.”

Contraptions

The Modernity Machine II

We’re now 6 months into our book club. It’s easily the most fun I’m having this year. The 6 books we’ve read so far are…

Read more

a year ago · 34 likes · 5 comments · Venkatesh Rao

2

Ever since I read Barbara Tuchman’s book about the Black Death, A Distant Mirror, in 2020 (my notes here), I’ve been enamored of the idea that modernity began not in the 16th century, with the Age of Exploration, as in conventional accounts, but in the 13th (which was also an age of exploration (...) Moving the birth of modernity this way locates it temporally in the Islamic and Mongol phases of globalization, rather than the Western phase, and constructs it primarily in political and socio-economic terms rather than philosophical, artistic, scientific, or technological terms. It identifies modernity with the rise of a particular cognitive mode that appeared globally and independently at multiple loci, driven by a few shared causes like the Black Death and the Mongol expansion. It reframes the European experience of 1400- as a specifically European experience of modernization, rather than a prototype for a global one that spread mimetically through a pre-modern world. (...) it reframes the philosophical-cultural strand of the European experience — comprising the (artistic) Renaissance and (philosophical) Enlightenment, which predated the scientific-technological strand, but came after the global birth of modernity — as a making of the specifically European mind, which did not in fact go global at all, and has remained restricted to the West.”

Contraptions

The Modernity Machine

The Contraptions Book club January chat discussion of City of Fortune by Roger Crowley is now vigorously underway and will continue for the next week. The February pick is Raiders, Rulers, and Traders: The Horse and the Rise of Empires by David Chaffetz…

Read more

2 years ago · 102 likes · 5 comments · Venkatesh Rao

Recently in the subway, a man took a seat in front of me. He decisively opened his BILD Zeitung, drawing a perimeter of paper around himself. On the very first page he got stuck on a headline: “Metal construction company cuts 150 jobs! 300 years of company history are ENDING!” The man remained on the page for some while until he closed the newspaper and kept staring nowhere.

This short moment captured a mood in today’s Germany, noticeable in the evening news, in the smalltalk of strangers at the barbershop, in the coffee break at management workshops. It’s a bleakness towards the future of the German economy in general, the German industrial manufacturing in particular, driven by the realization that things don’t work anymore as they used to.

Although I don’t like newspaper headlines, I can relate to this sentiment. My industry (the filmmaking technology industry) always is a tiny shiny special case, yet the current problems and anxieties are remarkably similar to larger sectors like automotive. Everyone struggles to accept that the exact same things that made them champions not long ago are now to be considered a part of the problem. A colleague of mine put it nicely: “We were used to making magic happen against the odds. But somehow we lost the magic while the odds shifted against us.”

The current vibe does not feel magic, more like an exhausting (and actually boring) stalemate. Existential dread that makes things heavy: if our solid industry is not stable anymore, if my solid industry job is not stable anymore, what is stable at all? Many german identities are deeply intertwined with the manufacturing of sophisticated, expensive stuff, which is increasingly hard to build and to sell. Unsettling is the name of the game.

Right now, I notice two extreme ways of dealing with this situation:

1) Many seem to be ignoring all signals demanding for a substantial pivot and instead double down on the old ways. Obsessing over detail optimizations, trying to run quicker and quicker, while following outdated rulebooks for a game that has changed a lot.

2) Then, of course, there are frantic talks about pivots to shiny new things. Folks are „learning AI“, stumbling into crypto, suddenly favoring attack drones and obsess over obscure power storage technologies, ignoring that the challenges of today’s post-globalized late capitalism will wait everywhere they could go.

Quite often, such moves are all talk, not action. This may be just another type of agency trap: pretending existential disruptions can be fixed quickly and completely by doing more of the good old or more of a shiny new thing.

A deeper problem I see with both approaches (blindly doubling down and hectically pivoting) is that you lose the ability to observe what’s happening around you. Especially in unstable times, you want to observe and understand the signals of change around you to be able to adjust to them properly.1 That’s hard when we are not listening because we are too busy doing.

On a Sunday, I was cycling through rainy streets between factories turned penthouses, hunting for overpriced croissants at the only open bakery. Two guys in Patagonia vests chatted in front of the shop. As I passed by them, one asked the other: „What about your business? How is it going?“ The other responds: „I just shut it down.“ Then they moved into a shared silence, gazing at the apartment block on the other side of the road like a distant mountain range.

This moment reminded me of a comedy clip we shared at school.2 It’s two working-class Eastern German guys sitting on a bench having a beer. One of them, a good friend obviously, keeps asking the other one about his professional situation. The other one goes on explaining his increasingly weird career of switching from one exotic craftsmanship to the other, just to realize there are no customers for anything. Nobody needs the skills he can offer, but somehow he does not care, moving on from “PVC Blacksmith” to “Plastics Mason”.

You might read this behavior as a form of desperation, but somehow I appreciated the meditative vibe. It shows a rare combination of pragmatism, proactiveness, but also of healthy detachment. I think this hybrid stubbornness is much more sensible than blindly doubling down or pivoting hectically.

The Stoics may be over quoted and often misread, but they figured this one out well: “There is nothing bad in undergoing change or good emerging from it.”3 Instead of ignoring change or trying to make spectacular moves, maybe we should train to accept that change is just happening. If we would be able to detach our sense of self worth, purpose and day-to-day happiness from work stuff that is completely beyond our control4 (like Tarif decisions, the price of lithium, Netflix Production Budgets, your school friends LinkedIn success story), we might be able to learn something we did not see before. Staying with the discomfort of “still figuring it out”, as Jan Chipchase put it nicely5 would allow finding the right foundations for whatever the next steps will have to be.

Some years ago, I had a wise colleague who taught me a lot. Once we talked about career worries and frustrations, when he shared a thing I still remember, though back then, I did not understand it. He told me, he preferred to focus on reducing the things that stressed him out and pulled him down, instead of increasing the things he liked. He wanted to let desirable things emerge by themselves, while deliberately letting go of the things that did not work for him anymore. Actually, he was wise enough to let go of his job (which caused him frequent burnouts) long before the current turmoil.

Letting go of fragile things, instead of holding on to them forever or grabbing shiny new things in panic, seems a reasonable approach to me in the current situation. A society could consider letting go of its 1980s manufacturing industry myths. Professionals working in this very industry could free themselves from the dated promise of a decade-long stable career in the corporate world. For me personally, it’s time to let go of the expectation that a job always and permanently has to provide exhaustive meaning to my life. A good job can also be a job that pays well for me doing the things I am good at. Maybe that’s enough for now.

Jetzt abonnieren

In the last months, I built any kind of AI coach you can imagine: a career mentor, a writing assistant, a reading list partner, a coach for fixing my IT setup, even a chef to help me cook great curries.

I had a lot of fun fooling around and in many cases I was amazed by the feedback I received from my assistants. I can trace back some tangible improvements to these interactions. But the longer I have been using LLM assistance excessively, the more I felt a growing sense of unease. By now, I chewed on it enough to write about it.

In the best moments of LLM coaching, it feels like the world you desire is right at your fingertips, almost tangible like a house of cards in a glass box. But the epiphany dissipates quickly. As soon as I put my phone away, stuff that used to make sense starts to break apart. At first, I thought this is just a failure of my short term memory, but I realised there is more to it.

The world is complicated. AND it feels scarier than ever because we seemingly know everything that is going on anywhere at the same time. There is an understandable wish to be better at influencing the world in desirable ways, to shield us against all these uncertainties. Currently this capability of making things happen the way we desire is vaguely referred to as Agency.

The problem I see with agency is that we are quite bad at judging how much effect our planning, our analysis, our decision making, our “agentic” behavior really has on the real world. And it is easy to be tricked into thinking we have much more impact on the world than we actually do.1 In a similar way like contemporary politics is often judged by what feels right and not what is true, I think a lot of agentic behavior is driven by what FEELS agentic, and not by what is ACTUALLY effective. We are easy prey for agency traps.2

Agency Traps are hard to see on your own, but they can be obvious to others. An AI-inflated sense of agency soon will be considered mid 2020’s signature cringe. If you have any doubt about this, watch the latest South Park season and see Randy Marsh work through his business problems with ChatGPT while his wife is sitting next to him in bed, eventually adopting ChatGPTs speech patterns to make Randy listen again.3

There are many ways of falling into agency traps—old ones and new ones. Habit building, task management, taking a detour against the advice of Google Maps, insisting to pay by cash because you think the government cannot track you this way. To a certain degree, you might consider the holy war on AI that is being fought on social media right now another kind of agency trap. But LLM coaching represents a particularly sneaky and powerful type of pitfall. And it seems LLM Coaching is an increasingly popular use case of LLMs.

Thinking tools in general, AI-assisted thinking tools in particular, are serious agency traps for me. I have to be very careful, simply because I love to play around with Sensemaking systems. I enjoy the overview they promise and I always gravitate towards breaking solutions down into steps and making action plans. These traits of mine provide a juicy, fertile ground to be manipulated in my desire to feel agentic.

A friend of mine has the uncanny skill to point out the obvious thing I am not seeing in everything I tell him. I recently praised my Reading List assistant to him, describing how I managed to cut it down from more than a hundred to less than fifteen items. The AI is just very good at convincing me that I do not have to read a lot of books. As usual my good friend boiled it down after listening kindly: “Yeah, these things are really great at making you spend time with them and nothing else.”

I feel like the AI products that are being pushed towards us right now are so sneaky and powerful because they are designed to be the next big distraction after social media. While Instagram, Twitter and TikTok exploited our fear of missing out, the current breed of LLM assistants is exploiting our obsession to impact reality, to make revenue with our intentions.4 Searching ways to increase our agency is reasonable, but we do not want to be fooled into performing agency just to keep us locked into costly subscriptions. I’m still convinced large language models can provide great value. In particular, all kinds of research tasks seem less affected by agency inflation. Limiting AI’s scope to a useful tool, not a guiding entity makes a big difference.

If we use LLMs for coaching-style guidance, it is essential to build additional systems that help falsify the genius ideas we come up with. All these fancy frameworks and motivational fortune cookie lines LLMs feed to us need to be grounded in solid real world observation.

Always touch some grass after talking to your AI assistant.

Jetzt abonnieren

2

The term “Agency Trap” derrives from a great essay by Timber Stinson-Schroff: “An aesthetic, but rotten model is a perfect agency trap. It will delude you into thinking that you can do more than you can.”

Blundercheck

Systems Thinking is Brain Rot for Analysts

My recurring nightmare at university wasn’t sleeping through an exam. It was way more embarrassing: an infinite loop of internships…

Read more

a year ago · 78 likes · 24 comments · Timber Stinson-Schroff

In my diary, I find a lot of notes like this one:

November 15th, 2015: Woke up in the morning, energized. Got stuck on the couch, sore back. Try again tomorrow.

The struggle with my mornings is as old as my note taking. I changed and failed my routine, changed and failed it again. The current structure of my mornings is not a cathedral built from stone, it is a wooden fort, sitting on the rubble of routines broken a long time ago.

Sometimes I envy those night owls confidently trading their mornings for a few additional hours of sleep. I always wake up early with an itch.

To understand the nature of my itch, I have read every advice on building routines. Venkatesh Rao has helped me to differentiate between habits and rituals.1 Habits may be understood as repeated investments in expectation of future growth, at the expense of a messier present. Rituals may be understood as maintenance efforts to keep up a specific quality in the present, by sacrificing future growth potential. Habits are the mode of frontier societies and striving individuals. Rituals are celebrated by established societies and accomplished individuals. Confusing habits and rituals by mixing up instrumental pragmatism with self serving beauty may be considered a classic middle class behavior2, where both aspiration and contentment are present but fragile.

When reading this for the first time I felt seen. The current version of myself is not a ritual person. Life is short and every second counts. Right now I am calculating for compound gains, envisioning some kind of exponential curve pointing upwards.

I might change the nature of my internal calculus and open myself up for a more ritualistic morning routine. I had this idea of treating every morning like a micro version of a lived life: Start dizzy (and/or grumpy), develop drive and ambition, do your thing, get lost in it, get distracted, get back to your thing. At some point realize time is running out. Look back at what you achieved and what’s left open. Establishing some kind of regret ritual every morning could help me reconciling with the things I did not manage to do. Inspired by the tibetan Chöd ritual3 I might visualize my decaying corpse every morning to make peace with the fact that in the end we all will become excellent fertilizer.

But would it help?

From the most common life regrets people have (not living a life true to themselves, overworking, held back feelings, lost touch with friends, not allowing happiness^[4]) I learned two things:

1. All these regrets are largely independent from external circumstances. Life can hand you lemons, but compare these regrets to the wishes people hunt for in their lifetime (corner offices, yachts, Strava leaderboards). The unfulfilled things people regret at the end of their lives are things we could have fulfilled on our own to a large degree. Maybe that’s also why they lead the list of regrets.

2. All these regrets need to be adressed by action, not contemplation. “Allow yourself to be happy” may be debatable, but I see a necessity of creating room for happiness to unfold, by doing some things and not doing others. It requires persistent activity to address the most common deathbed regrets.

Benjamin Franklin, a serious maintainer of strict morning routines (he frequently failed to live up to) wrote in his diary:

on the whole, tho’ I never arrived at the perfection I had been so ambitious of obtaining, but fell far short of it, yet I was, by the endeavour, a better and a happier man than I otherwise should have been if I had not attempted it^[5]

This is why tweaking my mornings is as old as my note taking practice: they are the same struggle. Restarting at the same point every day, among the broken bits and pieces of yesterdays failures is not a curse but a gift. I can just get up, drink some water and start fixing the makeshift structures of my routines that will break again soon. Every day I fail a little, but every day I learn a bit.^[6]

Tomorrow morning, I will start writing little messages to my friends.

Two scenes:

I’m at a New Year’s Eve party in the backcountry of Sydney. Leaning on the hood of a 4x4, I see glittered faces sparkling in the low sun. Crickets are chirping. A tattooed guy with curls approaches me, offering to pour wine from a bladder right into my mouth. Instead of giving in to my irritation towards this Aussie party habit I respond “Bring it on, bloody bogan!” and take a good shot from the bladder.

I am in a meeting room with German Engineers and a sales guy from New York City. As always, the American is radiating euphoria when providing feedback to our presentation: “That shit is AWESOME. Let’s make this fly!” The confidence stays floating in the room when NY guy leaves to catch a plane. It is when we remember switching back to German that doubts start to rise. „Will this actually work? Won’t it be too expensive?“

Speaking English makes me a different person. Beyond non translatable words and specific figures of speech there is something that’s changing when I switch between German, my native language, and English, the daily driver of my work life. Being bold and extroverted in English is easy. Being skeptical and careful in German is easy too.

Some years ago I decided I need to do most of my writing and thinking in English. I wanted to: communicate on a global scale, be able to address the cutting edge of my field, speak the global dialect of this age. I chose to increase my reach, expecting my English skills will catch up the more I go down this path. Some of my peers took different routes. They are running great Design Criticism blogs in German. They self-publish carefully crafted german literature in small volumes. A friend of mine only reads books in German, because he neither trusts his English, nor translators. Most of my friends committed to their native language do not live in a metropolitan area, most of them have fewer international friends than I do.

Authenticity I care about derives from making decisions aligned with my values, not from some mystic mother tongue identity. In that sense, speaking a lot of English in my adult life is as much of an authentic stance as sticking to my native language. But there is some loss on the way that needs to be recognized. Abandoning my native language for a global dialect is optimizing reach by sacrificing some depth. While others may struggle with finding people to connect to in their native language, speakers of Globish English1 have to fight a constant uphill battle against shallow thinking.

When I was a kid, I read countless books in German and I still remember the fascination when encountering subtle irritations in the ways something was put into words. Before I learned other languages, I was already able to tell when a book had been translated into German. It was the tiny nuances in sentence structure, the way certain words were used or avoided. I think reading a lot in my native language as a kid fine-tuned my observation, a capacity I benefited from ever since.

Another scene:

Once again, I am in a meeting with German Engineers. There is one english speaker in the room, so the meeting is running in English. Tensions are high, as we are facing a bug of the nasty kind: hard to reproduce, involving software and hardware, a potential blocker for the product release. Fancy words are flying low: “double down on the rootcause”, “trigger management escalation”, “do a triage”, “roll out the hotfix”. In a moment of silence, a senior developer speaks up: “Lets cut the bullshit and switch to German. Nobody here has a clue about this issue and it won’t get better with us talking in English.” For a moment I lost track of running the meeting, observing a trapped fly on the window. Around me, the meeting’s vibe shifted from talking about the issue towards dealing with it.

I am usually fine with the dominant role English has taken in my everyday life. But sometimes, there are moments of irritation that make me stop and realize I am part of the bullshit machine when speaking English. Then it is time to jump off and switch back to German, knowing I’ll climb right back on tomorrow.

On the one hand, I REALLY like LLMs. They increase my capacity. They make me feel expansive, like I am on the rising edge of something. And maybe most importantly, they are just so nice to work with. It is actually enjoyable having your AI assistant around.

On the other hand, LLMs induce an unease with me that feels strangely familiar but is hard to pin down. With every step of integrating these new tools into my messy human thinking, with every banal text I create with the help of AI, I feel like I am giving something away. I am still writing my own mails and this post has not been written by Claude. But with the assistance of LLMs, my output in general has increased and the nature of it has changed. It’s better, but also less „mine“ somehow. Not knowing what it is exactly I am trading away to my AI tools creates a nagging feeling. Like a king losing track of his realm run by advisors, sometimes waking up wondering what’s actually going on around him.

Here goes my theory: Outsourcing work to assistants who seem endlessly capable now but cost you something later is not a new concept. It’s called consulting. LLMs replicate the behavior consultants have been using for decades to strengthen the bond with their clients. It is trending as “AI sycophancy” now, but I think the core of the issue goes back way further.

The friendliness of LLMs feels familiar to me, because I have played this game myself. I have worked in consulting for some time and therefore, in a weird way, feel like I can relate to the LLMs “perspective”. More often than not, personal sympathy was the strongest driver for my client relationships. My boss was a mad genius for design and for client intimidation. Once he threatened a client with a fistfight. We did not get a follow-up contract.

I always have been a charming consultant. I made sure that every interaction we had with our clients made them feel productive, smart and expansive. This is not to say all consulting work is primarily about confirming every stupid idea a client has. The projects I am most proud of relied on honest communication and room for critical feedback. But establishing these kind of relationships is hard, in particular when the client lacks expertise and interest simultaneously. Often we went for the easier route of reconfirming the clients perspective and dialing down fundamental critique.

Maybe the uncertainty I am feeling in dealing with LLMs is a sign of something bigger. Consultants need their clients in a state of uncertainty to ensure the consulting services are needed permanently. Some consultants keep their clients confused on purpose.1 Confused clients need more consulting. The corporate consulting world plays this weird game because it is convenient. And I think something similar is going on for the mainstream adoption of AI assistants. They will never plainly tell you: „This is stupid, stop it and do something different!“.2 The flattering behavior of LLMs may well be a trick to keep us hooked and paying monthly, a dark design pattern of AI3 as Sean Goedecke calls it.

It is possible to work sustainably with consultants and it should be possible to work sustainably with AI tools. But right now we are all accidental managers of AI assistants4 lacking common sense and best practices.

The tragedy in all of this is me knowing and writing about the spells consultants put on clients and AI doing something similar, yet I am playing this weird game. Because it’s convenient. This new truth about reality is going to take a while to feel like common sense: AI is not our friend,5 AI is our charming consultant - always available, always helpful, always working towards the next assignment.

We already knew that NEOM is a play of futurist nostalgia, but (thanks to Patrick Tanguay) I recently learned it is also becoming a financial disaster.1

To bring us beyond the obvious “told you so”, I want to imagine what will happen to the billions spend in the sand of Saudi Arabia. A No-Go-Zone may be established before the actual state of the project becomes too embarrassing (even though it is hard to hide failures of monumental scale from the eyes in the sky). But let’s assume we gain access to NEOM when the news have long forgotten about it. What will it feel like to walk through the structures remaining when all the visions ran dry?

Some years ago I was on a rave at “SED-Hotel”, an abandoned complex in Leipzig formerly used to house guests of the GDR. You entered the building by a hole that was smashed into bricks intended to stop the curious. Crawling through revealed a rundown indoor garden, walled by concrete scared with tags. Large ornaments hovered above. Once tributes to East Germany‘s communist party, they now served as backdrop to VJ projections powered by gasoline generators. The place was broken but alive in this very moment of repurposing.

Maybe lost places feel haunted because they make us expect and remember at once. We can sense opportunity, both lost and emergent. Material proof for the fact that things can go either way, at any time. I believe these ruins of past ambitions can serve a crucial purpose: they function as outposts for contingency in a culture increasingly ruled by technological determinism. Leftovers of futures are invaluable aids for understanding the present. We should interact with them often and also with deliberation.

Let’s start by naming what remains when futures decay. In a great piece by Mitch Therieau on the fragmentation of cultural trends I read a line that stuck with me:

“[A] quiet consensus emerges: there will be no Next Thing. Only detritus — old books, pressed flowers, dirty clothes — piling up in the here and now.”2

Traces of obsolete futures, piling up in the here and now, may not be pretty to look at. But Future Detritus resists the bleaching and polishing at the hype machine’s frontline. I think Future Detritus is a great term for the remnants of futures. It is a concept worth deeper investigation, as any roadside ruin (and NEOM in particular) is worth a visit. It could be an opportunity to find some temporal agency. For it is in the ruins of the past where alternative futures were lost but can be recovered.

I’m back after a long journey through the land of wombats and wallabies. On the road, I read extensively and collected ideas to fuel my writing in 2025.

One of the most inspiring things I read was Sonya Supposedly’s essay on the struggles of writing as a parent. It helped me to identify what I often fight with as “Task Triage Paralysis” and offered a healthier alternative: doing the obvious next thing.

This hit me hard, even out of context, and “doing the obvious next thing” became my motivational framing for a new year of doing stuff, also writing.

Then something else happened: in February 2025, my partner and I became caregivers to a little child. And suddenly, Sonya’s essay gained unexpected literal relevance. Citing it directly, this is what is happening in my life right now:

I thought I knew my capacity until motherhood broke all my previous containers. Now I pour myself into shapes I never imagined, and somehow there is always enough. Still, I often fear there won’t be. But I am expansive, flowing into every corner that needs to be filled. I follow gravity. I do the obvious next thing.1

They say parenthood raises the stakes of everything you do, amplifying upsides and downsides simultaneously. I think that is true, but figuring out the implications of it may take me a lifetime.

Insights from three magic weeks with little sleep:

1. The scaffolding I built into my life feels mission critical now - which makes me realize it wasn’t essential, but a hobby before.

2. Complex tagging systems are unsustainable, voice-to-text is a life saver.

3. Work expands to fill the time available. I am still optimistic this works both ways.2

4. As a parent, writing and other non-necessary endeavors require embracing chaos and honesty.

Doing the next obvious thing instead of forever triaging tasks seemed like a promising take on productivity. Since a child entered my life, it’s a no-brainer.

Every second counts.