Here’s a snippet from Brett’s video, with minor voice-to-text edits, and emphasis mine:

Jensen Huang once again has announced we’ve got AGI—again. Who am I to gainsay him, though? He’s the CEO of NVIDIA, the most valuable company on the planet, almost single-handedly driving the progress in the hardware on which these models are running. He himself is worth over 200 billion US dollars. He’s an electrical engineer. He’s a very smart guy.

Sam Altman, also a smart guy, in a January 2025 blog said, “we are now confident we know how to build AGI, as we have traditionally understood it.” And in July 2026, he said, “we are now like in the singularity.”

Elon Musk has made similar remarks about that as well. Altman’s also said AGI “kind of went whooshing by and that the term is not that useful because the bar keeps moving.”

What bar?

Apparently, we’ve met the bar, we’ve exceeded the bar. And yet, isn’t the bar supposed to be the thing that tells us whether or not we’ve got AGI?

Who’s setting this bar?

It wasn’t any of us who were speaking from the philosophical side of things. The philosophically minded people on this have not set bars in the way the tech people do.

The tech people have, want, desire, think it’s possible to have quantifiable metrics when it comes to this. Why?

Well, all these announcements essentially amount to marketing. It’s like Altman saying from time to time as he does that his model can solve all of physics or is soon about to solve all of physics. Or when Musk says Grok is smarter than any PhD and he’s got an IQ of 300 or something.

Huang and others are speaking to their shareholders.

[…]

But if you’re interested in reality without the hype, the actual philosophy and science of this, then you have to go beyond these benchmarks.

The benchmarks and tests exist to be passed so that CEOs and others have something to report. Progress. They have a quantifiable metric of progress. […] But as for saying it’s a metric for or a benchmark for AGI, that’s another thing.

It’s all to the good to be able to talk about how much progress we’re having and how fast it’s going on. That gives us a little bit of an indication about how far we’ve come in terms of technology and wealth creation and all the great advantages that are coming to us as the users. But when it comes to the philosophy, never mind the financials of these big tech companies, benchmarks cannot matter a jot.

The term AGI as used by tech CEOs has not only become detached from the original definition, it’s become effectively meaningless. It no longer represents a mind with the same general — universal — explanatory ability as ours but instantiated on silicon instead of grey matter. The AGI tech CEOs tout means a machine capable of doing remarkably many different things remarkably well, according to whichever definition of many, well, and remarkably serves their purpose. It’s hot air. It’s something they say to signal: we’re doing great.

It’s not surprising when CEOs go around hyping up their products, especially in today’s media environment. Companies need marketing to attract new customers and, crucially, new investors. As users who benefit from LLMs and AI agents, we should appreciate the contribution of the investors who financed LLM and AI agent development as much as the researchers and engineers who built them.

So why take to the keyboard and rant about it? Isn’t this just CEOs being CEOs?

The thing is, there’s a difference between Apple-style, self-aggrandizing PR releases and the incessant misdirection and bullshitting we see from AI companies, boosted by a media industry interested only in sensationalist headlines. It’s one thing to introduce the latest incremental improvement as our best iPhone ever with great fanfare. It’s another to claim your product can solve all of physics! That’s intellectually dishonest. We should have no patience for that.

The hype, and its companion, constant fearmongering, might taint the idea of AGI at large in the same way that environmentalists tainted nuclear energy. We’re already seeing politicians jumping on the anti-AI bandwagon.

But AI research and development is useful. And AGI research even more so, not because we necessarily need more intelligence (although we do), but because to build AGI we first need to understand GI: general intelligence. That is, ourselves.

“What if the world had one GPU per person?” That’s the question the GPU World writing challenge is asking the internet. And with a 1,000–5,000 word scope and a $100,000 prize pool, imagining answers is both accessible and appealing.

The assignment: Today’s access to intelligence is capped by the short supply of GPUs. What if AI development never went beyond 2026 levels, but hardware became so cheap that every person in the world had their own dedicated GPU by 2040?

Implicit in the brief is the belief that more intelligence is good. That is, I think, correct. LLMs have unlocked improvements in coding, math proof formalization, and video rendering, to name just a few. It’s exciting to imagine what we could do with more, cheaper access to those capabilities.

So, sure, humanity needs more intelligence. But it needs something else even more.

Consider a technology as revolutionary as LLMs: the internet. We’ve had access to this groundbreaking technology for the past 30-odd years, with an estimated 6 billion people connected today. It enabled new businesses and products, new ways to work and new forms of entertainment. And with the advent of the smartphone, people have been able to carry the internet everywhere they go.

These technologies brought huge improvements, many of which life-changing — we wouldn’t be interacting right now without them. But when I think about the fact that billions of people have in their pocket a machine orders of magnitude more powerful than the computers that won the space race, with instant access to information, communication, and commerce, and then I look at the world around us, I wonder why all this power hasn’t translated to more freedom and abundance. As J. Storrs Hall asks, where is my flying car?

The technological capabilities are there, but there seems to be a bottleneck to how much value we can actually create with them. My guess is that the bottleneck is culture.

Culture is its own kind of technology. It’s the set of memes1 that guide how we behave and interact.

Maybe the biologist E.O. Wilson had a point when he said, “the real problem of humanity is the following: we have paleolithic emotions; medieval institutions; and god-like technology.”

Making progress in the realm of atoms and bits seems to be easier than making progress in how we relate to each other and ourselves. But that’s not surprising: atoms and bits are inert and predictable; people are creative and unpredictable. In fact, it’s thanks to creativity that we reached “god-like” technology despite our limitations. It’s creativity that allows us to overcome our emotions and improve our institutions.

Sure, we need more GPUs. We need more technological improvements of all kinds. But the one technology that we need to improve the most is our culture.

We need more of that cultural technology that gave us the Enlightenment, the West, and the scientific revolution. Without it, intelligence will end up like the internet: more and more people will get access to it, but the collective won’t benefit as much.

Karl Popper identified criticism, the rejection of dogma and search for errors in current explanations, as the crucial ingredient that made the scientific revolution and the Enlightenment possible, and with them, eventually, the institutional innovations that allowed people to invent the internet and GPUs. He traced it back to the pre-Socratic philosophers of Ancient Greece, such as Xenophanes, with his inquiry into the limits of certainty.

It’s through criticism and fallibilism, through looking for errors and trying to fix them, that we can improve our culture and institutions. And improve they must, if we want them to support progress in a world of 280-character tweets, 30-second TikToks, and sycophantic personal AIs.

So I’d like to propose a companion writing challenge inspired by GPU World, unfortunately with no prize pool, but hopefully just as thought-provoking. Let’s call it Enlightenment World.

Here’s the brief:

The Enlightenment revolution has only just begun to affect humanity, and billions of humans have yet to so much as experience living in an open society that values criticism and error detection and correction. This is in large part because anti-rational memes are easy to generate but hard to eradicate: humanity is culture-poor.

This could increase, however, as creative people show how conjecture and criticism generate progress and inspire others. Someday, the values of the Enlightenment might be part of every human being’s operating system and of the societies in which they live.

What would this world be like?

What will our world be like if our institutions, at every scale, private and public, engage in a search for truth instead of status, driven by criticism instead of tribalism? How far could personal liberties reach? Will education be revolutionized by encouraging individual pursuits? Will social media operate on an engine other than engagement? What developments will arise from across the globe?

The tradition of criticism that created the world we live in today holds the potential to be far more transformative than any of the individual technologies it produced.

I invite you to imagine this enlightened future.

I often type “who” instead of “how.” As I was fixing that typo in a sentence today, I realized that the replacement of who with how is a thread that connects many ideas that resonate with me.

Take Nullius in verba, the Royal Society’s Latin motto translating to “Take nobody’s word for it.” At a time when society was still ordered by rank, a group of elites began to see themselves as “experimental philosophers” and chose a motto stating that, at least when it came to science, how a claim could be demonstrated was more important than who pronounced it.

Later on, Karl Popper made the switch explicit in the field of politics. The question, he argued, is not “Who should rule?” but “How can we develop our political institutions in such a manner that even incompetent and dishonest rulers cannot do too much harm?”1

Naval Ravikant’s observation that truth-seekers take feedback from reality, while consensus-seekers from other people is another instance of the difference between following the who and focusing on the how.

That’s not to say that “who” should always be ignored. When you trust a “who”, you are using them as a proxy for the “how” behind their claim. Even the Royal Society ran on trust between its gentleman members, but that trust wasn’t an innate characteristic. It had to be earned and could be lost; it was based on a track record of detailed experiments, public trials, and witnesses. A “who” goes bad when they are given authority that cannot be revoked: when the “how” behind it becomes immune to criticism.

Progress happens when we move from who to how: from authority to good explanations. But beware. Nothing guarantees the process will continue in one direction only. The ideological capture of editorial boards at various scientific journals and academic institutions is proof of that: a “how” that can no longer be criticized. The replication crisis is the result of a system that over-indexed on the “who” with not enough care for the “how” behind it. Balaji Srinivasan’s exhortation to value independent replication over prestigious citation offers a way out, or, rather, a way back to the original values promoted by the Royal Society.

Adopting a how-over-who attitude offers no certainties; it will only show you where you were wrong, again and again. That’s how we make progress. The work is never over, and it starts every day with revoking the authority of the loudest “who” of all: our ego.

Imagine two companies, Company A and Company B. They have the same budget and the same deadline. They have teams comparable in size and skill. They are building on the same tech stack. The only difference: A builds in secret and will ship only on the deadline; B builds in the open, ships to testers, and iterates on what it learns.

If you had to bet on which will deliver the better product, which would you choose?

I would pick B. I think B works in a way that increases the opportunities for learning, and that will lead to a better product.

Software development is at its core a process of knowledge creation. Knowledge of the problem, the domain, the users. Code matters, it matters a lot, but only as the way to materialize the knowledge into a usable solution.

If you want to build better software, better products, it’s therefore useful to learn about how knowledge is created.

The 20th-century Austrian philosopher Karl Popper gave a compelling explanation of how knowledge grows: through conjecture and refutation.

Popper’s main claim is that all knowledge is conjectural. There is no authority or solid ground on which to justify knowledge. All we have are conjectures, explanations, guesses. But that is no cause for despair. We don’t need justifications to make progress, we just need to correct the errors in our guesses. The engine of the whole process is criticism.

People notice problems in the world and guess at solutions. That’s the conjecture step. The proposed solutions are then scrutinized through severe, strict, and honest criticism. That’s the refutation step. If a solution survives, it is adopted tentatively.

This process, Popper found, never ends. A new solution introduces new problems or reveals problems that were hidden before. That’s how progress happens.1

If software development is knowledge creation, and knowledge grows through criticism, then the more we criticize the theories behind the choices in our software and their implementation in code, the more chances we have to discover and correct errors.

This is why techniques like test-driven development, continuous integration, continuous delivery, and feature flags lead to better products. They are all ways to criticize, to identify errors, in our software. TDD and CI accelerate error detection in the developer’s understanding of how the code runs. Is the code doing what I expected it to? CD and feature flags accelerate error detection in the developer’s understanding of how users interact with the product. Is the code doing something worth doing?

Both Company A and Company B could be working using TDD, CI, and other practices for fast feedback on whether the code is correct. But only Company B exposes its early ideas to real world criticism, giving itself many chances to correct its errors. That’s the difference that makes Company B the best bet.

Notice, however, that there is no guarantee of success. Critical errors may remain hidden from Company B, or they might fail to create solutions for the ones they know about.

Popper offers no certainties. In fact, he shows how nothing is certain. What he does is shift the focus from “I have to get it right” to “I have to find and fix errors.”

Let’s go discover some errors.

When the first edition shipped in 2021, we wrote tests with XCTest and asynchronous code with Combine. But the Swift landscape is always evolving and improving. Between Swift Testing and async await, the way we write and test code has changed. This edition is a full revision to match.

What’s new

Swift Testing throughout. Every test in the book is now written with Swift Testing and making the most of its natural language support. But fear not, if your codebase still runs on XCTest, a new appendix holds a completed XCTest reference, so you can bridge the book’s lessons to your older tests.

Structured concurrency. The networking chapters have been rewritten with async/await. The old AnyPublisher<T, Error> signatures are now async throws -> T, and the tests await results directly instead of using cumbersome XCTestExpectation or sink.

A refreshed sample app. Alberto’s menu-ordering app now uses the latest SwiftUI APIs and look and feel. Chapter 7 keeps @Published/ObservableObject as the teaching path and adds a sidebar on where @Observable fits.

Get it

Grab a copy from Apress or Amazon, and pick up the free bonus material while you’re at it.

If you already own the first edition

The Test-Driven Development philosophy hasn’t changed, because it was never tied to a framework. The menu-ordering app, the Test Doubles (Stub, Spy, Fake, Dummy), and the red/green/refactor rhythm are all still here.

What changed is the how: Swift Testing, structured concurrency, modern SwiftUI. If you’re moving a codebase onto those, the second edition is a guided tour of the new patterns in context.

Questions about the new edition? Leave a comment below, find me on X, or at gio@mokacoding.com.

The crucial why? and how? questions are seldom asked. And when they are, they are in the form of “why is this bad?” and “how bad will this be?” Never “why are you saying this?” or “how is this going to work?”

Granted, someone working at Anthropic can provide better insight on upcoming innovations than your average Gio writing from his kitchen table. But there is a glaring conflict of interest at play that keeps being conveniently ignored in favor of sensationalist reporting.

Anthropic, OpenAI, and all the other players in the AI field are yet to turn a profit. These companies depend on massive influxes of cash from investors. You know what’s a great way to get investment? Hype!

Whenever a staffer in one of these companies makes grand declarations on the future of the industry, one cannot help but wonder how much of it is informed guess and how much is PR work.

But if one can forgive podcast hosts for inviting partisan or biased voices—after all, we are all biased in one way or another—what keeps happening is that they don’t push back. They don’t ask how does it work?

Take the claim that developers at Anthropic use Claude to improve Claude, and that this will result in the much dreaded exponential recursive self-improvement loop.

Ezra takes this as a given, without asking how that might happen.

When you dig beneath the surface, you find the claim of imminent AI takeover rests on flimsy foundations.

Claude Code, the tool, uses Claude, the model, let’s say Opus, to write code. Recently, the code Opus produces and the autonomy with which Claude Code integrates it have become so good that many developers use it every day, myself included.

But Opus, GPT, Gemini, and all the other large language models are not code. LLMs are mathematics.

For sure, Anthropic employees may use Claude Code to improve Claude, but only at the margins. They can make the apps we use to interact with their models faster or more secure. Likewise, they can improve all the tooling, infrastructure, and scaffolding that goes into training new models. All of that is welcome and remarkable, but comes from existing solutions already present in the models’ gargantuan training sets. Code writing agents cannot create the new explanations that are necessary for fundamentally better models and alternative AI architectures—see Vishal Misra’s explanation and Brett Hall’s commentary if you want to dig deeper.

I wish Ezra and his colleagues deployed the same relentless criticism toward AI as they do with politicians they disagree with. All this handwaving press about AI doom distracts from the real conversations we need to have: How to use these tools well, how to prevent their makers from harvesting our data in their chase for growth, and how to make the underlying economics scalable.

Claude Code and Codex have been great unlocks in my workflow. I also spun up an AI assistant that reminds me of my kids’ school schedule, monitors topics online, and is “learning” how to open issues to report its own bugs, then open PRs to fix them.

I can’t imagine going back to a world without these tools. And this is scaring me, because the current AI bonanza may not last.

The tokens we are burning for serious work and recreational projects are much more expensive than the prices companies are offering. We might be within a golden window of subsidized experience.

OpenAI and Anthropic internal documents report they are years away from being profitable. The introduction of ads in ChatGPT is essentially an admission that there are not enough paying customers to sustain the business in its current state. It’s a common pattern for tech startups: operate at a loss focusing on growth alone, gather as many users as possible, then start harvesting them by increasing prices without improving the experience. Is the future of AI one where the need to generate returns can be met only by squeezing the customers for all their worth?

Of course, one cannot predict the future. A lot of smart people are innovating on all fronts. The operational cost may go down significantly and new products might enter the scene creating paths for profitability that don’t require enshittification.

Regardless of where the industry will go, or perhaps especially because of this uncertainty, this is the best time to use AI. Let’s get as much leverage out of it as we can. It may not last.

At a time when anti-progress thought is rampant, we need more voices like Arjun’s and Logan’s. Authors who can argue with straight logic against nihilism and in favor of human prosperity.

Here are some passages that stood out.

Humanity finally kicked into high gear during the Enlightenment, when we realized that progress was both possible and achievable, when ideas that fostered creativity and criticism began to replace those that suppressed them, when we sought to explain the world around us with rigorous theories, both scientific and otherwise.

The pursuit of knowledge is an egalitarian enterprise—whether one is rich or poor, male or female, slave or king, no one’s ideas enjoyed privilege over another’s for any reason other than that they contained superior arguments.

The growth of knowledge is the fundamental driver of progress, the primary weapon in the fight against our problems. With this understanding in mind, we can see in clearer terms precisely why all of the ideologies we discussed are, in fact, Enemies of Civilization: They slow the growth of knowledge and wealth (wealth being the set of all transformations we know how to cause).

For most of our history, our cultures were static. Then, with the philosophical advances of the Enlightenment, the West figured out how to make continuous progress—it became a dynamic society. While the Enemies of Civilization had dominated static societies of the past, they continue to hamper progress to this day. All of them fail to appreciate that problems are due to lack of knowledge—and, therefore, that speed, creativity, and freedom are necessary for progress, rather than political and intellectual tyranny, reducing resource consumption, and ridding the Earth of humanity.

You may laugh at those environmentalists who throw paint at art, but they’ve been effective at halting the development of nuclear power, a potential source of abundant energy that we’ve known how to build for decades. We can’t calculate how much suffering could have been ameliorated had we been free to build nuclear power plants across the Earth.

Relativism might seem open-minded and fair, but it is neither. For it is not open to the possibility that one party is in the right, the other in the wrong.

If left unchecked, [environmentalism, relativism, and other anti-rational memes] could come to dominate our dynamic society and revert it back to the static societies of old. We therefore have a duty to not only recognize them for the threat that they are, but to do everything in our power to eradicate them entirely.

It’s taken a few centuries, but we’ve come back to the ancients’ view of the relationship between people and the cosmos. While we’ve rightly abandoned the majority of their beliefs, they were right about this much—to understand Nature at its deepest, we have to acknowledge the special role people play.

If we so choose, we can continue to make the world, the solar system, the galaxy, and the rest an infinitely better and more beautiful place. Human knowledge—our values, scientific theories, political ideals, and culture—can come to be the predominant cause of every physical structure in the cosmos.

The fate of the cosmos depends on the future history of knowledge.

I had the fortune of reading a pre-release version, then preordered the paperback as soon as it was available on Amazon. I want it where I can see it on my bookshelf, next to other inspiring works such as The Ascent of Man, The Rational Optimist, and, of course, The Beginning of Infinity, to serve as a reminder that we can all contribute to the fate of the cosmos.

I hope you’ll give Lords of the Cosmos a chance. If you do, get in touch and let me know.

This week, a personal reflection on patience, purpose, and the tension between internal and external metrics.

The last post, Depth Takes Time, was an invitation to push against the modern instinct of instant gratification and take the time to go deep.

The anecdote from Popper’s intellectual journey that prompted that reflection also made me think about patience and the purpose of writing.

Read more

The internet has given us unprecedented access to information. We have libraries at our fingertips, with tools to search and summarize them. When any question can be answered with a few clicks, it’s easy to fool ourselves into believing we understand the complex topics that make up whatever the current thing is.

From this false sense of understanding comes an entitlement to have an opinion, and, of course, to post it online. But there is a wide chasm between the thumbnail sketch one gets from a couple of internet searches and actually understanding a topic.

I was reminded that deep understanding takes time when reading Karl Popper’s intellectual autobiography, Unended Quest. Popper built his arguments by first understanding other people’s theories to the point where he could elevate them to their strongest version. Only from that vantage point, he went on to criticize them, and, at times, demolish them.

The Open Society and Its Enemies is an example of Popper’s method directed towards Platonism and Marxism. His work was so thorough that Marx’s biographer Isaiah Berlin described it as “the most scrupulous and formidable criticism of the philosophical and historical doctrines of Marxism by any living writer.”

One cannot reach such depth overnight. It requires years of reading and thinking.

Seventeen-year-old Popper encountered communism and Marxism in 1919. As he recounts in Unended Quest, he was initially suspicious but was eventually won over by the propaganda. His infatuation didn’t last long, though. When “a shooting broke out during a demonstration by unarmed young socialists who, instigated by the communists, tried to help some communists to escape who were under arrest in the central police station in Vienna” resulting in the death of several people involved, Popper was horrified. But it wasn’t just the brutality of the police that struck him. As a communist, he felt complicit because of the implications of Marxism, with its demand that class struggle intensify and become ever more violent.

From there began a long process of reflection:

It took me some years of study before I felt with any confidence that I had grasped the heart of the Marxian argument. […] But it was not till sixteen years later, in 1935, that I began to write about Marxism with the intention of publishing what I wrote. As a consequence, two books emerged between 1935 and 1943—The Poverty of Historicism and The Open Society and Its Enemies.

When we hold a book, or engage with any other work of creativity, we interact with a finished product. We can access the work in its entirety and without delay. It’s easy to gloss over the fact that it didn’t materialize all of a sudden.

We are used to getting everything we need instantly, be it a ride home with Uber or an item with Amazon, a movie with Netflix or a meal with DoorDash. This craving for instant gratification is also visible in entertainment apps like Instagram and TikTok, with their short videos that get swiped away unless they hook viewers within a few seconds.

But there is no same-day-delivery system for understanding.

Knowledge comes from within. It’s a lattice we weave one connection at a time, sequence after sequence of conjecture and criticism.

Complex ideas take time to digest. They cannot be summarized in 30-second soundbites.

Popper spent years grappling with the Marxist doctrine, and even longer thinking and discussing with friends, before sitting down to write its obituary.

To push back against reductionist narratives and bring nuance back into our discourse, we need to remember that depth takes time.

Status is given by others. Substance is earned, by yourself, for yourself.

Because status is given, it can also be taken away. No one can rob you of substance. What you’ve learned stays with you, and so do the things you create.

Higher status might open doors—your network is your net worth—but only substance gives you the tools and agency to solve your problems.

You don’t need others to tell you when you’ve done something of substance. You can just tell.

You gain status by pleasing others. You build substance by pleasing yourself.

When people stop talking, really bad stuff happens. When marriages stop talking, divorce happens. When civilizations stop talking, civil war ensues.

When you stop having a human connection with someone you disagree with, it becomes a lot easier to want to commit violence against that group.

What we as a culture need to get back to is being able to have reasonable disagreement where violence is not an option.

— Charlie Kirk, 1993 - 2025

Those words hit harder today, after Charlie Kirk’s assassination.

Some saw him as a champion of the free exchange of ideas, others as a “dishonest huckster.” Regardless, his murder is an unjustified tragedy.

Before today, I only knew him from the viral clips that made their way on my feeds from time to time. I didn’t know how popular he was or the influence of his Turning Point organization.

You cannot form an opinion about someone from a sample of clips edited for virality, but take the one from which I took the quote at the start of the post. In it, I see echoes of Popper’s description of rationalism:

Perhaps I am wrong and you are right; anyway we can both hope that after our discussion we will both see things more clearly than before, just so long as we remember that our drawing closer to the truth is more important than the question of who is right.

— On Freedom, from the All Life Is Problem Solving essays collection.

What makes Charlie’s death even more tragic is the vicious callousness of the reaction in certain corners of the internet. Sickening. Despicable. Morally rotten.

At such an emotionally charged moment, it’s crucial to remember that those celebrating are a vocal but tiny minority. Most people are recoiling, including Charlie’s political opponents.

Once again, I return to Popper: “It might be well for all of us to remember that, while differing widely in the various little bits we know, in our infinite ignorance we are all equal.”

fears Charlie’s death will be a turning point for the worse, “tonight feels like some sort of invisible line has been crossed that we didn’t even know was there.”

If this horrible death marks a turning point, my hope is that it will be a return to an honest, earnest, polite exchange of ideas. Let it be the moment we all say enough with the nonsense, and bring back the nuance.

Let’s go back to talking instead of shouting.

Rest in peace, Charlie Kirk.

According to Business Insider, Julia Liuson, president of the Microsoft developer tools division, shared a memo encouraging managers to consider AI as “part of your holistic reflections on an individual’s performance and impact.”

Microsoft is heavily invested into the AI game, so it’s no surprise they push hard for it internally. But they are not the only ones.

Shopify’s CEO Tobi Lutke considers using AI efficiently “a fundamental expectation of everyone at Shopify,” and plans to “add AI usage questions to our performance and peer review questionnaire.”

The sensationalist media has capitalized on leaders pushing AI with headlines such as “No AI, no job” but Tobi’s mandate, and I’d like to think those of most other leaders, is far more nuanced.

The e-commerce giant’s CEO wants his employees to explore what he considers a promising opportunity together. “Learning to use AI well is an unobvious skill,” Tobi writes, “My sense is that a lot of people give up after writing a prompt and not getting the ideal thing back immediately. Learning to prompt and load context is important.”

It’s one thing to advocate for AI literacy, it’s another to essentially mandate people change their workflows to introduce AI. Tobi says employees are “welcome to try” opting-out of learning how to integrate AI in their craft, but when performance reviews explicitly include questions about AI usage, what choice do they really have?

For sure, it’s a leader’s job to set the company’s technological direction. 37 Signals CEO David Heinemeier Hansson caught heat online for requiring employees to transition from macOS to Omarchy, his bespoke Linux distribution. His response to critics is that picking technologies to bet on and invest in is “literally the job description.” “Omarchy is our distribution,” David explained on the Rework Podcast, “It was born at 37 Signals, it was born from me pouring 37 Signalness into it.” Going all in on Omarchy is a bet on a development environment the company controls and can tailor to its needs, like it already does with Ruby on Rails, Kamal, Hotwire, and other non-negotiable tools.

David’s mandates at 37 Signals, from Omarchy to Ruby on Rails, set environmental constraints that focus development. Can we say the same about mandating AI use?

Leaders who push for AI adoption believe it will help with a variety of tasks. Many critics have argued the promised productivity increase has yet to materialize, but let’s assume it’s just a matter of time, a combination of tools maturing and people learning how to use them. If that’s the case, it will soon be pretty easy to identify who uses AI well: they will be the most productive teams and individuals.

If overall productivity is what businesses care about, does it matter how a person or team becomes more productive? A good employee delivers what they promised within the timeframe they agreed upon. If they can improve output quality or delivery time, does it matter if that’s done by using AI or by, say, improving their work environment to reduce distractions?

My concern is that adding AI usage to performance reviews distracts people from useful work. On top of “build great products” they now have to “prove you use AI.” This will inevitably result in a lot of theater, with folks showcasing how they use AI, instead of actually using AI to solve problems.

I get that leaders, especially at big companies, have the long-term interests of the business in mind. By pushing teams to master a new promising technology, they are laying the groundwork for remaining competitive in the future. But what we need is convincing, not coercing.

If a new tool or workflow helps people do more with less, those who are committed to their craft will naturally adopt it. The only thing they need is for someone to show the benefit and demonstrate the adoption cost is worth it. No mandates required!

Leaders who want to bet on the revolutionary effects of AI should invest in teaching employees how to use it well. Incentivize learning by rewarding those who demonstrate how integrating AI in their work made them more productive. Adding AI usage metrics to performance reviews does the opposite: It penalizes those that don’t use AI—regardless of why they don’t.

Mandates also make me question the trust leadership has in their teams and the hiring process that got them on board. Who doesn’t want to automate toil in favor of higher bandwidth for applying their creativity? Trust your teams will appreciate the benefits and do all they can to take advantage of them.

A good leader sets a bold vision for the future. A great leader trusts teams to make the best decisions to turn that vision into reality.

I benefit daily from using AI tools, but am skeptical of AI mandates. They seem a misguided way to promote AI usage. Mandates might result in AI adoption, but it will be in a performative way, one that proves compliance over generating leverage.

What about you? What’s your company’s expectation on AI usage?

The AI cheerleaders seem to be forgetting that when a technology is truly useful, valuable, and enjoyable, people will adopt it of their own free will. No one had to force anyone to buy an iPhone or to install GPS in their cars.

Mari contrasts the organic mass adoption of GPS and iPhones with the failure of AR devices, from Google Glass to Vision Pro.

Mari’s comment feels right, but one should be skeptical of feelings when forming an opinion. Is it actually true that if a technology is truly useful, people will naturally adopt it?

A few counter examples. Nuclear power is the best energy source we’ve implemented so far, yet it’s actively banned in many countries. Vaccines are among the greatest health innovations of the 20th century, yet there are vocal groups of people that reject them. Genetically modified crops produce more food with less land, yet the European Union has various regulations curtailing them.

What nuclear, vaccines, and GMOs have in common is that their opposition is in large part ideological. Clearly, doing more for less is not enough; a new technology also needs to fit into the user’s value system.

Advocates promote adoption by explaining how a new tech aligns with people’s values. At times, the advocate’s work is less about promoting the tech itself and more about changing people’s worldview. See, for example, ’s The Ecomodernist.

So, yes, technology needs advocates, especially when a new technology requires a paradigm shift. A good advocate helps bridging the gap between the status quo and the future that new technology enables.

Looking back at Mari’s observation, I guess what resonated was not the implicit critique of cheerleaders the but the proviso “of their own free will.”

Convincing is always better than coercing.

There is a huge difference between advocating from the bottom and mandating from the top.

If a technology is useful, a good advocate will be able to showcase its benefits and plant the seed for people to change their mind about rejecting it. A top-down mandate, no matter how well intentioned, does not explain why the new technology should be used.

Sometimes, we need to be shown how something is better to understand why we should use it.

The problem is not with cheerleaders, but with mandates, bureaucracy, and regulations. Technology needs advocates, not enforcers.

In a recent monologue, Ed shares his frustration at journalists and editors who have been boosting the AI narrative over the past couple of years, forfeiting their intellectual integrity and journalistic commitment to dig beneath the surface.

I share Ed’s frustration. As I wrote back in AI Does Not Need Welfare, too many journalists accept exceptional claims without asking for exceptional evidence—or explanations.

Unfortunately, the explanation-less pattern does not stop with AI. We see it at play, for example, in the reporting on the Gaza conflict, where prominent outlets share victim counts published by Hamas, a terrorist organization that boasts of using human shields and propaganda. We saw it during the final years of Biden’s presidency, when we were assured the president was fit and lucid despite it being visibly not the case.

The problem is not limited to the so-called “legacy media.” Many of the independent podcasters and writers that filled the trust void left by traditional publications are no better. They might be worse, actually. In rejecting the attitude of established outlets, some also reject rigor, fact-checking, and accountability.

For sure, I’m not accusing all journalists of having lost their way, but the recent flip on AI is a clear example of an overall preference for sensationalism and narrative in favor of truth-seeking.

I don’t pretend to have a good explanation for this complex phenomenon, but perhaps the diagnosis Neil Postman offered in his 1985 Amusing Ourselves To Death could be a good place to start.

As television became ubiquitous in the West, discourse and information morphed into entertainment. But entertainment is not an exercise in truth seeking. Television’s aim, Postman wrote, is “applause, not reflection.”

Television’s paradigm reshaped discourse across all other media. Social media, with its algorithms promoting the most engaging content with no regard for depth or correctness, fits perfectly within TV’s framework.

With a media landscape that incentivizes engagement, is it any surprise that those working in it chase emotion-inducing partisan headlines?

How off the mark do all those interviews by Ezra Klein and others on the imminent collapse of education and white-collar work look now, after GPT-5’s disappointment?

But also, how predictable. Doomerism is more engaging than grounded optimism. Nuanced discussions on trade-offs and how much work still needs to be done are plain and boring compared to stories of impending ruin and injustice, never mind how unfounded.

If you’ve been seeking explanations below the flashy headlines, only to be left wanting, then you won’t be surprised that so many news outlets got it so wrong. They haven’t been in the business of looking for truth for a long time.

They are ruining the environment with their private jets, scheming to replace us with AIs, and — worst of all — bending the knee to Donald Trump. They have a dark agenda. They are screwing us all. They should be abolished.

For many, tech billionaires, with their unbounded wealth, their excesses, and their nefarious plans are the new enemies of humanity.

Wrong!

Poverty is the enemy. Not wealth.

I’m not here to defend the current crop of tech billionaires. Think of them whatever you like. But wealth and wealth creation are good things.

Wealth creation is not a zero-sum game — it’s net positive.

Profit is proof of service. Money is a neutral indicator of value. Acquiring wealth requires creating something useful in the world.

Let people solve problems, sell their solutions, and get rich doing so. The more problems get solved, the better off we all are for it.

The issue is not that some people are incredibly wealthy, it’s that not enough people are given a chance to become wealthy.

That’s not because of the already wealthy conspiring to prevent access to their cabal. It’s a result of errors in politics and philosophy that have accumulated over time. Those errors can be corrected, little by little. Problems are soluble.

Don’t abolish billionaires. Abolish poverty!

This post is an extension of a comment I left on ’s excellent post Beggars or Billionaires: Who Would You Rather Get Rid Of?

The Popperian business consultant Bart Vanderhaegen offers an alternative and more actionable way of categorizing work. In episode 167 of his podcast, Seeking Good Explanations, Bart defines three working modes:

• Monotony

• Politics

• Problem solving

Monotony is operating on routine work. It’s executing an already proven solution. This work might generate useful output, but it can’t sustain a company in the long run. Problems are inevitable and today’s solutions are unlikely to work for tomorrow’s problems.

Politics is navigating the inevitable social dynamics that emerge when people collaborate. This mode requires creativity, but it directs it toward winning zero-sum games that improve your position or that of your tribe instead of improving the customer’s life.

Problem solving is the quintessential creative mode. It consists of generating, exploring, criticizing, and testing ideas. This mode is the opposite of performative busyness: every minute spent in it is productive. Operating in problem solving mode is the only way to progress.

The other modes are not necessarily wasteful, but the results they produce will at best make existing processes run more smoothly. That might be okay in the short term, but for a business to remain viable — and for people to feel they have an impact — progress is required.

Compared to “let your best people cook,” thinking in terms of Bart’s work modes has the advantage of being applicable to any role in an organization, regardless of how directly involved they are in creating the product. Whether in engineering or HR, design or middle management, everyone benefits from a working environment that optimizes for problem solving over monotony and politics.

Yet despite all the new tools and possibilities, many knowledge organizations stagnate. So much potential remains untapped.

One culprit is a pervasive culture of performative busyness: using visible activity as a proxy for productive work1.

A lot of time at work is spent doing theater.

• Communication Theater. Being always online and replying to messages and emails as soon as they are received to signal presence.

• Meeting Theater. Daily stand-ups and other ritualistic meetings where people take turns talking about the work they plan to do instead of actually doing the work.

• Reporting Theater. Status updates and progress reports rich in detail, screenshots, and graphs that take as long to compile as the work they describe.

• Task Theater. Bloated project management software with hyper-granular, multi-dimensional, interconnected tickets to curate like gardeners.

• Cultural Theater. Joining every culture event and strategically interacting on highly visible projects to show engagement and alignment.

None of these activities are inherently bad or completely wasteful. Teams need a way to track work in progress and spread workload, internal seminars are great learning opportunities, and asynchronous communication channels ought to be monitored. These tasks often look and feel like “real” work, but notice the difference in customer-facing progress between a day spent pruning the backlog and one spent writing code2.

For a knowledge organization to operate at its full potential, we need to minimize performative busyness in favor of actual problem solving.

Cal Newport^* has a proposal to move in this direction: Let your best people cook.

Cal starts from a Brandon Sanderson interview where the prolific author described how his company is organized around the principle “let Brandon cook.” Creating stories is the most valuable thing Sanderson can do for his business, so the whole organization is optimized for it.

Most knowledge organizations have their own Brandon Sandersons: people with high-return creative skills. When these people apply their craft, they disproportionately bring the product forward. It’s in a business’s best interest to let its high-impact people apply their skills for as long and with as few distractions as possible.

Obviously, not everyone in a company has a role that directly and disproportionately advances the product. But it seems reasonable to expect that, as we optimize for real-world progress (shipping) over performative busyness, the workflow changes involved will also benefit those with supporting roles. Alternatively, a knowledge organization with a more focused creative workforce might need far fewer auxiliary roles, and the people in those roles will have more ownership and opportunities to use their creativity.

There are so many problems waiting to be solved and we have never been in a better position to take them on. It’s truly tragic to think how much progress we are forfeiting because we’re projecting busyness instead of actually applying our creativity.

As Cal urges, “let the Brandon Sandersons of your company cook!”

But what counts as a problem?

Your house being on fire is a problem. Whether to have Italian or Indian for dinner is also a problem. Obviously, the two differ greatly in magnitude.

A house on fire is a risk-of-ruin kind of problem. What to have for dinner is a first world problem.

When thinking about problem solving and knowledge creation, I find it useful to categorize problems based on the consequences of failing to solve them.

Here’s a tentative taxonomy:

• Threats

• Challenges

• Puzzles

• Choices

• Trivialities

On one end, we have threats. These are problems that, if not solved, will have catastrophic consequences.

After threats we have challenges. Solving these problems brings an improvement, but, unlike threats, failing to solve them won’t be devastating. Examples of challenges include learning new skills, getting a promotion, and improving as an athlete.

Next, we have puzzles. These are gaps in one’s understanding, the “that’s funny” moments when reality doesn’t match your expectations and you want to know why. For a scientist, this could be the need to explain experimental results that don’t match the most advanced theory.

Choices are problems with more than one possible solution. The task is to find the best solution in context or create an entirely new and better one.

Finally, we have trivialities. A fly buzzing around your desk, cold coffee in your mug, a flat phone battery, these are examples of trivial problems with inconsequential solutions.

The categorization is somewhat subjective. The problem of the buzzing fly might be trivial for me, with my noise-canceling headphones, but far more serious for someone taking an exam who is distracted by the noise.

Of course, classifying problems this way does little in the way of solving them.

But my hope is that a richer vocabulary can be a helpful tool for prioritization and introducing novices to problem-solving via conjecture and criticism.

This week, a personal story on being stuck consuming without creating, and finding a way out by… consuming more?!

The past few weeks have been frustrating.

I had that rare kind of time off when the kids are at school. In the lead-up to it, I dreamed of getting so much done. Live like someone with the luxury of working without having to work. Write, read, code, train.

But when the time came, I couldn’t get started on anything. All those shiny ideas had lost their sparkle. Absolutely no gumption.

Read more