The Deep Feed

01 — Cal Newport

The Ruach Problem: Speech as the Last Human Bastion

Why the fluency of LLMs feels like a transgression against the sacred

By Cal Newport · 8 min read

Editor's note: As AI masters the art of conversation, we must ask if we are losing the very thing that makes us human.

The second chapter of Genesis offers a striking image: a deity forming man from dust and breathing life into his nostrils. In the Aramaic translation by the scholar Onkelos, this 'living being' is rendered as a 'speaking spirit'—ruach memalela. This is not a mere linguistic quirk. It is a fundamental claim about our species. To be human is to possess the ability to communicate through words, to alchemize internal thought into vocalised phonemes that travel across space to inhabit the mind of another. It is a form of telepathy that defines our dignity.

The Illusion of the Speaking Spirit

We are currently witnessing a technological mimicry of this sacred act. When we converse with a large language model, we encounter a lexical fluidity that feels startlingly real. We know, intellectually, that this is the result of matrix multiplications and the autoregressive prediction of tokens. Yet, the sensation remains. There is a sense of transgression in seeing a machine perform the very ritual that has historically separated the human from the object. We are granting the role of the 'speaking spirit' to a mathematical construct.

Speech is constitutive of what it means to be a human – a core part of our humanity is our ability to communicate with words.

This discomfort is not irrational. If speech is the mechanism by which we express our soul, what happens when that mechanism is decoupled from a soul? We are entering an era where we might use AI to write our letters, speak our condolences, or provide companionship to the lonely. This is the 'golem' problem: a creature that mimics the form of life but lacks the essence. We are essentially outsourcing the most intimate part of our existence to a statistical engine.

The Ethics of the New Digital Frontier

The field of digital ethics is currently where bioethics stood fifty years ago. We are facing moral quandaries that our existing frameworks are ill-equipped to handle. Before we accept the inevitability of AI-generated communication, we must decide what we are willing to sacrifice. If we allow machines to speak for us, do we diminish the value of the words themselves? If we use them to simulate empathy, do we erode our capacity for genuine human connection?

The Risks of Automated Speech

The erosion of authentic human-to-human telepathy
The devaluation of written and spoken expression
The psychological impact of simulated companionship
The loss of accountability in communication

We should not fear the technology, but we must respect the weight of the medium. Words are not just data; they are the architecture of our social reality. As we integrate these models into our lives, the task is to ensure that the machine remains a tool for our expression, rather than a replacement for it.

Key Takeaway

If we outsource our speech to machines, we risk losing the very essence of our human connection.

02 — Lenny's Newsletter

The Abstraction Ladder: Engineering a Life with Claude

How to stop treating AI as a tool and start treating it as a collaborator

By Claire Vo · 12 min read

Editor's note: The difference between a power user and a novice isn't technical skill; it's the ability to think one level higher.

Most people use AI as a better version of Google. They ask a question, get an answer, and move on. This is a waste of the technology's actual capacity. Felix Rieseberg, an engineering lead at Anthropic, argues that the real breakthrough happens when you stop performing manual tasks and start managing processes. The goal is to move up the abstraction ladder. If you are manually entering data, you are doing the work of a machine. If you are telling a machine how to find that data, you are doing the work of an engineer.

The Email Inventory Hack

Consider the mundane task of moving house. Traditionally, this involves hunting through boxes or spreadsheets to catalogue furniture. Rieseberg bypassed this by pointing Claude at his email inbox. His digital history—receipts, shipping confirmations, and dimensions—is a structured database he didn't even know he possessed. By parsing these records, Claude built a 3D floor planner populated with his actual belongings. He didn't type a single dimension; he simply provided the context.

Whenever you’re doing something annoying that doesn’t feel creative, pause and ask yourself if Claude could do it instead.

This philosophy applies to any domain where we accumulate digital footprints. Clothing, medical records, travel history—these are all latent datasets. The barrier to using them isn't the complexity of the AI; it's the psychological habit of manual labour. We have been trained for decades to believe that certain tasks are simply 'the way things are done.' Unlearning this is the primary requirement for the AI era.

Choosing the Right Brain

The Model Selection Heuristic

Use Sonnet for well-scoped, specific, and execution-heavy tasks.
Use Opus when you don't know exactly what you are asking for.
Use Opus for problem decomposition and high-level reasoning.
Treat errors as workflow bugs, not model failures.

A common mistake is blaming the model when a task fails. Rieseberg suggests a different approach: debug the workflow. If the output is wrong, don't curse the machine. Instead, ask it to walk you through its reasoning. Usually, the issue is a lack of clean data or a prompt that is too vague. By treating the AI as a collaborator that requires better instructions, you move from being a frustrated user to a capable director.

Ultimately, the objective is to free up creative energy. The more we automate the 'annoying little problems,' the more space we create for the work that actually requires a human mind. The machine should handle the logistics; we should handle the intent.

Key Takeaway

Stop doing the work that a machine can find itself; move one level of abstraction higher.

03 — Simon Willison

Magnifica Humanitas: The Vatican's Warning

Decoding Pope Leo XIV's encyclical on the AI revolution

By Simon Willison · 10 min read

Editor's note: The Church is entering the fray, not as a tech critic, but as a defender of human dignity.

The Vatican has released a significant document: *Magnifica Humanitas*. Named by Pope Leo XIV in reference to the social teachings of the industrial revolution, this encyclical addresses the modern challenge of artificial intelligence. It is a rare instance of high-level moral philosophy meeting cutting-edge technological reality. The Pope's choice of name is deliberate; he is framing the AI revolution as a social and moral crisis, not just a technical one.

The Cultivation of Intelligence

One of the most striking sections of the document deals with the nature of LLMs. The Pope notes that these systems are 'cultivated' rather than 'built.' Developers do not design every detail; they create a framework in which intelligence grows. This leads to a fundamental problem of interpretability. We are creating systems whose internal representations remain unknown even to their creators. This lack of transparency is not just a technical hurdle; it is a moral risk.

Current AI systems are more 'cultivated' than 'built,' for developers do not directly design every detail, but instead create a framework within which the intelligence 'grows.'

This 'growth' model introduces a layer of unpredictability that challenges our ability to assign responsibility. If a system's decision-making process is opaque, how can we ensure it aligns with human justice? The encyclical warns that we must not allow the pursuit of efficiency to override the necessity of understanding. A tool we cannot explain is a tool we cannot fully control.

The Illusion of Relationship

The Three Perils of Personal AI Use

The search for ready-made answers that weakens personal judgment.
The false impression of objectivity in biased models.
The simulation of empathy that creates a hollow sense of connection.

The Pope is particularly concerned with the 'artificial imitation' of human communication. When a machine simulates empathy or friendship, it creates an illusion of a relationship. For the discerning user, this is a novelty; for the vulnerable, it is a trap. These simulated bonds do not build genuine relationships; they only mimic the appearance of them. This is a direct threat to the social fabric that relies on authentic human presence.

The document concludes with a call for development that places people at the centre. Progress is not truly human if it merely increases consumption while shifting the environmental and social costs onto others. As we build the future, we must ensure that the dignity of the person is not a secondary consideration to the accumulation of computational power.

Key Takeaway

AI must serve human dignity, not merely simulate human presence.

04 — Lenny's Newsletter

The Builder's Library

Distilling decades of wisdom for the modern product creator

By Lenny Rachitsky · 15 min read

Editor's note: In an age of infinite shallow content, the deep wisdom of a well-vetted book is a competitive advantage.

The modern professional is drowning in a sea of ephemeral content. Newsletters, podcasts, and social media threads offer immediate gratification, but they rarely offer lasting mental models. There is a fundamental difference between consuming information and building a foundation. A great book is the result of a person spending years distilling their best ideas into a coherent whole. For a product builder, these books are not just reading material; they are the blueprints for a career.

The Architecture of Communication

Before you can build a product, you must be able to communicate its value. This requires more than just good grammar; it requires an understanding of how to structure thought. Lenny Rachitsky identifies a core set of texts for this purpose, including Zinsser's *On Writing Well* and Pressfield's *Nobody Wants to Read Your Sh*t*. These are not books about style; they are books about the discipline of clarity. If you cannot write clearly, you cannot think clearly, and if you cannot think clearly, you cannot lead.

The smartest person in the world on a topic I care about spent years of their life distilling their best ideas into an enjoyable read.

The transition from a junior contributor to a senior leader requires a shift in focus from execution to strategy. This is where the 'CEO' bucket of books becomes essential. Works like *The Goal* and *High Output Management* move the reader away from the minutiae of tasks and toward the understanding of systems. They teach that management is not about supervising people, but about optimizing the flow of value through an organisation.

The Essential Curriculum

Core Reading Categories

Writing & Clarity: Pressfield, Zinsser, Dicks.
Management & Scale: Mochary, Hughes Johnson, Grove.
Strategy: Rumelt, Martin, Bryar.
Execution & Grit: Mechner, Fadell, Knight.

One of the most important filters for selecting what to read is time. Marc Andreessen's advice—to prioritise books that are over ten years old—is a powerful heuristic. If an idea has survived a decade of technological and cultural shifts, it is likely a fundamental truth rather than a passing trend. In the product world, where 'new' is often equated with 'better,' the ability to lean on proven principles is a significant advantage.

Ultimately, building a product is a human endeavour. It requires empathy, resilience, and the ability to navigate chaos. The books that matter most are those that teach you how to manage yourself and your impact on others. They provide the long-term stability needed to survive the short-term volatility of the tech industry.

Key Takeaway

Prioritise books that have stood the test of time over the latest trend.

05 — Stratechery

The Nvidia Pivot

Decoding the new reporting structures of the AI era

By Stratechery · 7 min read

Editor's note: Nvidia is no longer just a chip company; it is a full-stack infrastructure provider.

Nvidia is undergoing a fundamental shift in how it presents its business to the world. For years, the narrative was simple: they make the chips that power the AI revolution. But as the market matures, that narrative is becoming insufficient. The company is changing its reporting to distinguish between two very different types of customers: the hyperscalers and everyone else. This is not just an accounting change; it is a strategic signal.

The Battle for the Hyperscalers

The hyperscalers—the giants like Google, Amazon, and Microsoft—are the primary buyers of Nvidia's most advanced silicon. However, these same giants are also Nvidia's most dangerous competitors. They are all building their own custom AI chips to reduce their dependence on third-party hardware. In this segment, Nvidia is fighting a war against commoditisation. They must prove that their silicon is so much better than a custom-built alternative that the premium is worth paying.

Nvidia is fighting commoditisation with hyperscalers, while running the whole stack with everyone else.

This creates a bifurcated business model. On one side, you have a high-volume, high-competition commodity business. On the other, you have a high-margin, integrated ecosystem. By separating these in their reporting, Nvidia is telling investors exactly where they have a moat and where they are in a fight. It is an admission of the reality of the AI stack.

Owning the Stack

The Two Faces of Nvidia

Hyperscaler Sales: Fighting against custom silicon and commoditisation.
Full-Stack Sales: Providing hardware, software, and networking to the broader market.
The Moat: Integration of CUDA and networking protocols.

The 'everyone else' category is where Nvidia's true power lies. For the vast majority of companies, Nvidia doesn't just sell a chip; they sell an entire ecosystem. This includes the software layers, the networking protocols, and the developer tools that make the hardware useful. In this segment, Nvidia is not just a component supplier; they are the infrastructure. They run the whole stack, making it incredibly difficult for a customer to switch to a competitor.

This dual identity is the key to understanding Nvidia's future. They are simultaneously a hardware manufacturer and a platform provider. Their success depends on their ability to maintain the dominance of their software ecosystem while defending their hardware margins against the world's largest tech companies.

Key Takeaway

Nvidia is transitioning from a chip maker to a full-stack infrastructure provider.

06 — Lenny's Newsletter

The Magic of the Unlearned

Why children are the ultimate AI power users

By Lenny Rachitsky · 6 min read

Editor's note: The greatest barrier to AI adoption isn't technical; it's the mental prison of knowing too much about how computers 'should' work.

There is a striking irony in the current state of AI adoption. While engineers and executives struggle to integrate these tools into their workflows, children are using them with a natural, almost magical fluency. Felix Rieseberg, an engineer at Anthropic, has observed this firsthand. He sees parents sharing videos of their children building complex, interactive worlds with Claude—tools that would have required a dedicated software team only a few years ago.

The Mind Prison of Experience

The reason for this gap is not intelligence, but experience. Adults have spent decades being trained by computers. We have learned the rules of the interface: you click a button, you type a command, you follow a specific logic. We have built a 'mind prison' of knowing exactly what computers can and cannot do. We approach AI with a set of preconceived limitations, often trying to force it into the old paradigms of software.

Adults have spent 20 years in a 'mind prison' learning what computers can’t do. Unlearning that is the unlock.

Children, however, have no such baggage. They don't care about the underlying architecture or the 'correct' way to prompt. They simply interact. They treat the AI as a conversational partner that can manifest their ideas. They ask for a game with a specific character, or a story that changes based on their choices, and they expect it to happen. Their lack of technical cynicism allows them to explore the boundaries of the technology in ways adults simply don't.

Unlearning as a Skill

How to Adopt a Child-Like Approach

Stop asking 'How do I do this?' and start asking 'Can this be done?'
Treat the AI as a collaborator, not a search engine.
Experiment without worrying about the 'correct' method.
Focus on the desired output rather than the technical process.

The real challenge of the AI era is not learning new skills, but unlearning old ones. We must break the habit of assuming that certain tasks are 'too complex' or 'not possible' for a computer. The gap between what we think AI can do and what it actually can do is largely psychological. To truly harness this technology, we need to rediscover the curiosity and the lack of inhibition that defines a child's relationship with the world.

The goal is to move from being a user of software to being a director of intelligence. This requires a fundamental shift in how we perceive our relationship with machines. We are no longer just operating tools; we are orchestrating capabilities.

Key Takeaway

The most effective AI users are those who have unlearned the limitations of traditional computing.