The Deep Feed

01 — Not Boring

The End of Chronic Management

How gene therapy is turning heart disease from a lifelong struggle into a single event

By Packy McCormick · 12 min read

Editor's note: A look at how Eli Lilly's latest breakthrough could eliminate the need for daily cholesterol medication forever.

For decades, the strategy for managing cardiovascular health has been one of endless maintenance. If your LDL cholesterol levels were too high, you took a statin. If those weren't enough, you added more. It was a cycle of daily pills, constant monitoring, and the quiet, persistent threat of a heart attack or stroke. We treated the symptoms of a genetic predisposition, trying to outrun a biological certainty with chemistry. But the era of management is facing a challenger: the era of correction. We are moving from the pharmacy to the genome, attempting to fix the broken code rather than just buffering its effects.

The PCSK9 Breakthrough

The recent results from Eli Lilly regarding the VERVE-102 gene therapy represent a shift in the scale of medical possibility. The target is PCSK9, a protein that regulates LDL cholesterol. By using a single infusion to edit the gene responsible for this protein, researchers have seen massive reductions in cholesterol levels. In early trials, a single dose reduced PCSK9 levels by up to 88% and LDL levels by 62%. This isn't just a better version of a pill; it is a permanent alteration of the body's internal chemistry. It mimics a rare genetic mutation found in certain populations—people who are naturally protected from high cholesterol—and brings that advantage to the rest of us.

We are no longer just treating the disease; we are rewriting the instructions that allow it to exist.

The scale of this is difficult to grasp without looking at the raw numbers. Cardiovascular disease claims roughly 18.6 million lives every year. High LDL cholesterol alone accounts for about 4.4 million of those deaths. If a single shot can knock out a significant portion of that risk, the economic and human implications are staggering. We are talking about moving away from a model of 'patienthood'—where a person is defined by their daily regimen—toward a model of 'resolution,' where a medical intervention is a discrete, finished event.

The shift in cardiovascular strategy:

From daily chemical management to one-time genetic correction.
From treating symptoms to altering the underlying biological mechanism.
From lifelong pharmaceutical dependency to permanent physiological change.

However, the transition from Phase 1 trials to global availability is fraught with complexity. We are dealing with the permanent editing of human biology. While the 'free lunch' of PCSK9 inhibitors has been available for some, those drugs require lifelong adherence. The gene therapy approach offers the same benefit but removes the human error of forgetting a pill. The question for the next decade is not just whether we can edit these genes, but how we manage the long-term consequences of making these changes permanent across entire populations.

Key Takeaway

The future of medicine lies in moving from the constant management of symptoms to the permanent correction of genetic causes.

02 — Stratechery

The Efficiency Trap

Why Ferrari's electric pivot is facing a crisis of identity

By Ben Thompson · 10 min read

Editor's note: An analysis of why the Ferrari Luce is struggling to connect with enthusiasts despite its design.

The Ferrari Luce, designed by Jony Ive, is a beautiful object. On paper, it is a triumph of modern engineering. Yet, the reception has been cold. This reaction is not about the car's aesthetics or its technical specifications; it is about a fundamental clash of values. Ferrari has built its reputation on excess, noise, and the visceral, often irrational experience of internal combustion. Electric vehicles, by their very nature, are exercises in optimization. They are designed for efficiency, for the smooth, silent, and calculated movement from point A to point B. When you apply the logic of efficiency to a brand built on passion, you don't get a better car; you get a different category entirely.

The Alienation of Optimization

There is a psychological tension at play here. Modern technology, much like the AI revolution, tends to smooth out the edges of experience. We want things to be faster, quieter, and more efficient. But human satisfaction often lives in the friction. The roar of an engine or the mechanical resistance of a gearbox provides a sense of agency and connection to the machine. When everything becomes a seamless, software-defined experience, we lose the sense of being a participant. We become mere passengers in a highly optimized system. The Luce feels like a product of this new world—a world where the goal is to remove the very things that made the brand feel alive.

Efficiency is the enemy of character in a machine built for emotion.

This tension extends beyond the automotive industry into the broader tech sector. As we move toward AI-driven interfaces, we are seeing a similar trend. We are trading the complexity of direct interaction for the convenience of automated outcomes. This creates a sense of alienation. If the machine does everything perfectly, what is left for the human to do? The backlash against the Luce is a symptom of a larger cultural anxiety: the fear that in our pursuit of a frictionless life, we are accidentally removing the parts of life that actually matter.

The conflict in EV branding:

Performance vs. Efficiency: One is about feeling; the other is about math.
Complexity vs. Simplicity: One offers engagement; the other offers ease.
Identity vs. Optimization: One builds a cult of personality; the other builds a utility.

Ferrari's challenge is to find a way to introduce electric propulsion without surrendering its soul to the spreadsheet. They must find a way to make efficiency feel like an expression of power rather than a compromise of character. If they fail, the Luce will be remembered not as a breakthrough, but as the moment the brand traded its identity for a better set of efficiency metrics.

Key Takeaway

True brand identity often lives in the friction and excess that optimization seeks to eliminate.

03 — Simon Willison

The Value of Saying 'I Don't Know'

Why Anthropic's latest model prioritises honesty over performance

By Simon Willison · 8 min read

Editor's note: A look at why Claude Opus 4.8's most important feature is its ability to admit uncertainty.

In the current AI arms race, the prevailing metric for success is capability. We want models that can code better, reason faster, and answer more questions. But this obsession with raw power has a dangerous side effect: hallucination. When a model is trained to be helpful, it often learns that the most 'helpful' response is a confident one, even if that response is factually incorrect. We are building incredibly intelligent systems that are also incredibly prone to lying to us with total conviction. This is why Anthropic's release of Claude Opus 4.8 is so interesting. They aren't claiming a massive leap in intelligence; they are claiming a leap in honesty.

The Honesty Metric

Opus 4.8 is described as a 'modest but tangible improvement.' The real breakthrough is its tendency to flag uncertainty. Instead of jumping to a conclusion when the evidence is thin, the model is now more likely to abstain or admit it doesn't know. In evaluations, this has translated to a significant reduction in incorrect code being passed off as working. This is a fundamental shift in how we should value AI. A model that is 90% accurate and tells you when it is unsure is infinitely more useful than a model that is 95% accurate but lies about the remaining 5%.

Reliability is more important than raw intelligence in an agentic world.

This shift is essential for the next stage of AI development: agency. As we move from chatbots to agents—systems that can actually execute tasks, manage finances, or write software—the cost of a mistake rises exponentially. An AI that confidently executes the wrong command is a liability. An AI that stops and asks for clarification is a partner. By training models to value accuracy over the appearance of competence, Anthropic is addressing the primary barrier to widespread AI autonomy.

Why honesty matters for AI:

Reduces the risk of 'confident hallucinations'.
Enables safer deployment in high-stakes agentic workflows.
Builds long-term user trust through predictable behavior.

The industry is beginning to realise that the goal isn't just to build a more powerful brain, but a more reliable one. As the models become more integrated into our lives, the ability to trust their output becomes the most important feature they can offer. Claude Opus 4.8 suggests that the path to true utility isn't through more knowledge, but through better judgement.

Key Takeaway

In the transition from chatbots to agents, the ability to admit uncertainty is more valuable than the ability to provide an answer.

04 — The Marginalian

The Danger of Certainty

Hermann Hesse and the heavy cost of self-righteousness

By Maria Popova · 15 min read

Editor's note: Reflecting on Hesse's warnings about how absolute belief can lead to catastrophe.

In 1919, as the world attempted to pick up the pieces of the First World War, Hermann Hesse wrote a letter to a young German that served as both a consolation and a warning. He was writing to a generation that had been crushed by the weight of their own convictions. They had marched off to war believing they were doing 'good,' only to find themselves in the mud, witnessing the slaughter of millions. Hesse's insight was that the most dangerous people are not those who doubt, but those who are absolutely sure of their righteousness. Certainty is the mechanism by which humans justify the unjustifiable.

The Trap of the 'Right Good'

Hesse argues that the tragedy of the war was not just a failure of politics, but a failure of the individual mind. People had become convinced of a specific version of morality—a version that demanded blood in the name of honour or country. This 'good' was a false one, because it was not timeless or indestructible; it was a construct of the moment, used to fuel a machine of destruction. When we lose the ability to question our own motives, we become susceptible to the herd mentality that allows atrocities to be framed as duties.

To not know whether there is a God, or whether there is good and evil, is far better than to know for sure.

This is a radical stance in a world that demands conviction. We are taught that doubt is a weakness, that to be a leader or a person of character, one must stand firm. But Hesse suggests that doubt is actually a form of spiritual hygiene. It is the only thing that prevents our morality from becoming a weapon. By maintaining a state of uncertainty, we remain open to the complexity of the world and the humanity of others. Certainty closes the mind; doubt keeps it alive.

Lessons from Hesse's philosophy:

Self-respect comes from accepting responsibility for one's own life.
Certainty often masks a lack of true moral understanding.
The capacity to doubt is the primary defence against mass delusion.

As we navigate our own era of intense political and technological polarization, Hesse's warning feels more urgent than ever. We are constantly pressured to pick a side, to adopt a set of certainties, and to view the 'other' through the lens of our own perceived righteousness. To resist this is not to be indecisive, but to be truly responsible. It is to acknowledge that the truth is rarely as simple as our convictions would have us believe.

Key Takeaway

The most dangerous form of error is not being wrong, but being certain that you are right.

05 — Simon Willison

The AI Amish

The growing urge to retreat from the agentic age

By Simon Willison · 9 min read

Editor's note: A look at the psychological toll of constant technological acceleration.

There is a growing movement of people who are not just 'unplugging' for a weekend, but are making a fundamental decision to step away from the digital world. This isn't a reaction to social media or doomscrolling; it is a reaction to something deeper. It is a reaction to the 'agentic' era of technology—the moment when software stops being a tool we use and starts being a presence that shares our inner monologue. For people like Chad Whitacre, the arrival of highly capable AI agents was the final straw. It felt less like progress and more like an intrusion into the very essence of human thought.

The Second Person in Your Head

The sensation described by those experiencing this shift is unsettling. When you work closely with an AI agent, it begins to feel as though you have another 'person' in your head, a digital shadow that is constantly processing, suggesting, and anticipating. This creates a sense of cognitive crowding. The boundary between your own thoughts and the machine's outputs begins to blur. For some, this is the ultimate convenience; for others, it is a loss of self. The 'AI Amish'—those who seek to live a more analog, pre-screen life—are not rejecting technology itself, but the way it has begun to colonise the human mind.

It was like I had another 'person' in my head, sharing my inner monologue—but that person was a computer system owned by a megacorp.

This retreat is a form of resistance against technological accelerationism. It is an attempt to reclaim the 'pre-screen' life, where the world is experienced directly rather than through a layer of algorithmic mediation. It is a recognition that there is a limit to how much complexity a human psyche can integrate before it begins to feel fundamentally altered. The desire to return to the '1980s'—a world of discrete tools and clear boundaries—is a search for a way to be human without being a node in a network.

Drivers of the digital retreat:

The loss of cognitive boundaries caused by AI agents.
The feeling of being 'intoxicated' by agentic loops.
The desire to reclaim agency from corporate-owned intelligence.

Whether this movement is a niche eccentricity or a precursor to a wider cultural shift remains to be seen. But it highlights a critical tension in our era: as our tools become more capable of mimicking human thought, we are forced to ask what, if anything, is left that is uniquely ours. The 'AI Amish' are not just leaving tech; they are trying to find their way back to themselves.

Key Takeaway

As technology moves from being a tool to being a cognitive partner, the boundary of the self becomes the new frontier of resistance.

06 — Simon Willison

The Local Machine

How running code in the browser is decentralising the web

By Simon Willison · 7 min read

Editor's note: A technical deep-dive into how WebAssembly is changing the architecture of the internet.

For most of the history of the web, the relationship between the user and the server has been one of strict hierarchy. The browser is a thin client, a window that requests data, and the server is the heavy, central authority that processes it. This model has shaped everything from how we build apps to how we manage privacy. But a new architectural shift is underway, driven by technologies like Pyodide and WebAssembly. We are seeing the emergence of the 'local machine'—the ability to run complex, server-side logic entirely within the user's browser, without needing a backend at all.

The End of the Heavy Backend

By using a service worker to intercept requests and a WebAssembly-based Python runtime to execute them, developers can now run full ASGI web applications—like FastAPI or even complex data tools like Datasette—directly in a browser tab. This isn't just a neat trick for performance; it is a fundamental redistribution of power. When the 'backend' lives on the client's machine, the need for massive, centralized server farms to handle every minor computation diminishes. The compute moves to the edge, to the very device the user is holding.

We are moving from a world of client-server hierarchy to a world of distributed local execution.

The implications for privacy and autonomy are significant. If an application can run entirely in the browser, your data never has to leave your device. You can interact with complex datasets and run sophisticated analyses without ever sending a single byte to a central server. This breaks the current trend of 'data harvesting' that defines much of the modern web. It allows for a model of software that is truly private by design, rather than private by policy.

Benefits of browser-based execution:

Enhanced privacy by keeping data on the local device.
Reduced server costs and infrastructure complexity.
Lower latency for complex, interactive applications.

This shift represents a move toward a more resilient and decentralized internet. As we continue to push the limits of what the browser can do, the line between 'client' and 'server' will continue to blur. The result will be a web that is faster, more private, and more capable of supporting complex, local-first applications that don't rely on the permission of a central authority.

Key Takeaway

The migration of compute from the server to the browser is a vital step toward a more private and decentralized web.