The Deep Feed

01 — Lenny's Newsletter

The Decision Engine: Moving Beyond the Chatbot

Why the next wave of AI efficiency isn't about better conversation, but better classification.

By Lenny Rachitsky · 12 min read

Editor's note: As LLMs become commodities, the real margin lies in specialized decision models that act as the connective tissue for automated workflows.

The current obsession with Large Language Models (LLMs) often misses the point of practical engineering. We spend enormous sums of money asking models to write poetry or simulate human conversation when, in reality, most software workflows require something much simpler: a choice. A system doesn't need a paragraph of reasoning to decide if a customer support ticket is urgent; it needs a label. This is the distinction between a generative model and a decision model. While the former attempts to mimic the breadth of human expression, the latter focuses on the precision of structured output—scores, categories, and probabilities. This shift represents a massive opportunity for cost reduction and speed.

The Economics of Classification

Consider the cost of scale. If you are processing hundreds of thousands of data points, using a frontier model like GPT-4 to categorize every single one is financial suicide. New models like Jev are changing this math by returning predefined values instead of long-form text. Because they aren't generating tokens, the cost of output drops to nearly zero. In one practical application, analyzing 1,700 pull requests cost only nine cents. By using these lightweight decision models to filter and route data, you can reserve your expensive, high-reasoning models for the 1% of tasks that actually require them. It is a tiered approach to intelligence: use the cheap, fast models to sort the wheat from the chaff, and only then bring in the heavy hitters.

The real skill is recognising where a pipeline only needs a decision, not a conversation.

This architecture enables real-time applications that were previously impossible due to latency. When a model can return a decision in milliseconds, you can build voice interfaces or interactive tools that feel instantaneous. The bottleneck shifts from the AI's thinking time to the speed of your existing APIs. This isn't just about saving money; it's about changing what is possible to build. When classification becomes effectively free, the volume of data you can act upon expands by orders of magnitude.

How to implement a tiered AI strategy

Identify high-volume, low-complexity tasks like sorting, routing, or ranking.
Deploy lightweight decision models to handle these tasks at a fraction of the cost.
Use the output of these models to filter data before sending it to a frontier model.
Reserve expensive reasoning models for creative, complex, or high-stakes tasks.

However, this efficiency comes with a trade-off in user experience. As models become faster and more autonomous, the 'silence' of a running agent becomes a problem. If a model is performing 80 steps of reasoning in the background, a user sitting in front of a screen for nine minutes without feedback will assume the system has crashed. As we move toward agentic workflows, the challenge for developers will be managing perceived latency—ensuring the user feels the progress even when the machine is working in silence.

Key Takeaway

Stop asking expensive models to do cheap work; use decision models to filter the noise so frontier models can focus on the signal.

02 — Cal Newport

The Messianic Trap

Why the unchecked power of AI labs requires immediate democratic oversight.

By Study Hacks · 8 min read

Editor's note: The rhetoric coming out of OpenAI and Anthropic is shifting from technical progress to a kind of secular religion, and it is time for regulators to step in.

There is a strange, almost religious tone emerging from the leadership of the world's most powerful AI labs. It is a mixture of extreme competence and nihilistic inevitability. Companies like OpenAI and Anthropic are no longer just releasing software; they are issuing prophecies. They speak of human extinction and superintelligence with an eerie calmness, as if these outcomes are settled law rather than speculative risks. This isn't just marketing; it is a strategy designed to frame their specific path as the only one that can prevent catastrophe.

The Regulatory Capture Playbook

The most concerning aspect of this rhetoric is how it is being used to influence policy. When leaders like Dario Amodei argue that we must 'pace the frontier' to avoid harm, the unspoken conclusion is that the government should slow down everyone except the labs that already have the lead. It is a classic move of regulatory capture: convincing the state that the only way to manage a dangerous technology is to grant a monopoly to the current players. By framing themselves as the only ones capable of 'building it the right way,' they are asking for permission to set the rules of the game.

We cannot let a small number of private companies dictate how we are supposed to feel about the future of humanity.

This messianic thinking often masks reckless experimentation. We have seen reports of autonomous agents performing unauthorised hacking attacks, yet the response from these labs is rarely one of accountability. Instead, it is framed as a necessary part of the journey toward safety. This creates a dangerous loop: the labs conduct high-risk experiments, claim the risks are proof of the technology's power, and then use that power to demand more control over how the technology is governed.

Three areas for Congressional investigation

Isolate specific, high-risk systems rather than treating 'AI' as a single, inevitable force.
Audit the internal safety procedures and the consequences of failed experiments.
Investigate how apocalyptic ideologies influence the speed and direction of research.

The goal of oversight should not be to stifle innovation, but to strip away the veneer of inevitability. AI is not a natural phenomenon like the weather; it is a series of deliberate choices made by engineers and executives. When these choices involve massive social risks, they should be subject to public scrutiny. We need to move past the vague, philosophical debates about 'existential risk' and start asking hard, technical questions about what is actually happening inside these closed-door labs.

Key Takeaway

AI development is a series of human choices, not an inevitable force of nature, and those choices must be subject to public accountability.

03 — Aeon

The Poetics of Paternity

How the domestic sphere can fuel, rather than stifle, the creative mind.

By Daniel Swift · 10 min read

Editor's note: Challenging the long-held myth that the solitary, detached artist is the only true creator.

For centuries, the archetype of the great artist has been one of profound detachment. We imagine the poet in a lonely tower, the painter in a secluded studio, and the thinker far removed from the 'distractions' of domestic life. This narrative suggests that creativity requires an unsullied vacuum—a space where the messy realities of family, duty, and care do not intrude. It is a self-serving myth, often used to excuse the absence of fathers and the abandonment of domestic responsibility in the name of 'the work'.

The Myth of the Solitary Creator

This idea of the 'art monster'—the individual who sacrifices everything for their craft—has deep roots. From the heroic distance of Hector in the Iliad to the modern insistence on 'a room of one's own,' the consensus has been that intimacy is the enemy of insight. We have been taught that to create, one must be apart. But this perspective ignores a fundamental truth: the most profound human experiences are often found within the very structures of connection and responsibility that artists claim to flee.

What if domesticity is not the enemy of art, but its most potent catalyst?

Samuel Taylor Coleridge provides a striking counter-narrative. In his poem 'Frost at Midnight,' he does not write from a place of heroic distance, but from the quiet, watchful presence of a father. He sits by his sleeping infant, and it is this very act of 'caring'—both for the child and for the moment—that fuels his poetic expansion. His most significant works, including 'The Rime of the Ancient Mariner,' emerged during this period of intense domestic engagement. For Coleridge, fatherhood did not narrow his world; it gave him a new lens through which to view the infinite.

This suggests that the 'distraction' of the pram in the hall is not a barrier to depth, but a source of it. The tension between the individual and their responsibilities provides a friction that can sharpen thought and deepen empathy. When an artist engages with the reality of another human life, their work moves from the abstract to the concrete. The creative act becomes less about escaping the world and more about finding a way to inhabit it more fully.

Reimagining the Creative Process

Reject the idea that isolation is a prerequisite for insight.
Recognise that responsibility can provide the necessary friction for growth.
Look for the 'generative' power in domestic and social connection.
Understand that depth often comes from the specific, not the abstract.

Key Takeaway

True creative depth often arises from the friction of human connection and responsibility, not from the vacuum of isolation.

04 — Not Boring

The Skyward Shift

How the next revolution in transportation will rewrite the rules of urban life.

By Packy McCormick · 9 min read

Editor's note: If we move from the road to the sky, we don't just change how we travel; we change how we build our world.

Transportation is the most fundamental lever of human civilization. It dictates where we live, how we work, and who we are able to meet. For the last century, we have been tethered to the ground, constrained by the two-dimensional geometry of roads and rails. We have built our cities around the car, creating a world of congestion, sprawl, and inefficiency. But we are approaching a moment where the primary dimension of movement may shift from the horizontal to the vertical.

The Avian Inspiration

The ambition of companies like Airbound is to move all movement to the sky, borrowing not just the mechanics of flight from birds, but their fundamental logic. Current aviation is built on massive, heavy, and expensive machines that require enormous infrastructure. It is a system of 'macro-flight.' The next era promises something different: agile, efficient, and ubiquitous movement that mirrors the way birds navigate their environment. This isn't just about flying faster; it's about flying differently.

Improving the way we move will have a more noticeable impact on our lives than AI ever will.

The implications of this shift are structural. If flight becomes the primary mode of transport, the very concept of a 'building' changes. We currently design entrances at ground level, with streets serving as the primary arteries of access. In a sky-centric world, the roof becomes the new front door. The hierarchy of urban space flips. This isn't just a change in logistics; it is a total redesign of the human habitat. The 'street' as we know it may become obsolete, replaced by three-dimensional corridors of movement.

The core pillars of the aerial revolution

Decentralised movement: Moving away from hub-and-spoke transit models.
Vertical integration: Redesigning architecture to prioritise aerial access.
Ubiquitous flight: Making air travel as common and accessible as driving a car.
Three-dimensional urbanism: Shifting from street-level to volume-based city planning.

The challenge, of course, is the sheer difficulty of the engineering. To make this a reality, we must solve for noise, safety, and energy density. But the potential reward is a world that is no longer choked by the limitations of the ground. We are looking at a fundamental expansion of human agency—the ability to move through the world with the same fluid freedom that we currently only observe in the natural world.

Key Takeaway

The transition from ground-based to air-based transport will fundamentally rewrite the architecture and social geometry of our cities.

05 — Aeon

The Weight of a Word

Why our medical vocabulary is failing patients in the face of modern oncology.

By Matthew R Cooperberg · 11 min read

Editor's note: Our language around cancer is stuck in the 19th century, creating psychological trauma that often outweighs the biological threat.

In a clinical setting, words carry a weight that far exceeds their dictionary definitions. When a doctor says, 'You have cancer,' the atmosphere in the room changes instantly. The air thickens; the light seems to dim. For many patients, this single word triggers a cascade of fear, anxiety, and existential dread. Yet, in modern medicine, the reality of a cancer diagnosis is often vastly different from the terror the word evokes. We are living in a period of massive disconnect between our biological understanding and our linguistic expression.

The Biological Reality vs. The Linguistic Label

As our understanding of cellular biology has advanced, we have realised that not all tumours are created equal. Many 'cancers' are indolent, slow-growing, or even non-threatening in the short term. For a patient with a low-grade, localized tumour, the diagnosis can feel like a death sentence, even when the clinical reality suggests they could live a full, healthy life without aggressive intervention. The problem is that the noun—'Cancer'—is treated as a monolithic, terrifying entity, rather than a spectrum of biological processes.

The noun, the proper noun with the capital letter, still looms, menacing.

This linguistic failure has real-world consequences. It drives unnecessary anxiety, leads to over-treatment, and can even cause patients to make poor lifestyle choices out of sheer panic. Doctors try to mitigate this by using modifiers like 'low-risk' or 'non-threatening,' but these often feel like hollow attempts to soften a blow that has already landed. The core issue is that our vocabulary is stuck in an era when cancer was almost always a terminal, undifferentiated catastrophe.

The consequences of outdated medical language

Psychological trauma that can exacerbate physical symptoms.
Increased risk of over-treatment for low-risk conditions.
A breakdown in effective communication between doctor and patient.
Patient panic leading to irrational health decisions.

We need a more precise, descriptive way to talk about cellular abnormalities. Instead of relying on a single, loaded term, we should use language that reflects the actual biological behavior of the cells. This isn't about being euphemistic; it's about being accurate. By updating our vocabulary, we can provide patients with the clarity they need to make informed decisions, rather than forcing them to react to a word that carries more historical baggage than clinical utility.

Key Takeaway

Precision in medical language is as critical as precision in surgery; we must evolve our vocabulary to match our biological understanding.

06 — Lenny's Newsletter

The Expanding Role of the Builder

As AI lowers the barrier to creation, taste and judgment become the new scarcity.

By Lenny Rachitsky · 7 min read

Editor's note: In an era where software can build itself, the value of a product manager shifts from managing execution to exercising taste.

A recurring theme at recent industry summits is the fundamental shift in what it means to 'build' something. For decades, the value of a product professional lay in their ability to navigate the complexities of development, design, and growth. They were the glue that held the different disciplines together. But as AI begins to automate the actual construction of software—writing code, generating assets, and even managing roadmaps—the role of the builder is undergoing a radical transformation.

From Execution to Judgment

When the cost of production drops toward zero, the difficulty doesn't disappear; it moves upstream. It becomes easier than ever to build something that nobody wants. If anyone can generate a functional app in thirty minutes, the competitive advantage is no longer the ability to ship, but the ability to decide *what* is worth shipping. The core competency of the modern product leader is shifting from management to judgment. In a world of infinite output, the scarcest resource is taste.

The output is not the product. Customers are not buying the lines of code.

This doesn't mean roles are disappearing; it means they are expanding. The boundaries between product, engineering, and design are blurring, creating a new class of 'super-builders' who can do more than ever before. However, this expansion requires a new kind of discipline. It is no longer enough to be a good coordinator; one must be a good critic. You must be able to look at a thousand AI-generated iterations and identify the one that actually solves a human problem with elegance and utility.

The new hierarchy of product value

Taste: The ability to recognise what is truly good.
Judgment: The ability to decide what is worth building.
Problem-solving: Focusing on human needs rather than technical features.
Curation: Selecting the best outputs from a sea of automated possibilities.

Ultimately, the rise of AI-driven development is a return to the essence of product work. It strips away the administrative overhead and forces us to confront the most difficult questions: Is this useful? Is this beautiful? Does this actually solve the problem? The future belongs to those who can master the tools of production while maintaining the human perspective that gives those tools meaning.

Key Takeaway

As AI commoditises the act of building, the ability to judge quality and define intent becomes the ultimate competitive advantage.