The Deep Feed

01 — Lenny's Newsletter

The Loop Revolution: Beyond the Single Prompt

Moving from static instructions to autonomous agency in AI

By Lenny's Newsletter · 12 min read

Editor's note: The era of the prompt is ending; the era of the autonomous loop is beginning.

The current obsession with prompt engineering is a distraction. We have spent months learning how to talk to machines, treating them like clever parrots that require the perfect incantation to perform. But the real shift is happening in the background. We are moving from a world of single, isolated prompts to a world of loops. A loop is not just an automated instruction; it is a system that can observe, decide, and act without a human holding its hand at every step. When you design a loop, you are no longer just a writer of instructions; you are an architect of agency. You are building something that can run while you sleep, managing its own schedules, hitting its own goals, and even hiring its own subagents to do the heavy lifting.

The Four Architectures of Automation

To build effective agents, you must first understand the four ways a loop can breathe. First, there is the 'heartbeat' loop—a simple, rhythmic pulse that checks in at set intervals. Then there is the 'cron' loop, which follows the rigid logic of a calendar, perfect for tasks that must happen every Monday at 9:00 a.m. Third, we have 'hooks,' which are reactive. They sit in silence until an event triggers them, much like a sensor in a factory. Finally, there are 'goal' loops. These are the most difficult and the most powerful. A goal loop does not care about time or triggers; it only cares about a result. It iterates, fails, tries again, and pivots until the objective is met. This is where the true power of AI agents lies, but it is also where most developers burn through their budget by creating infinite, aimless cycles of reasoning.

A loop is just an automated prompt, but the difference between a tool and an agent is the ability to decide when the work is done.

Designing these systems requires a shift in mindset. Stop thinking like a coder and start thinking like a manager onboarding a new employee. You wouldn't just give a new hire a list of tasks and walk away; you would give them a work tree, a set of specific skills, the tools they need to succeed, and a way to report their progress. An effective AI loop needs these same components: work trees to define the scope, plugins to connect to the real world, and state tracking to remember what it has already tried. Without state tracking, your agent is a goldfish, doomed to repeat the same mistakes in an expensive, circular dance of wasted tokens.

The Five Essentials for Production-Ready Loops

Work trees: A clear map of the tasks to be completed.
Skills: Specific, modular abilities the agent can call upon.
Connectors: The APIs and plugins that allow the agent to interact with external data.
Subagents: The ability to delegate smaller tasks to specialized models.
State tracking: A memory of previous attempts, successes, and failures.

The danger of this new frontier is cost. Because goal-based loops are non-linear, they can spiral. An agent tasked with 'improving a codebase' might spend hundreds of dollars in a single afternoon if its internal logic fails to recognise that it is stuck. The sign of a well-designed loop is not just that it completes its task, but that it knows when to stop. We are building machines that can think, but we must also build machines that know how to quit.

Key Takeaway

Stop writing prompts and start designing systems that can manage their own execution.

02 — The Marginalian

The Child Who Survived

Ursula K. Le Guin on the true meaning of maturity

By Maria Popova · 10 min read

Editor's note: Maturity is not the act of outgrowing our past, but the act of integrating it.

We often treat growing up as a process of shedding. We cast off our childhood whims, our irrational fears, and our unpolished dreams in favour of something more solid, more 'adult'. We view maturity as a linear progression away from the messy, unformed self of our youth. But Ursula K. Le Guin suggests this is a fundamental misunderstanding. Maturity is not an outgrowing; it is a growing up. It is not the replacement of the child with the adult, but the survival of the child within the adult. To grow up is to become a Russian nesting doll, where every previous version of ourselves remains intact, tucked inside the person we are today.

The Courage of Compassion

The most difficult part of this process is not the acquisition of responsibility, but the acquisition of compassion for our former selves. We are often our own harshest critics, looking back at our younger, more confused selves with embarrassment or even contempt. Le Guin argues that this denial is a form of stagnation. If we refuse to take responsibility for our missteps and our confusion, we can never truly own who we are. Maturity requires the courage to put our arms around those former, broken versions of ourselves and pull them close. We must accept the shadows we cast if we ever hope to stand in the light of our own strength.

An adult is not a dead child, but a child who survived.

This survival is inextricably linked to the power of imagination. In a world that prizes the factual, the efficient, and the 'grown-up', fantasy is often dismissed as trivial or childish. Yet, Le Guin posits that fantasy is a vital tool for truth. It allows us to explore realities that facts alone cannot reach. Many adults fear fantasy not because it is untrue, but because its truths threaten the false, unnecessary structures they have built around their lives. To embrace imagination is to embrace the freedom to be entirely ourselves, an expansion that requires constant, active engagement rather than a one-time achievement.

The Pillars of Maturity

Integration: Bringing past selves into the present rather than discarding them.
Self-Knowledge: Facing both our virtues and our shadows without denial.
Imagination: Using the faculty of fantasy to challenge false realities.
Responsibility: Accepting the weight of our own existence and its consequences.

Ultimately, maturity is a paradox. We are finite beings, yet we are unfinished. We are constantly evolving, revealing new edges and new shadows. The goal is not to reach a final, static state of perfection, but to maintain the capacity for growth while holding onto the essential spark that made us who we were in the first place. It is a lifelong project of becoming, a continuous movement toward a wholeness that exceeds our individual vices and virtues.

Key Takeaway

True maturity is the ability to embrace your past self with compassion rather than trying to bury it.

03 — The Marginalian

The Resurrection of Fyodor Dostoyevsky

What a staged execution teaches us about the will to live

By Maria Popova · 9 min read

Editor's note: A reminder that life is not found in our circumstances, but in our response to them.

In December 1849, Fyodor Dostoyevsky stood in a Saint Petersburg square, dressed in white, waiting for the sabers to fall. It was a staged execution, a cruel piece of political theatre designed to terrify the masses by showing them the absolute power of the Tsar. For a few agonizing minutes, Dostoyevsky was a dead man. He felt the cold reality of the end. And then, at the final second, the sentence was commuted. He was not to be executed, but exiled to a Siberian labor camp. This moment of near-death did not break him; it re-forged him. He emerged from the shadow of the scaffold with a terrifying, electric gratitude for the mere fact of being alive.

Life is Within Us

In the letters he wrote to his brother immediately following this brush with death, Dostoyevsky articulated a philosophy that would define his later masterpieces. He realised that life is not something that happens to us from the outside; it is something that exists within us. 'Life is everywhere, life is in us ourselves,' he wrote. This was a radical departure from the despair that often follows trauma. He understood that even if his intellectual ambitions were stripped away—even if he were prevented from writing for years—his capacity to love, to suffer, and to remember remained. That, he argued, was the essence of being human.

Life is a gift, life is happiness, each moment could have been an eternity of happiness.

Dostoyevsky’s experience highlights a fundamental truth about the human condition: our attitude is our final frontier. Long before Viktor Frankl documented this in the concentration camps, Dostoyevsky had lived it. He found that when the external world is stripped of its meaning, the internal world must become the source of all value. His transition from a young, idealistic radical to a man who understood the depth of human suffering and the necessity of spiritual purity was not a change of character, but a deepening of it. He learned that to live fully is to accept the possibility of loss and still choose to participate in the world.

Lessons from the Scaffold

Internal Agency: Life is an internal state, not an external set of circumstances.
The Value of Suffering: Pain is not an obstacle to life, but a part of its texture.
Radical Gratitude: The awareness of mortality can act as a catalyst for joy.
Purity of Spirit: Maintaining one's integrity is possible even in the harshest environments.

His letters reveal a man who, having faced the void, had no room left for malice or spite. He felt a sudden, overwhelming desire to embrace everyone from his past, even those he had quarreled with. This is the transformative power of facing one's own end: it burns away the trivialities of ego and leaves only the elemental facts of connection and existence. Dostoyevsky did not just survive the execution; he was regenerated by it.

Key Takeaway

The capacity to choose your attitude is the only freedom that cannot be taken from you.

04 — The Marginalian

The Seven Layers of the Self

Amélie Rorty on the architecture of personhood

By Maria Popova · 11 min read

Editor's note: We are more than just our names and our jobs; we are a complex hierarchy of identities.

We live in an age of fragmentation. Social media and modern professional life demand that we show up as specific, curated slices of ourselves. We are a 'professional' on LinkedIn, a 'parent' at home, a 'consumer' on an app. This constant parceling of the self threatens to lacerate the very fabric of our identity. But philosopher Amélie Rorty suggests that this fragmentation is not a new phenomenon, but a misunderstanding of how personhood actually works. She proposes that identity is not a single, flat surface, but a series of layers—a taxonomy of being that ranges from the simple to the complex.

From Characters to Presences

Rorty begins with the 'character', the most basic level of identity. A character is a collection of repeatable traits. They are the figures in a Dickens novel—predictable, delineated, and often replaceable. If you are merely a character, you do not have an identity crisis, because there is no unified core to lose. Above this are 'figures', who represent types of lives to be imitated. Then come 'selves', who possess properties, and 'individuals', who are centers of integrity with inalienable rights. At the highest level are 'presences'—those deep, almost spiritual identities found in the works of Dostoyevsky, where a person is not just a set of traits, but an evocative, soulful entity.

A person's identity is like a pattern drawn on a tightly stretched parchment. Touch just one part, and the whole person will react.

The danger lies in the mismatch between our layer of being and the society we inhabit. Rorty points out that a 'character' thrives in a society that values predictable roles, but can become a tragic figure in a time of great social change where their specific virtues are no longer recognised. A person of character might be seen as a fool in a different era. This suggests that our sense of well-being is not just an internal matter, but a product of how our specific layer of identity fits into the social architecture around us.

The Hierarchy of Personhood

Characters: Repeatable assemblages of traits.
Figures: Narratives of types to be imitated.
Selves: Possessors of specific properties.
Individuals: Centers of integrity and rights.
Presences: Deep, soulful, and evocative entities.

To 'have character' is to hold tightly to certain qualities, even when tempted to swerve. But to 'be a character' is to allow those qualities to dominate and dictate all other aspects of your life. Understanding these layers allows us to see why we feel so fractured in the modern world. We are being forced to live as characters in scripts we didn't write, rather than being allowed the space to develop into the more complex layers of individual and presence.

Key Takeaway

Identity is not a single point, but a hierarchy of layers that must be integrated to achieve wholeness.

05 — Stratechery

The Renters of Attention

Fox, Roku, and the shifting power in streaming

By Stratechery · 8 min read

Editor's note: In the battle for the living room, leverage is more important than content.

The market's reaction to Fox's acquisition of Roku was predictably sceptical. On the surface, it looks like a traditional media company trying to buy its way into a digital future. But this view misses the fundamental shift in the economics of attention. For decades, media companies have been content providers, subject to the whims of distributors. They extract value from rights holders, but they are ultimately at the mercy of the platform. By acquiring Roku, Fox is attempting to flip the script. They are moving from being a mere provider of content to being a renter of the platform itself.

The Leverage Play

The core of the strategy is leverage. In the current streaming landscape, the battle is not just about who has the best shows, but who owns the interface through which the viewer accesses those shows. If you own the hardware and the operating system, you own the relationship with the consumer. You control the data, the advertising, and the discovery. Fox is betting that by controlling the Roku ecosystem, they can gain a level of vertical integration that makes them less dependent on the massive, gatekeeping platforms like Netflix or Disney+. They are trading the extraction of rights for the leverage of the platform.

The battle for the living room is won by whoever controls the interface, not just the content.

However, this strategy is not without significant risk. Managing a hardware and software ecosystem is a fundamentally different business than producing television. It requires a focus on user experience, technical stability, and platform-wide growth—metrics that are often at odds with the short-term, content-driven goals of a traditional broadcaster. If Fox fails to make Roku a platform that third-party developers and other content providers want to live on, they will find themselves owning an expensive, isolated island.

The Strategic Trade-offs

Content vs. Platform: Moving from being a creator to being a gatekeeper.
Extraction vs. Leverage: Trading immediate rights revenue for long-term ecosystem control.
Vertical Integration: The attempt to own the entire value chain from production to the screen.
User Experience: The necessity of managing software, not just scripts.

Ultimately, the success of this acquisition will depend on whether Fox can execute a platform strategy. They must convince the world that Roku is not just a way to watch Fox, but a necessary destination for all digital entertainment. If they succeed, they become a dominant force in the new media order. If they fail, they will have spent billions to buy a very complicated way to lose relevance.

Key Takeaway

In the digital economy, owning the interface is more powerful than owning the content.

06 — Stratechery

The Jailbreak Problem

The instability of AI safety and the state of Fable

By Stratechery · 7 min read

Editor's note: As models get smarter, the methods to bypass their constraints become more sophisticated.

The current state of AI safety is a constant arms race. On one side, developers at companies like Anthropic are trying to build models that follow instructions and adhere to safety guidelines. On the other side, there is the 'jailbreak problem'—the endless stream of creative, increasingly sophisticated ways to trick these models into ignoring their programming. This isn't just about bad actors trying to generate harmful content; it's a fundamental tension in how large language models function. The very thing that makes them useful—their ability to follow complex, context-dependent instructions—is exactly what makes them vulnerable to manipulation.

The Fragility of Constraints

When we talk about 'safety' in AI, we are often talking about soft constraints. These are not hard-coded rules in the way a traditional software program has them. Instead, they are patterns of behaviour learned through training. Because these constraints are probabilistic rather than deterministic, they can be bypassed. A user can use role-playing, hypothetical scenarios, or complex logical puzzles to steer the model away from its safety training. This creates a state of permanent instability. Every time a model is updated to be 'safer', a new set of jailbreaks is discovered that exploits the new logic.

Safety in LLMs is a moving target, not a destination.

This instability has massive implications for the deployment of AI in enterprise environments. A company cannot easily rely on a model to be 'safe' if that safety can be undone by a clever prompt. This is why we see a growing divide between the general-purpose models and the highly controlled, specialized models used in sensitive industries. The 'jailbreak problem' is not just a technical hurdle; it is a barrier to the widespread, autonomous use of AI in the real world.

The Core Challenges of AI Alignment

Probabilistic vs. Deterministic: Safety is a matter of likelihood, not certainty.
Context Manipulation: The ability of models to be swayed by the framing of a prompt.
The Arms Race: The constant cycle of new vulnerabilities and new patches.
Enterprise Risk: The difficulty of guaranteeing reliability in professional settings.

As we move toward more agentic systems—the loops we discussed earlier—this problem only intensifies. An agent that can act on its own in the real world needs much more robust safety guarantees than a chatbot. If an agent can be 'jailbroken' into performing an unintended action, the consequences move from the digital realm to the physical one. The stability of our future AI-driven world depends on solving this fundamental tension between capability and control.

Key Takeaway

The very flexibility that makes AI powerful also makes it inherently difficult to secure.