The Deep Feed

01 — Lenny's Newsletter

The Harness: Moving Beyond the Prompt

Why general-purpose AI is a blunt instrument for professional workflows

By Claire Vo · 8 min read

Editor's note: As AI moves from novelty to utility, the real winners won't be those who prompt well, but those who build the machinery around the model.

The current obsession with 'prompt engineering' is a distraction. It treats large language models like magic lamps rather than the engines they are. When a professional spends their day typing 'please fix this bug' into a chat window, they aren't working; they are babysitting. The real leap in productivity happens when we stop talking to the model and start building around it. This is the concept of the 'harness'—a dedicated layer of software that wraps an AI agent in specific permissions, tools, and structured workflows. It turns a general intelligence into a specialist that knows exactly how to interact with your specific stack of tools, from Sentry to GitHub.

The Architecture of Agency

A functional harness requires more than just an API key. It needs four distinct components to move from a chatbot to an agent. First, there are the runs: the execution context. Second, the tasks: the specific, repeatable objectives the agent must achieve. Third, the tools: the adapters that allow the AI to actually touch the real world—reading a ticket in Linear or pushing code to Vercel. Finally, there are the artifacts: the structured outputs that the team can actually use, rather than a wall of conversational text. Without these, you are just playing with a very expensive toy.

The real leap in productivity happens when we stop talking to the model and start building around it.

Building such a system involves a specific kind of technical orchestration. You might use a model like Claude Sonnet to power the logic inside the harness, while using a more powerful model like Opus to actually write the harness code itself. The goal is to encode specific permissions so the agent cannot wander off-task. For instance, a bug-triage harness shouldn't just 'look' at errors; it should be able to gather evidence from logs, perform a root-cause analysis, and then create a follow-up task in your project management tool. It is the difference between a person who can read a map and a driver who knows the route.

The Three Pillars of a Harness

Contextual Runs: Defining the specific environment and boundaries of the task.
Opinionated Adapters: Custom connections to your specific software stack (Sentry, GitHub, etc.).
Structured Artifacts: Ensuring the output is a usable file or ticket, not just a chat response.

The transition to harness-based work marks the end of the 'chatbot era' for serious enterprises. We are moving toward a world where AI is a silent, integrated part of the infrastructure. In this new reality, the competitive advantage lies not in how well you can talk to a machine, but in how effectively you can constrain and direct it to perform the repetitive, structured work that currently drains your best talent.

Key Takeaway

Stop prompting and start building the machinery that makes prompting unnecessary.

02 — Lenny's Newsletter

The Great Bifurcation

How AI is splitting the tech workforce into the amplified and the shaken

By Noam Segal, Lenny Rachitsky · 10 min read

Editor's note: The data shows a workforce in crisis, not because of job loss, but because of a fundamental shift in professional identity.

The tech industry is no longer a monolith. A massive survey of workers in 2026 reveals a workforce splitting in two distinct directions. On one side are the 'amplified': professionals who use AI to expand their capabilities, feeling more confident and capable than ever before. On the other side are those who feel 'shaken': individuals who see AI as a force that destabilizes their value and threatens their place in the industry. This isn't just a difference in opinion; it is a divide that predicts everything from career optimism to burnout levels.

The Burnout Paradox

While 82% of workers report that AI makes them more productive, this efficiency is not translating into happiness. In fact, burnout has jumped significantly in the last year. The fear isn't necessarily that a robot will take your job; it's that a robot will make you do three times as much work for the same pay. There is a growing anxiety that the 'productivity gains' promised by AI will simply be captured by management in the form of higher expectations and an unsustainable pace of output.

The defining feeling about AI is not fear of replacement, but the dread of an endless, automated treadmill.

The demographic split is telling. Designers and researchers report the highest levels of anxiety and the lowest willingness to recommend their fields to newcomers. They are seeing the tools that once required deep, human intuition being mimicked by models, leading to a sense of diminished value. Conversely, founders and those in small companies remain the happiest, likely because they have the agency to use these tools to build rather than just to comply with existing corporate structures.

The Four Identity Groups

Amplified: Using AI to do more and better work.
Redefined: Seeing the role change shape without a clear positive or negative sentiment.
Destabilized: Feeling unsure of where they stand or what is uniquely theirs.
Diminished: Feeling less essential or valuable in the professional ecosystem.

This bifurcation suggests that the 'AI revolution' is actually a massive re-sorting of human capital. Those who can bridge the gap between human intent and machine execution are thriving. Those who find themselves competing directly with the machine's output are being pushed into a state of permanent instability. The question for the next decade is not whether AI will replace workers, but which workers will be able to maintain their sense of agency in an automated world.

Key Takeaway

Your career survival depends on whether you use AI to expand your agency or let it dictate your pace.

03 — Experimental History

The Cost of Safety

Why prosperity is killing the deviant

By Adam Mastroianni · 9 min read

Editor's note: As we become more prosperous, we are trading the chaos of innovation for the safety of the status quo.

There is a quiet tragedy occurring in our modern social fabric: the decline of deviance. Since the 1990s, the markers of risk-taking and rule-breaking—from substance abuse to crime—have plummeted. While this is objectively a victory for public safety, it comes with a hidden tax. Deviance is often the precursor to innovation. When we eliminate the 'weird' and the 'risky,' we also eliminate the friction that produces new ideas. We are becoming a society that is safer, more predictable, and significantly more boring.

The Prosperity Trap

The decline isn't a result of better policing or more effective social programs. It appears to be a byproduct of increasing prosperity. When people have more to lose, they act with more caution. As the stakes of life rise, the appetite for risk falls. We have effectively traded the artistic and social golden ages of previous, more chaotic generations for a stable, middle-class existence where the primary goal is to avoid error. We are living in the aftermath of an unearned victory.

We have traded the possibility of greatness for the certainty of safety.

Some blame the internet, suggesting that the digital panopticon makes everyone afraid to be strange. But the data suggests otherwise. The decline of deviance was well underway long before the smartphone became ubiquitous. Instead, the internet has acted as an accelerator for three specific trends: cultural carcinization, where ideas converge on a few safe strategies; de-frictioning, which removes the obstacles that force people to adapt; and the 'flat-Earth' problem, where algorithmic curation prevents us from seeing anything outside the norm.

The Drivers of Cultural Flattening

Carcinization: The convergence of creative strategies into a few 'proven' models.
De-frictioning: The removal of the struggle required for true growth.
Algorithmic Curation: The narrowing of the human experience to what is most likely to be consumed.

If we want to see a resurgence of genuine innovation, we cannot simply demand it. We have to find ways to reintroduce productive friction into our lives. We need to create spaces where it is safe to be wrong, where the cost of failure isn't total social or economic exile. Otherwise, we will continue to produce endless sequels to the same stories, living in a world that is perfectly safe and entirely hollow.

Key Takeaway

A society without risk is a society without a future.

04 — The Marginalian

The Soul of the Machine

Revisiting the boundaries of consciousness and the animal spirit

By Maria Popova · 7 min read

Editor's note: As we build increasingly complex intelligence, we must ask if we are missing the very thing that makes life meaningful.

We have long held the dogma that the soul is a uniquely human invention—a tool we created to explain our capacity for art, love, and existential dread. But this distinction is increasingly fragile. As our understanding of animal consciousness expands, through everything from the dreaming patterns of birds to the complex social lives of octopuses, the line between 'human' and 'animal' begins to blur. We are forced to ask: is the soul a biological miracle, or merely a biological byproduct?

Beyond Survival

To speak of a soul is to speak of something that exists beyond the mere struggle for survival. It is the capacity for beauty, the awareness of mystery, and the ability to experience wonder. When we witness an animal displaying curiosity, grief, or even a sense of play, we are seeing the outward manifestations of an inner life. To deny them a soul is often not a scientific conclusion, but a failure of our own language and imagination—a refusal to acknowledge a depth we cannot measure.

Soul is what makes each of our lives a microcosm—not merely a meaningless fragment of the universe, but a reflection of the whole.

The concept of the soul does not require a religious framework. It can be understood as the 'marrow of existence'—the quality that makes a life feel significant. It is found in the moments of profound connection, the awe felt under a night sky, and the capacity for irrational kindness. If we define the soul by these qualities, then it is not a substance, but a way of being that is shared across the spectrum of sentient life.

Signs of the Non-Human Soul

Individual Agency: The presence of unique likes, dislikes, and temperaments.
Social Complexity: The capacity for friendship, loyalty, and altruism.
Cognitive Wonder: The expression of curiosity and engagement with the environment.

As we enter an era of synthetic intelligence, this question becomes urgent. If we can build machines that mimic our reasoning, will we also attempt to build machines that mimic our 'soul'? Or will we realize that the soul is not found in the complexity of the logic, but in the vulnerability of the living? The answer will define how we relate to everything we create.

Key Takeaway

The soul is not a thing to be measured, but a quality to be recognized in the living.

05 — The Marginalian

The Architecture of Understanding

Thich Nhat Hanh's remedy for the friction of human perception

By Maria Popova · 6 min read

Editor's note: In an age of instant reaction, the most radical act is to pause and question our own assumptions.

The human brain is a storytelling machine, but it is a flawed one. We do not react to reality; we react to the stories we tell ourselves about reality. When someone hurts us, we immediately construct a narrative of motive, intent, and injustice. Most of these stories are wrong. They are built on our own fears, our own insecurities, and our own cognitive biases. We live in a self-generated dream, responding to shadows rather than the substance of the world.

The Error of Perception

The great Buddhist teacher Thich Nhat Hanh argued that much of human suffering stems from these wrong perceptions. We assume we know why someone acted a certain way, and we build our resentment on that assumption. This is a form of self-inflicted harm. By clinging to a false narrative, we close the door to connection and trust. The remedy is not to fight the other person, but to fight our own misconceptions.

When we make the effort to listen and hear the other side of the story, your understanding increases and your hurt diminishes.

Hanh proposed a three-step process for repairing relationships that is as practical as it is profound. It begins with internal acknowledgement: admitting that your current picture of events might be inaccurate. It continues with a vulnerable request for help: approaching the other person not with an accusation, but with a confession of your own suffering and a request for clarity. Finally, it requires the hardest step: listening with the genuine intent to understand, rather than to defend.

The Three Steps to Repair

Internal Pause: Breathe and acknowledge that your perception may be flawed.
Vulnerable Inquiry: Tell the other person you are suffering and ask for their help in understanding.
Deep Listening: Listen to their response without the urge to counter-argue.

This approach requires a level of humility that is often absent in modern discourse. It demands that we prioritize truth over being 'right.' In doing so, we find that most conflicts are not battles between good and evil, but collisions between two people, both of whom are victims of their own incomplete perceptions. To listen is to perform an act of profound care—for the other, and for ourselves.

Key Takeaway

Most conflict is a battle between two people fighting shadows of their own making.

06 — Stratechery

The Xbox Failure

Bundling, sunk costs, and the limits of the subscription model

By Stratechery · 5 min read

Editor's note: Microsoft's struggle with Game Pass reveals the inherent difficulty in maintaining value through pure bundling.

Microsoft's recent layoffs in the Xbox division are a stark signal that the Game Pass strategy is hitting a wall. For years, the industry has been chasing the 'internet solvent'—the idea that bundling enough content together will create a recurring, unstoppable revenue stream. But bundling is not a magic bullet. As the cost of producing high-end content rises, the math of the subscription model begins to break. You cannot simply add more content to justify a price; eventually, the cost of the content outstrips the value of the bundle.

The Sunk Cost of Strategy

The problem for Xbox is one of coordination and transaction costs. To make a subscription work, you need a critical mass of both content and users. But as the market becomes more crowded, the cost of acquiring those users increases, and the quality of the content required to keep them becomes astronomical. Microsoft has found itself trapped in a cycle of trying to out-spend competitors to maintain a bundle that is increasingly difficult to sustain without massive, constant investment.

Bundling works until the cost of the parts exceeds the value of the whole.

This is a classic case of the sunk cost fallacy playing out at a corporate scale. Having invested billions into the infrastructure of Game Pass, the company is hesitant to pivot, even as the underlying economics shift. The strategy assumes that users want access to everything, but in reality, users often want access to the *best* thing. When the 'best' thing is a single, expensive title, the bundle becomes a secondary, less valuable consideration.

Why Bundling Fails

Content Inflation: The rising cost of AAA titles outpaces subscription growth.
Value Dilution: Too much mediocre content can devalue the premium offerings.
Coordination Friction: The difficulty of aligning content release cycles with user retention needs.

The Xbox situation serves as a warning to all platform players. The era of easy growth through aggregation is ending. The next phase of the digital economy will likely favor those who can manage the tension between scale and specificity—those who can provide deep, focused value rather than just a broad, shallow collection of assets.

Key Takeaway

Aggregation is a strategy for growth, but specificity is the strategy for survival.