The Deep Feed

01 — Lenny's Newsletter

The Software Factory: Beyond the Coding Assistant

How Warp is moving from helping engineers write code to managing entire production lines

By Claire Vo · 12 min read

Editor's note: As AI moves from autocomplete to autonomy, the role of the engineer is shifting from creator to supervisor.

The era of the coding assistant is ending. For the last two years, the industry has been obsessed with tools that help a developer write a single function or debug a specific error. These are helpful, but they are incremental. They do not change the fundamental physics of how software is built. Zach Lloyd, CEO of Warp, is betting on something different: the software factory. Instead of a person sitting at a terminal, typing commands and waiting for a compiler, the factory is a continuous loop that takes a requirement from a Slack message and turns it into a merged pull request. This is not about making a person faster; it is about making the process autonomous.

The Death of the Sequential Workflow

Traditional engineering follows a predictable, linear path: a ticket is created in Linear, a developer pulls the code, writes the logic, tests it locally, and pushes a pull request for review. It is a slow, human-heavy process. Warp’s 'Wilson' factory breaks this sequence. When a trigger occurs—perhaps a bug report in Sentry or a request in Slack—the factory begins working in parallel. It doesn't just suggest code; it executes the entire workflow. It manages the version control, runs the tests, and handles the QA. The engineering leader is no longer looking at a list of tasks, but at a dashboard of agentic runs.

The bottleneck is no longer writing code; it is the human capacity to review it.

This shift creates a new problem: the review bottleneck. If an AI factory can ship 2,000 pull requests a month, the human engineers will be buried under a mountain of code they didn't write. To solve this, the factory must become self-improving. Warp uses an 'LLM-as-a-judge' system to score every agent run. If a run fails, the system doesn't just stop; it analyses the failure mode, attempts to fix its own workflow, and tries again. This turns the engineering process into a feedback loop where the software itself learns how to be a better developer.

The Factory Framework

Trigger: Input from Slack, Sentry, or Linear
Execution: Parallel agentic tasks rather than sequential steps
Evaluation: LLM-as-a-judge scoring every output
Optimization: Using cost-quality Pareto charts to select the right model

The economic reality of this shift is found in the Pareto chart of model selection. Not every task requires the intelligence of GPT-4o or Claude 3.5 Sonnet. A factory must be able to match the complexity of a task to the cost of the model. Writing a unit test for a simple utility function is cheap; architecting a new microservice is expensive. By tracking the cost per pull request across different model configurations, companies can scale their output without their compute bills spiraling out of control. We are moving toward a world where the primary skill of a CTO is not managing people, but managing the throughput and cost of an automated production line.

Key Takeaway

The future of engineering is not about writing code, but about designing the systems that write the code.

02 — Aeon

The Acoustic Erasure of Borneo

How the construction of Indonesia's new capital is silencing the rainforest

By Aeon · 8 min read

Editor's note: Conservation is often thought of in terms of what we see, but what we hear is just as critical to the survival of an ecosystem.

For Abidin, an elder in East Kalimantan, the rainforest is not a silent space. It is a dense, layered communication network. The calls of birds, the rustle of insects, and the specific rhythms of the canopy are the maps he uses to navigate his world. But the map is being erased. As the Indonesian government moves forward with Nusantara—the massive new capital city designed to replace the sinking Jakarta—the sounds of construction are beginning to drown out the natural world. This is not just a change in volume; it is a fundamental disruption of the biological signals that wildlife relies on to survive.

Bioacoustics as a Record of Loss

Researchers are racing against the noise. Bioacoustic scientists, such as those from the Cornell Lab of Ornithology, are working with local residents to record the sounds of the rainforest before they are altered forever. They are placing microphones in caves and high in the treetops, capturing a digital archive of an ecosystem in transition. These recordings serve a dual purpose: they provide a baseline for scientists to measure the impact of urbanisation, and they preserve the ancestral knowledge embedded in the sounds of the land.

We are not just losing species; we are losing the acoustic history of a living world.

The danger of urban encroachment is often measured in hectares of cleared trees, but the sonic impact is more insidious. When the frequency of a forest changes—when the low rumble of heavy machinery replaces the calls of primates—the entire social structure of the wildlife can collapse. Animals may abandon territories not because the food is gone, but because the communication lines are broken. The noise creates a wall, effectively fragmenting the habitat even if the trees are still standing.

The Threats to the Soundscape

Mechanical noise from heavy construction equipment
Habitat fragmentation caused by new road networks
The displacement of specialist species by generalist urban dwellers

The project to record Nusantara's surrounding wilderness is a struggle against time. As the city grows, the window for capturing the 'true' sound of the Borneo rainforest closes. What remains will likely be a hybrid soundscape: a mixture of biological calls and the persistent hum of a modern metropolis. For people like Abidin, this is more than a scientific loss; it is a cultural one. The sounds he grew up with are the foundation of his relationship with the land, and as they fade, so does the connection to the history of his people.

Key Takeaway

The survival of an ecosystem depends as much on its soundscape as it does on its physical structure.

03 — Aeon

The Wisdom of Probabilia

Why Cicero's ancient scepticism is the best tool for a chaotic modern world

By Massimo Pigliucci · 15 min read

Editor's note: In an age of extreme polarisation, the middle ground isn't a place of weakness, but a place of intellectual rigour.

Most people treat belief like a binary switch. You either hold a position with armoured certainty, or you reject it with the performative doubt of a professional contrarian. This is a false choice. It leaves no room for the reality of human knowledge, which is almost always provisional. In 45 BCE, as the Roman Republic was collapsing around him, Marcus Tullius Cicero proposed a different way to live: the path of the Academic Sceptic. He didn't advocate for the denial of truth, but for the pursuit of 'probabilia'—the most probable accounts of reality.

Scepticism vs. Contrarianism

The term 'sceptic' has been hijacked in recent decades. Today, it is often used as a shield for denialism—climate sceptics, vaccine sceptics, and so on. These individuals are not actually enquirers; they are dogmatists who have reached their conclusions in advance and use doubt as a rhetorical weapon to stall action. A true sceptic, by contrast, is someone who is willing to update their position when presented with new evidence. The contrarian seeks to obstruct; the sceptic seeks to understand the strength of the evidence.

A genuine sceptic updates when the evidence changes; a contrarian merely shifts the goalposts.

This leads to the concept of calibrated confidence. To live by 'probabilia' is to acknowledge that while we cannot be 100% certain of anything, we can be 70%, 90%, or 99% sure. This allows for decisive action without the arrogance of absolute truth. It is the difference between saying 'This is the only way' and 'This is the best available account, and I will act on it while remaining open to being wrong.' This is how science actually functions, and it is how effective leadership must operate.

The Three Pillars of Probabilism

Provisional Commitment: Acting on the best available data
Calibrated Confidence: Assigning a degree of certainty to beliefs
Intellectual Humility: The readiness to pivot when evidence shifts

The value of Cicero's approach is most evident in times of crisis. When the world feels unstable, the instinct is to grab onto something solid—any dogma, any certainty, no matter how flawed. But the most resilient thinkers are those who can navigate the grey areas. They do not seek the comfort of being right; they seek the utility of being accurate. By adopting a philosophy of provisional but committed belief, we can make better decisions in an uncertain world without falling into the trap of fanaticism.

Key Takeaway

Intellectual strength is found in the ability to hold a position with confidence while remaining ready to abandon it.

04 — Lenny's Newsletter

The Vocabulary of Trust

How Meta's Muse is redefining the consumer AI experience

By Lenny Rachitsky · 10 min read

Editor's note: The next frontier of AI isn't more capability; it's better design and more intuitive trust.

Most AI agents feel like tools for engineers. They use technical language—crons, artifacts, plugins, and connectors—that creates a barrier to entry for the average person. Meta's Muse is attempting to break this barrier. Instead of a technical interface, Muse organises itself around human concepts: ideas, goals, and a library. This is a deliberate design choice. By changing the vocabulary, Meta is changing who feels invited to use the technology. They are moving AI from a specialized utility to a consumer companion.

The Permission Economy

Trust in AI is not a vague feeling; it is built through specific interaction patterns. One of the most significant challenges for consumer agents is how they handle sensitive information. Muse avoids the two extremes of 'unlimited access' and 'constant interruption'. Instead, it asks for permission at the exact moment access becomes relevant. When it reads an email to build a profile, it summarises what it learned and asks for consent. This creates a sense of agency for the user, rather than a feeling of being surveilled.

The words a product uses determine who feels comfortable using it.

Transparency is another pillar of this trust. While most consumer products hide their internal processes behind a polished result, Muse provides an activity feed. This feed shows the tool calls, scripts, and searches the agent performed. While the average user might ignore it, the presence of this data provides a 'paper trail' that makes the agent's actions feel grounded and verifiable. It moves the experience from 'magic' (which is suspicious) to 'process' (which is understandable).

Design Principles for Consumer AI

Human-centric language over technical jargon
Contextual permission requests
Visible process through activity feeds
Goal-oriented rather than task-oriented interaction

However, the technology still faces a significant hurdle: the unreliability of the open web. While Muse excels at managing calendars and generating newsletters, it struggles with the messy, unpredictable nature of browser-based shopping. The inability to reliably complete a purchase for a specific item is a reminder that until agents can navigate the web with the same fluidity as a human, they will remain assistants rather than true proxies. The gap between 'helpful' and 'indispensable' lies in the ability to handle these low-level, high-friction tasks.

Key Takeaway

Consumer AI success will be defined by design and trust, not just raw computational power.

05 — Lenny's Newsletter

The Product Oxymoron

Unfiltered lessons from the front lines of Snapchat and Discord

By Lenny Rachitsky · 14 min read

Editor's note: Great product management often looks like chaos to those on the outside.

Peter Sellis, a veteran of both Snapchat and Discord, argues that the standard model of product management is often broken. In many organisations, the PM is a glorified project manager, a person who sits in meetings, writes tickets, and ensures that features are shipped on time. This is not product management; it is administration. To build products that actually move the needle, Sellis suggests a more radical approach: designing teams that function like highly autonomous, mission-driven cells.

Systems Thinking over Feature Lists

The most common mistake in product leadership is focusing on the feature roadmap rather than the underlying system. Growth does not come from a clever new button or a marketing campaign; it comes from the core product mechanics. If the core loop of the product is not engaging, no amount of feature additions will save it. Sellis advocates for 'systems thinkers'—people who understand how a change in one part of the product will ripple through the entire ecosystem. This is a rare skill that separates the mediocre PM from the elite leader.

Growth almost always comes from the core product, not from the marketing department.

This requires a level of autonomy that many corporations find uncomfortable. Sellis describes high-performing teams as being almost like 'terrorist organisations'—small, fast, and operating with a level of independence that allows them to strike at problems and move on before the bureaucracy can react. This isn't about being destructive; it's about being agile. It's about removing the layers of permission that stifle innovation and replace them with a clear sense of mission and accountability.

The Sellis Framework for Product Leadership

Prioritise core product loops over feature bloat
Empower small, autonomous, mission-driven teams
Value systems thinking over specialist execution
Avoid the 'median PM' trap of administrative management

Ultimately, the goal of product management should be to create a platform for discovery and engagement, not just a checklist of delivered code. The best products are those that feel inevitable, where the features emerge naturally from the needs of the users and the logic of the system. Achieving this requires a rejection of the traditional, top-down management style in favour of a more organic, iterative approach that respects the complexity of human behaviour and the unpredictability of growth.

Key Takeaway

True product leadership is about managing systems and autonomy, not features and people.

06 — Psyche

The Sound of Memory

The tension between nostalgia and trauma in the history of choice

By Psyche · 7 min read

Editor's note: Documentaries often use aesthetic tools to bridge the gap between personal memory and historical reality.

In Courtney Stephens' film 'Lesser Choices', the past is not a distant country, but a series of overlapping textures. Through an interview with her mother, Estelle, Stephens explores the clandestine reality of seeking an abortion in 1969 Mexico City. This was a time before the legal protections of Roe v Wade, a time when reproductive agency was a matter of shadows and secrets. The film does not attempt to provide a political lecture; instead, it focuses on the raw, personal experience of a woman who was 'willing to do anything' to maintain control over her own life.

The Super 8 Aesthetic

The film's power lies in its visual dissonance. Stephens uses Super 8 footage of Mexico City—grainy, warm, and nostalgic—to frame a conversation that is fraught with tension and trauma. This creates a jarring effect. The visual language suggests a time of simplicity and charm, while the spoken word recounts a period of fear and high-stakes decision-making. This tension mirrors the complexity of memory itself: the way we often wrap our most difficult experiences in the soft, hazy light of the past.

The freedom of choice is often found in the most clandestine of spaces.

Estelle's story is one of two competing truths. There is the fear of the process—the physical and legal risks of navigating an underground medical system. And there is the profound sense of freedom that comes from being able to decide when and on what terms to grow a family. The film suggests that these two truths are not mutually exclusive; the fear is precisely what makes the freedom so significant. The risk taken was the price paid for autonomy.

Themes of Lesser Choices

The intersection of personal memory and political history
The use of aesthetic dissonance to convey emotional truth
The fundamental necessity of reproductive agency

By focusing on the intimate rather than the institutional, Stephens forces the viewer to confront the human cost of legal restrictions. The documentary is a reminder that history is not just a series of laws and court decisions, but a collection of individual lives lived under the pressure of those laws. The 'lesser choices' of the title refer not just to the medical options available, but to the limited ways in which women were able to navigate their own destinies in a world that sought to decide for them.

Key Takeaway

Autonomy is not a gift of the state, but a hard-won necessity of human existence.