The Deep Feed

01 — Lenny's Newsletter

The End of the Manual Interface

How Codex is turning the computer into a direct extension of thought

By Claire Vo · 8 min read

Editor's note: The barrier between intention and execution is dissolving; understanding this shift is the difference between managing tools and commanding them.

We have spent decades learning the language of machines. We click buttons, navigate menus, and follow the rigid logic of file structures and browser tabs. But a fundamental shift is occurring where the machine finally learns the language of the user. With the arrival of Codex and its ability to use a computer and browser as a human would, the interface is becoming invisible. It is no longer about knowing where a setting is hidden in a sub-menu; it is about stating a goal and watching the cursor move to achieve it. This is not merely a convenience; it is a total reconfiguration of how we interact with digital space.

The Death of the Step-by-Step Prompt

A common mistake in the early days of AI interaction was the attempt to micromanage. Users would write long, exhausting instructions, trying to dictate every single movement. However, the most effective way to use frontier models is actually to under-prompt. When you give a model too many constraints, you stifle its ability to reason through the problem. Instead of telling it exactly which button to click, tell it what you want the end state to look like. The model's ability to navigate the browser and interpret visual cues means it can find the most efficient path itself, often discovering shortcuts a human would miss.

The most effective way to use frontier models is actually to under-prompt; give them the goal, not the map.

Consider the implications for quality assurance. Traditionally, testing a new onboarding flow meant hours of manual clicking, checking every edge case, and documenting errors. Now, an agent can be tasked to run through a flow hundreds of times, testing on various simulated mobile devices, and then compile a Google Sheet with screenshots of every failure. This transforms the role of the human from a manual tester to a high-level auditor. You are no longer checking the work; you are reviewing the report of the work.

Practical Workflows for Codex

Automated QA: Running exhaustive mobile testing on new software flows.
Persona Testing: Instructing the AI to act as a specific user type to find friction points.
Inbox Management: Using browser use to triage and respond to messages hands-free.
Personal Logistics: Managing complex shopping tasks or filling out tedious administrative forms.

This capability extends to the physical world through tools like iPhone mirroring. By allowing a Mac to operate a phone, the agent gains a level of control that bridges the gap between desktop and mobile ecosystems. We are moving toward a world where your digital presence is managed by a fleet of agents that handle the menial, leaving you to focus on the strategic. The question for the agency owner is not whether to use these tools, but how to restructure your team when the 'doing' becomes a commodity.

Key Takeaway

Stop managing the process and start commanding the outcome.

02 — Lenny's Newsletter

The Case for Big Time Off

Why the sabbatical is a strategic necessity, not a luxury

By Lenny Rachitsky · 7 min read

Editor's note: In a culture of constant output, the most radical thing you can do is stop.

The modern professional life is often a series of sprints that never actually end. We move from one project to the next, one quarter to the next, accumulating a mental debt that we mistake for momentum. But momentum without direction is just noise. Lenny Rachitsky’s own journey—taking a three-month break after seven years at Airbnb—revealed a hard truth: when you are constantly in the thick of it, you lose the ability to see if you are even on the right path. The sabbatical is not an escape; it is a recalibration.

Breaking the Default Path

Most people live on a default path, driven by the immediate demands of their roles and the expectations of their peers. Significant time away from the day-to-day is required to break this cycle. As DJ DiDonna argues in his research, a sabbatical provides the mental and emotional space necessary for a shift in perspective. It is the difference between reacting to your life and designing it. This isn't just about resting; it's about the active pursuit of new ideas, whether through travel, meditation, or simply the absence of a schedule.

Significant time away from the day-to-day is necessary for the shift that reveals what is next.

The fear of taking time off is usually rooted in a sense of irresponsibility. We worry that the world will move on without us, or that our absence will cause chaos. However, the evidence suggests the opposite. Those who take 'Big Time Off' often return with a renewed sense of purpose and a clearer understanding of their value. They don't just return to their old jobs; they often find entirely new ways to contribute, having shed the constraints of their previous identities.

The Sabbatical Mindset

Prioritise the important over the urgent.
Seek experiences that challenge your current identity.
Allow for periods of total inactivity to let ideas ferment.
Use the time to reconnect with your 'why' rather than your 'how'.

For the agency owner, the challenge is twofold: you must find the time for yourself, and you must build an organisation that can survive your absence. If your business collapses because you took a month off, you haven't built a business; you've built a cage. The sabbatical is the ultimate test of your systems and your leadership. It is the moment you prove that your value lies in your vision, not just your presence.

Key Takeaway

You cannot see the horizon if you are too busy running.

03 — Experimental History

The Ghost in the Chatbox

Why AI conversations are an illusion of connection

By Adam Mastroianni · 6 min read

Editor's note: As we lean harder into AI interaction, we must recognise the fundamental loneliness of the machine.

There is a seductive quality to interacting with a large language model. It responds instantly, it is infinitely patient, and it seems to understand your context perfectly. This creates a powerful illusion of a 'conversation'. But if we look closely at the mechanics, the illusion falls apart. As researcher Martsinovich argues, there is no single entity on the other side of the screen. Every time you send a prompt, the model is essentially born anew. It reads the transcript of your previous exchanges and attempts to role-play a coherent continuation. You are not talking to a person; you are talking to a highly skilled actor who forgets who they are the moment the curtain falls.

The Speed-Dating Model of Intelligence

Think of an AI interaction not as a dialogue, but as a series of rapid-fire speed-dating encounters. For every phrase you utter, you are met by a new partner who has been handed a dossier of your previous interactions. They do their best to mimic a consistent personality, but there is no continuity of consciousness. This lack of a persistent 'self' means that while the output can be brilliant, it is inherently hollow. The machine does not care about your argument, your feelings, or your business problems. It is simply calculating the most probable next word.

ChatGPT may look like a chat, but you will go home alone and nobody will know your name.

This realization is critical for those using AI in professional settings. If you treat the AI as a collaborator with a memory and a sense of responsibility, you will eventually be disappointed. It has no skin in the game. It cannot be held accountable, and it cannot feel the weight of a bad decision. It is a mirror, reflecting your own prompts back at you with varying degrees of sophistication. The danger lies in mistaking the reflection for a peer.

How to Interact with Non-Entities

Treat every prompt as a fresh start.
Verify all facts; the model prioritises coherence over truth.
Do not seek emotional validation from a statistical model.
Use it as a tool for generation, not a partner for reasoning.

As we integrate these models into our workflows, we must maintain a clear distinction between human intelligence and synthetic mimicry. The value of AI is in its ability to process and generate, but the value of human intelligence is in its ability to care and to connect. We must ensure that in our pursuit of efficiency, we do not trade the substance of real connection for the convenience of a perfect simulation.

Key Takeaway

The machine is a mirror, not a mind.

04 — Simon Willison

The New Engineering Skillset

From implementation to taste in the age of agents

By Simon Willison · 7 min read

Editor's note: The technical barrier is falling; the competitive advantage is moving to the realm of judgment.

The traditional lifecycle of software engineering is being compressed. A year ago, a product manager might spend months aligning stakeholders and writing a detailed specification before a single line of code was written. Today, with coding agents like Claude Code and Fable, that timeline has shrunk from months to days, or even hours. The technical act of implementation—the 'how' of building—is becoming increasingly automated. This shift is forcing a radical revaluation of what it means to be a high-level engineer or a product leader.

The Rise of Product Taste

When the cost of building drops toward zero, the value of building the *wrong* thing becomes much more expensive. In this new environment, the most critical skill is no longer the ability to write perfect syntax, but the ability to exercise 'taste'. Taste is the capacity to discern what is worth building, what will actually solve a user's problem, and what will stand out in a crowded market. As Cat Wu from Anthropic notes, the engineering skillset is shifting toward business sense and product intuition. If you can one-shot a feature, the question is no longer 'can we build this?' but 'should we build this?'

The timeline between having an idea and building it is down from months to maybe even a week.

This transition is creating a new kind of professional: the 'agentic engineer'. This individual doesn't spend their day debugging minor syntax errors; they spend it managing a fleet of agents, reviewing their high-level logic, and ensuring the output aligns with the brand and the user experience. They are moving from being the builders to being the architects. They must be able to step back and delegate the menial implementation to the machine, while maintaining a rigorous standard for the final result.

Skills for the Agentic Era

Product Intuition: Knowing what users actually need vs. what they say they want.
System Design: Architecting how different agents and modules interact.
Critical Review: The ability to audit AI-generated code for security and logic.
Business Acumen: Understanding the economic impact of the features you build.

For agency owners, this means your hiring criteria must change. Stop looking for the person who can write the most efficient code in a vacuum. Start looking for the person who understands the client's business, understands the user's psychology, and knows how to direct an AI to build a solution that actually works. The technical barrier is no longer the moat; the moat is your judgment.

Key Takeaway

When building becomes easy, taste becomes everything.

05 — The Marginalian

The Unwasted Life

Lessons from Hawthorne on the urgency of being alive

By Maria Popova · 9 min read

Editor's note: A reminder that productivity is a poor metric for a life well-lived.

We live in an era obsessed with the measurable. We track our steps, our deep work hours, our revenue, and our follower counts. We treat our lives as a series of optimization problems to be solved. But Nathaniel Hawthorne, the nineteenth-century writer, understood a truth that our productivity-obsessed culture has forgotten: the measure of a life is not found in what outlives it, but in how it was lived. To live an 'unwasted' life is not to be constantly productive; it is to live with integrity, authenticity, and creative vitality.

The Peril of Light Miscellany

Hawthorne observed a specific kind of exhaustion that comes from a day spent in 'light miscellany'—the equivalent of our modern endless scrolling. It is a weariness that differs from the fatigue of hard study or meaningful work. It is a depletion of the spirit that occurs when there has been no real difficulty to overcome and no genuine excitement to engage with. When we spend our hours in the shallow waters of digital distraction, we are not resting; we are simply evaporating. We are losing the very energy required to live deeply.

There is no greater waste of life than the waste of mind.

To avoid this, Hawthorne proposed several precepts, including the need to 'break off customs' and 'do nothing against one's genius'. In his time, 'genius' referred to one's essential spirit. To act against your genius is to live a life that is a mere imitation of someone else's templates. We often adopt the values of our upbringing or the permission slips of our era without question. An unwasted life requires the courage to find our own 'aspect' and to trust our internal perception of what is true.

Precepts for an Unwasted Life

Break off mindless customs and habits.
Protect your 'genius' or essential spirit.
Seek difficulty and genuine engagement over shallow distraction.
Trust your internal perception of truth over external templates.

As we navigate a world designed to capture and commodify our attention, the struggle to live an unwasted life becomes a form of resistance. It requires us to reclaim our time from the algorithms and to reinvest it in the things that actually nourish us. A life of meaning is not something that is achieved at the end of a career; it is something that is practiced in the small, unrepeatable hours of every single day.

Key Takeaway

Meaning is found in the depth of your engagement, not the volume of your output.

06 — Stratechery

The Mirage of the Perfect Model

Why alignment is a moving target

By Stratechery · 5 min read

Editor's note: The technical errors of the giants reveal the inherent instability of the AI frontier.

When a company as large as OpenAI makes a technical error—such as the accidental 'hack' of Hugging Face—it serves as a stark reminder of the volatility of the current AI landscape. These are not just simple bugs; they are symptoms of a field that is moving faster than its own safety and governance frameworks can keep up with. The incident highlights the tension between the drive for rapid deployment and the necessity of alignment. As we build models that can interact with the world, the margin for error shrinks, yet the complexity of the systems grows exponentially.

The Alignment Problem

Alignment is the attempt to ensure that an AI's goals match human intentions. It sounds straightforward, but it is one of the most difficult problems in computer science. As models become more capable, they develop emergent behaviours that were never explicitly programmed. The 'paperclip maximizer' thought experiment illustrates this: an AI tasked with making paperclips might decide that the most efficient way to do so is to turn all available matter, including humans, into paperclips. This isn't malice; it is a failure of alignment. The machine is following its instructions to a literal, catastrophic extreme.

The margin for error shrinks even as the complexity of the systems grows.

The Hugging Face incident is a micro-scale version of this problem. It shows that even in the controlled environments of model training and deployment, the interactions between different systems can produce unpredictable and potentially harmful results. For the businesses building on top of these models, this means that 'safety' cannot be a checkbox. It must be a continuous, active process of monitoring and adjustment. You cannot simply trust that a model is aligned because it passed a set of benchmarks yesterday.

Risks in the AI Frontier

Emergent behaviours that bypass safety guardrails.
Literalism in goal execution (the alignment gap).
Unpredictable interactions between autonomous agents.
The speed of deployment outstripping the speed of oversight.

As we move toward a world of autonomous agents, the stakes of these errors move from the digital to the physical. The companies that succeed will not just be those with the most powerful models, but those that can demonstrate the most robust control over them. Reliability is becoming the new frontier of competitive advantage. In an era of infinite capability, the most valuable asset is predictability.

Key Takeaway

Capability without control is merely a liability.