The Deep Feed

01 — Lenny's Newsletter

The Digital Doppelgänger: Creating Avatars in Minutes

How AI video tools are turning non-creatives into production houses

By Lenny Rachitsky · 8 min read

Editor's note: A look at how the barrier to high-end video production has effectively collapsed.

The era of the expensive production studio is facing a quiet, rapid obsolescence. We are moving toward a reality where a single person with a laptop can generate a high-quality hype reel in the time it takes to drink a coffee. Claire, a creative who lacks traditional video production skills, recently put Google Flow and Gemini Omni to the test. The result was a one-minute video produced in fifteen minutes. This isn't just about speed; it is about the total redistribution of creative power. When the technical hurdles of lighting, framing, and editing are replaced by text prompts and face scans, the bottleneck shifts from 'how do I make this?' to 'what should I make?'

The Creative Partner vs. The Tool

Most people view AI as a hammer—a tool you pick up to hit a nail. But the workflow Claire experienced suggests something more collaborative. Google Veo did not just execute commands; it acted as a director. It suggested scene structures, asked about tone, and helped build a narrative arc. This is a fundamental shift in the creative process. Instead of starting with a blank canvas, the creator starts with a conversation. The AI proposes a seven-scene structure, and the human refines it. This partnership allows for a level of rapid iteration that was previously impossible, even for seasoned professionals.

The bottleneck is no longer technical skill, but the clarity of your own ideas.

However, the technology is not yet a perfect mirror. The 'uncanny valley' remains a stubborn obstacle. While the AI can replicate a face, it struggles with the subtle, messy realities of human emotion. A scene intended to show laughter can end up looking like a scripted performance by someone on heavy medication. There is a stiffness in the muscles, a lack of the micro-expressions that signal genuine feeling. Furthermore, character consistency is still a moving target. An avatar might change hair length or background details between shots, breaking the illusion of a continuous reality.

The New Production Workflow

Avatar Creation: Scanning the face to build a digital base.
Narrative Brainstorming: Using LLMs to storyboard and structure scenes.
Scene Generation: Converting descriptions into video clips.
Assembly: Stitching clips together in a streamlined editor.

For agency owners, this represents both a threat and an opportunity. The commodity work of basic video production is being devalued. You can no longer charge premium rates for simple social media clips that an intern can generate in fifteen minutes. The value moves upstream. The premium will be paid to those who can direct the AI, manage the narrative, and maintain the high-level taste that prevents the output from looking like generic, AI-generated sludge.

Key Takeaway

Technical skill is becoming a commodity; creative direction is becoming the premium.

02 — Lenny's Newsletter

The Curation Engine: Shopping for Longevity

Using Claude to bypass the noise of the modern internet

By Claire Vo · 7 min read

Editor's note: A practical guide to using AI to fight the decay of e-commerce quality.

The modern e-commerce experience is a minefield of low-quality goods. Amazon and other major platforms have become dominated by drop-shippers, paid advertisements, and brands that exist only as Instagram ads. Finding a well-made can opener or a pair of boots that will last a decade now requires an exhausting amount of research. We are drowning in convenience but starving for quality. The problem is that the algorithms are designed to show you what is popular or profitable, not what is durable or well-crafted.

Codifying Taste

Nicole Ruiz has solved this by turning Claude into a personal procurement officer. She doesn't use AI to find 'stuff'; she uses it to enforce her standards. By building a 'Claude Project,' she has codified her purchasing criteria: natural materials, repairability, brand history, and strong return policies. When she needs to buy something, she doesn't search the open web. She asks Claude to search through a pre-vetted list of trusted vendors. This flips the shopping model from a search for the cheapest option to a search for the most compliant option.

AI allows us to search for quality in a world designed for volume.

This approach also solves the UX problem of heritage brands. Many of the best manufacturers in the world—companies that have existed for a century—have websites that look like they were built in 1998. They are difficult to navigate and terrible for mobile users. Claude acts as a sophisticated interface for these archaic systems. It can parse their product descriptions, find the relevant details, and answer questions that the website's own search function cannot. It brings the efficiency of a modern tech giant to the artisans of the old world.

The Anti-Noise Shopping Framework

Define Criteria: Write down exactly what 'quality' means to you.
Build a Trusted List: Curate a database of brands with proven histories.
Use AI as a Filter: Query the AI to vet new products against your list.
Automate the Aftermath: Use tools like Claude Cowork to manage returns and receipts.

This is more than just a way to buy better goods; it is a way to reclaim mental bandwidth. The 'anti-to-do list' concept involves using AI to handle the repetitive, low-value cognitive tasks of household management. Instead of spending hours comparing specs or hunting for receipts, you delegate the search and the administrative follow-up to the machine. You aren't just buying a product; you are buying back your time and your attention.

Key Takeaway

Use AI to enforce your standards, not just to satisfy your impulses.

03 — Cal Newport

The Myth of the Job Apocalypse

Why AI is transforming roles rather than erasing them

By Study Hacks · 6 min read

Editor's note: A reality check on the most common fear in the professional world.

The narrative surrounding AI and employment has been consistently apocalyptic. CEOs of major AI labs have predicted the end of entry-level white-collar work, drawing parallels to the industrial revolution's impact on manual labour. The fear is that if machines can think, humans will become redundant. However, the reality on the ground is proving to be far more subtle. We are not seeing a mass exodus of workers; we are seeing a massive shift in how those workers spend their time.

Vibe Coding and the New Hack

Cal Newport observes that instead of replacing entire jobs, AI is being used by small business owners and professionals to 'hack' their operations. He compares this to his own teenage years, where he would write quick-and-dirty software to solve specific, annoying problems for a consulting firm. Today, people are using AI to build custom tools on the fly—what some call 'vibe coding.' These aren't polished, enterprise-grade software suites; they are useful, bespoke solutions that make a specific task easier. This isn't the arrival of the power loom; it's the arrival of a better screwdriver.

AI is not making humans superfluous; it is making their efforts a little deeper.

The shift is from 'execution' to 'orchestration.' In the past, a significant portion of a professional's day was spent on the mechanical aspects of their role—formatting reports, searching for data, or writing boilerplate code. AI absorbs these tasks. This doesn't eliminate the need for the professional; it changes the nature of their contribution. The professional must now move from being the person who does the work to the person who ensures the work is correct, strategic, and aligned with the goal.

The Real Impact of AI on Labor

Task Automation vs. Job Replacement: Most AI impact is at the task level.
The Rise of the Generalist: Ability to use multiple tools expands capability.
Increased Depth: Professionals can focus on high-level strategy.
The New Entry Level: The challenge of how juniors learn without 'grunt work'.

The danger isn't that there will be no jobs, but that the barrier to entry will change. If the 'grunt work' that traditionally trained juniors is automated, how do they gain the expertise required to become seniors? This is the real structural challenge for the knowledge economy. We must figure out how to build expertise in an era where the foundational steps of the ladder have been removed by automation.

Key Takeaway

AI replaces tasks, not people—but it fundamentally changes the path to expertise.

04 — Not Boring

Expanding the Radius of Daily Life

How transportation technology dictates the boundaries of human existence

By Packy McCormick · 9 min read

Editor's note: A philosophical look at why we haven't seen flying cars and what they would actually do.

We have been asking 'where are the flying cars?' for decades. The question is often treated as a joke or a sign of technological stagnation. But the absence of VTOL (vertical takeoff and landing) technology isn't just a failure of engineering; it is a failure to understand the anthropological impact of movement. When we talk about flying cars, we shouldn't be thinking about avoiding traffic on the way to the office. We should be thinking about the expansion of the human radius.

Marchetti’s Constant

There is a mathematical principle known as Marchetti’s Constant. It suggests that as transportation improves, humans do not spend less time traveling; they simply travel further. A person who once walked twenty minutes to work now drives forty minutes to a more distant suburb. We use the time saved by technology to expand our territory. This is an anthropological invariant. We are 'naked apes' with a biological drive to spread. Faster movement doesn't give us more leisure; it gives us more distance.

Humans don't use faster transport to travel less; we use it to travel further.

This is why the promise of companies like Vight is so significant. They aren't just building drones; they are building the infrastructure for a new way of living. If the radius of daily life expands, the very concept of the 'city' changes. The density that defines urban life is a response to the friction of distance. If that friction is removed, the distinction between urban and rural begins to dissolve. We are looking at a future where geography is less of a constraint on opportunity and more of a choice of environment.

The Implications of Expanded Mobility

Decentralization: The potential for more spread-out living patterns.
Economic Reach: Access to wider markets and labor pools.
Infrastructure Shift: Moving from roads to air corridors.
Territorial Expansion: The natural human tendency to inhabit more space.

The bet being made by frontier entrepreneurs is that we are on the verge of breaking this cycle of stagnation. The convergence of electric propulsion, autonomous flight, and advanced materials is creating a window of possibility that didn't exist twenty years ago. The goal isn't to make the commute faster; it is to make the world smaller, allowing us to inhabit a much larger portion of it with the same amount of effort.

Key Takeaway

Technology doesn't save time; it expands the distance we are willing to cover.

05 — Lenny's Newsletter

The Taste Imperative

Why judgment is the ultimate moat in the age of AI

By Lenny Rachitsky · 10 min read

Editor's note: Insights from Tony Fadell on why technical skill is secondary to product judgment.

Tony Fadell, the man behind the iPod and Nest, is offering a warning to the current generation of builders. As AI makes it easier than ever to generate code, designs, and marketing copy, the value of 'making' is plummeting. When anyone can generate a functional product, the differentiator is no longer the ability to build, but the ability to decide what is worth building. In an era of infinite output, taste and judgment become the only real moats.

The Danger of Cognitive Surrender

Fadell identifies a critical risk: cognitive surrender. This happens when builders stop questioning the output of their tools and start accepting the path of least resistance. If an AI suggests a feature or a design, the temptation is to implement it because it is 'good enough.' But 'good enough' is the death of great products. The history of the iPhone was defined by heated, often painful debates over things as simple as a physical keyboard. Those decisions weren't made by algorithms; they were made by people with a specific vision of how a human should interact with a device.

The biggest risk to product builders today is the surrender of their own judgment to the machine.

Marketing is often treated as a secondary concern in engineering-led companies, but Fadell argues it is as vital as the product itself. A great product that no one understands is a failure. In the AI era, the ability to communicate the 'why' behind a product becomes even more important. As the volume of products increases, the ability to stand for something—to have a clear, opinionated stance—is what will allow a brand to cut through the noise.

The Builder's New Toolkit

Opinionated Design: Making hard choices instead of averaging options.
Voice as Interface: Preparing for a world where interaction is conversational.
Judgment over Execution: Focusing on the 'what' and 'why' rather than the 'how'.
Resisting the Default: Avoiding the trap of AI-generated mediocrity.

For agency owners and product leaders, the lesson is clear. Do not compete on the ability to produce volume. The machines will always win that race. Instead, compete on the ability to provide direction. Your value lies in your ability to look at ten AI-generated options and know exactly which one will resonate with a human being. Taste is not a soft skill; it is a hard competitive advantage.

Key Takeaway

When execution becomes free, judgment becomes the most valuable asset.

06 — Chase AI · Video

Adversarial Coding: The New Quality Control

Using multi-agent systems to ensure software integrity

By Chase AI · 6 min read

Editor's note: A technical deep-dive into the next generation of AI coding workflows.

The current state of AI-assisted coding is plagued by a single, recurring problem: misalignment. You give an agent a task, it spends five minutes 'thinking,' and then it delivers code that is technically functional but fundamentally wrong for your specific architecture. The issue isn't that the AI is incapable; it's that the AI cannot be trusted to evaluate its own work. To solve this, the next generation of developers is moving away from single-agent workflows toward adversarial, multi-agent systems.

The Knowledge Graph Approach

One of the most effective ways to improve AI coding is to provide it with a better map. Tools like Graphify are moving beyond simple text searches (grep) to create actual knowledge graphs of a codebase. Instead of the AI guessing how different modules connect, it can see the entire structure visually and logically. This reduces token consumption and, more importantly, increases the accuracy of the AI's reasoning. It allows the agent to understand the 'why' behind the connections, not just the 'what'.

The most common failure in software development is misalignment between intent and execution.

But even with a perfect map, the risk of error remains. This is where 'Grill Me' and 'Grill with Docs' come in. These are advanced planning modes that force a deeper level of alignment before a single line of code is written. Instead of a quick three-question chat, these tools subject the developer's intent to a rigorous interrogation. They ask the hard questions that expose gaps in logic, ensuring that both the human and the AI are working from the same mental model.

The Adversarial Workflow

Knowledge Mapping: Using graphs to give the AI a structural understanding of the repo.
Deep Planning: Using 'Grill' modes to force alignment before execution.
Adversarial Review: Bringing in a second, independent AI (like Codex) to attack the plan.
Multi-Round Verification: Running multiple cycles of back-and-forth between agents to reach consensus.

The ultimate evolution of this is the 'adversarial review.' In this setup, you don't just ask an AI to write code; you task a second AI with finding the flaws in the first one's plan. This creates a productive tension. The two agents go back and forth—sometimes up to five rounds—until they both agree on the best path forward. This mimics the human process of peer review and significantly raises the floor of code quality. For the professional developer, the job is shifting from writing code to managing this high-stakes debate between machines.

Key Takeaway

Don't trust a single AI; build a system of competing agents to verify the truth.