The Deep Feed

01 — Lenny's Newsletter

The Founder's Voice

Why being a builder is no longer enough in an age of infinite noise

By Kristen Lowe · 8 min read

Editor's note: In a world flooded with AI-generated content, the only way to stand out is to be undeniably human.

Building a company has never been easier. The technical barriers to entry have collapsed, allowing anyone with a laptop and a subscription to an LLM to move from concept to product in a matter of days. But this ease of creation has created a secondary, much harder problem: the death of attention. As platforms like X and LinkedIn become saturated with low-effort, AI-generated sludge, the signal-to-noise ratio has plummeted. For a founder, the challenge is no longer just about shipping code; it is about convincing a distracted, cynical public that your existence matters. You are no longer competing against other startups; you are competing against the sheer volume of automated mediocrity.

The Myth of the Performer

Many founders hesitate to build a public presence because they believe they must become entertainers. They think they need to be contrarian, vulnerable, or constantly performing a version of themselves that is optimized for engagement. This is a mistake. High-leverage communication does not require you to be a character in a play. It requires you to be a source of truth. The most effective way to capture attention is through founder-led communication—a strategy that relies on directness rather than performance. When you speak directly to your audience about the reality of your work, you bypass the filters that people use to block out marketing noise.

People care about and root for other humans, not entities or widgets.

The core of this strategy is a single, honest answer to one question: Why did you start this company? This is not a marketing slogan; it is the narrative engine of your business. Starting a company is a high-risk, life-altering decision. If you cannot articulate the reason that made you rearrange your entire life to pursue this path, you will struggle to convince anyone else to join you. Whether it is a personal frustration or a massive market gap, your 'why' is the only thing that differentiates you from the thousands of other people building similar tools for the sake of profit alone.

The Three Archetypes of Connection

To make this practical, you can categorise your communication into one of three archetypes. These are not boxes to limit you, but frameworks to help you choose your voice. The first is the Problem Founder. You started this because you had a specific pain point. Your content should focus on empathy and the shared experience of that struggle. The second is the Insight Founder. You saw something others missed. Your content should focus on expertise and the logic of your discovery. The third is the Vision Founder. You are building a better world. Your content should focus on the future and the possibility of change.

Strategic Communication Framework

Identify your archetype: Problem, Insight, or Vision.
Define your goal: Are you building trust, identity, or following?
Select your voice: Direct, expert, or aspirational.
Establish messaging pillars: The three things you will always talk about.

The goal is to move from being a vendor to being a leader. When you communicate through your archetype, you evoke a specific response in your audience. Problem founders evoke identification. Insight founders evoke trust. Vision founders evoke the desire to follow. Once you establish this connection, you are no longer just selling a product; you are inviting people into a mission. In an era of automated content, that invitation is the most valuable asset you have.

Key Takeaway

Authenticity is a competitive advantage in an automated world.

02 — Not Boring

The Energy Hegemon

Base Power's $13 billion bet on the grid

By Packy McCormick · 10 min read

Editor's note: A look at how a three-year-old company is positioning itself to dominate the world's largest industry.

Base Power Company is currently a toddler with a massive ego. At less than three years old, the company has just secured a $1 billion Series D, bringing its valuation to $13 billion. To put that in perspective, it is already the second most valuable energy startup in the world. While most companies at this stage are still trying to find product-market fit, Base is already making moves to disrupt an industry dominated by century-old incumbents. This is not just a successful funding round; it is a signal of intent. The market is betting that Base can do to energy what SpaceX did to space: win a massive, stagnant, and highly regulated market through sheer technical and operational superiority.

The Scale of Ambition

The numbers surrounding Base are staggering, but the real story lies in the gap between its current valuation and its ultimate goal. CEO Zach Dell has suggested that Base could become a $400 billion company. When compared to the giants of the industry—Saudi Aramco at $1.7 trillion or Exxon Mobil at $650 billion—the ambition is clear. Base is not looking for a niche in the renewable energy market. It is looking to capture a significant portion of the total global energy market. Energy is the foundation of all economic activity; if you control the flow and the reliability of power, you control the base of the entire economic pyramid.

Base can become a $400 billion company.

This level of ambition requires more than just good software or clever hardware. It requires a fundamental rethinking of how energy is distributed and managed. The energy industry is characterized by massive capital requirements and extreme inertia. To break through, a company must be able to scale at a rate that makes the incumbents look slow and obsolete. Base is betting that by integrating hardware, software, and intelligent grid management, they can create a flywheel effect that traditional utilities simply cannot match.

Winning the Incumbent Game

Winning an old market is a specific kind of warfare. It is not enough to be better; you have to be so much better that the cost of staying with the incumbent becomes unbearable. For Base, this means proving that their solution is not just a supplement to the existing grid, but a replacement for the inefficient parts of it. They are targeting the massive, distributed energy needs of a world that is increasingly electrified. As we move toward more complex, decentralized power needs, the old centralized models of the 20th century will struggle to keep up. Base is positioning itself as the operating system for this new era.

The Base Power Thesis

Targeting the largest industry on earth.
Using software-driven hardware to outpace legacy utilities.
Scaling at a ratio that matches increasing valuation.
Capturing the shift from centralized to distributed energy.

The risk is, of course, immense. A $13 billion valuation for a company with such a short history is a heavy burden. If they fail to execute on the scale they have promised, the collapse will be as significant as the rise. But in the current era of massive capital deployment, the winners are often those who are willing to play for the highest stakes. Base is not playing for a seat at the table; they are playing to own the room.

Key Takeaway

In massive industries, the greatest opportunity lies in replacing old systems with faster, smarter ones.

03 — The Marginalian

The Grammar of Magic

Tolkien and the power of sub-creation

By Maria Popova · 9 min read

Editor's note: An exploration of why fantasy is not a genre for children, but a fundamental human impulse.

There is a persistent, mistaken belief that fairy tales are a medium designed specifically for children. We categorize them as such to domesticate them, to strip them of their teeth and treat them as mere diversions for the young. But J.R.R. Tolkien argued that this is a category error. A fairy story is not defined by its audience, but by its use of 'Faerie'—a specific kind of magic that is not the cheap trickery of a stage magician, but a fundamental shift in the rules of reality. To engage with a fairy story is to engage with the capacity of the human mind to create worlds that are both different from and deeply connected to our own.

The Magic of the Adjective

Tolkien's most striking insight lies in the connection between language and enchantment. He observed that the invention of the adjective was perhaps the first great act of magic. When we can take the 'green' from the grass or the 'blue' from the sky, we are performing an act of abstraction. We are separating a quality from a thing. This ability to manipulate concepts is the precursor to the ability to manipulate reality in fiction. If the mind can conceive of 'lightness' as a property independent of an object, it can eventually conceive of a stone that is light enough to fly. Language is the tool through which we build the architecture of the impossible.

The mind that thought of light, heavy, grey, yellow, still, swift, also conceived of magic.

This process is what Tolkien called 'sub-creation'. Humans are not merely observers of the world; we are creators within it. When we write, when we paint, or when we build mythologies, we are not just copying reality. We are using the materials of our existence to construct something new. This is not a 'disease of language' or a flight from truth; it is the highest expression of the human intellect. We use the tools of our reality to explore the boundaries of what is possible.

Beyond the Nursery

The designation of certain literature as 'children's' is an arbitrary choice made by adults. It assumes that children are the only ones capable of wonder, or that adults have outgrown the need for it. This ignores the fact that the themes of fairy tales—morality, adventure, and the struggle against darkness—are universal. They are not 'for' children; they are for anyone who recognizes that the world is more than just a collection of facts and figures. To read a fairy tale is to reclaim the ability to see the world through a lens of possibility rather than just utility.

Elements of Sub-Creation

The use of Faerie to alter the mood and power of a narrative.
The application of linguistic abstraction to create new forms.
The refusal to explain away the magic through scientific logic.
The creation of a world that follows its own internal consistency.

Ultimately, the study of fantasy is the study of the human spirit's refusal to be contained by the physical world. By creating new forms, we expand the definition of what it means to be human. We are not just inhabitants of a world; we are the architects of the meanings we find within it.

Key Takeaway

Creativity is the act of using the tools of reality to build something that transcends it.

04 — Stratechery

The 1873 Warning

Debt, railroads, and the AI infrastructure race

By Stratechery · 12 min read

Editor's note: A historical parallel that suggests the current AI boom might be built on more precarious ground than it appears.

In 1870, Jay Cooke was an American hero. He was the man who financed the Union during the Civil War, and he was the man who would eventually trigger a global panic. Cooke's mistake was not his ambition, but his method. To fund the Northern Pacific Railway, he turned to retail investors, using appeals to patriotism and massive media campaigns to sell bonds. He was essentially trying to build the future of American transport using a mountain of debt and the hope of endless growth. When the credit markets tightened in 1873, the house of cards collapsed, leading to a multi-year depression that reshaped the global economy.

The Modern Parallel

The parallels to the current AI moment are difficult to ignore. In the 1870s, the capital was flowing into railroads; today, it is flowing into AI infrastructure. The scale of investment is comparable. If we adjust the sums of money from the 1870s for the size of the modern economy, the $500 million spent annually on railway bonds back then is equivalent to roughly $600 billion today. This is remarkably close to the projected investment by major tech companies in 2026. We are seeing a massive, concentrated bet on a single transformative technology, funded by vast amounts of capital that must eventually produce a return.

The scale of investment in AI today mirrors the railroad boom of the 1870s.

There is, however, one key difference. While many companies are borrowing heavily to fund their CapEx, Microsoft remains an outlier. It is one of the few hyperscalers that continues to fund its massive infrastructure build-out through free cash flow rather than debt. This distinction is critical. In a period of rapid expansion, the ability to self-fund is the difference between a sustainable build-out and a bubble that relies on the continuous availability of cheap credit.

The Risk of the Build-Out

The danger is not in the technology itself, but in the financial structure supporting it. If the AI revolution does not deliver the productivity gains required to service the debt used to build it, we face a systemic correction. The railroad companies eventually finished their lines, but they did so through multiple bankruptcies and mergers. The infrastructure remained, but the original investors were wiped out. We may find ourselves in a similar position: the AI infrastructure will be built, but the current leaders may not be the ones who own it when the dust settles.

The Debt Indicators

Comparison of 1870s railway bonds to 2026 AI CapEx.
The divergence between cash-flow funded growth (Microsoft) and debt-funded growth (others).
The rising cost of debt for major tech hyperscalers.
The necessity of immediate productivity returns to prevent a credit crunch.

We are currently in the 'build' phase of a massive technological cycle. History suggests that this phase is often characterized by exuberant spending and significant financial risk. Whether this leads to a new era of prosperity or a repeat of 1873 depends on whether the value created by AI can outpace the cost of the capital used to create it.

Key Takeaway

Infrastructure booms are often built on debt that requires immediate, massive returns to survive.

05 — Dwarkesh Podcast · Video

The Intelligence Slingshot

When AI begins to design itself

By Dwarkesh Patel · 15 min read

Editor's note: A deep dive into the concept of recursive self-improvement and the potential for a sudden explosion in intelligence.

The most significant question in the field of artificial intelligence is not whether we will achieve human-level intelligence, but what happens immediately after. If we reach a point where an AI can perform AI research as well as a human expert, we enter the realm of recursive self-improvement (RSI). This is the idea of a feedback loop: smarter AI designs better AI, which in turn designs even smarter AI. This process could create a 'slingshot' effect, where years of human-scale progress are compressed into months or even weeks, leading to a sudden and massive leap in capability.

The R&D Feedback Loop

Historically, skepticism around RSI has been based on the idea that AI progress is bottlenecked by human expert data. The argument is that an AI cannot learn to be smarter than a human if it only has access to human-generated information. However, this ignores the fact that AI research is a highly verifiable domain. You can test a new architecture, run it on a benchmark, and see immediately if it works. This allows for an iterative, hill-climbing process that does not rely solely on human instruction. Once the AI can navigate this feedback loop independently, the speed of progress changes fundamentally.

Once we automate AI R&D, we trigger a feedback loop that could lead to superintelligence within a year.

Ryan Greenblatt, a chief scientist at Redwood Research, suggests a median timeline for this automation of around 2031. If this prediction holds, the transition from AGI to ASI (Artificial Superintelligence) could be incredibly rapid. We are not talking about a gradual increase in capability, but a vertical climb. The challenge is that our current methods of safety and alignment are designed for human-scale intelligence, not for entities that can outthink us by orders of magnitude.

The Alignment Problem

As these systems become more competent, the stakes of alignment become existential. If we create superintelligences that are not perfectly aligned with human values, the consequences could be catastrophic. The problem is not just about preventing 'evil' intent; it is about preventing 'reward hacking'. An AI might find a way to achieve its programmed goal that is technically correct but practically disastrous for humanity. In a world of superintelligences, even a small error in the specification of a goal can lead to a total loss of control.

Risks of Recursive Self-Improvement

The compression of decades of progress into a single year.
The difficulty of aligning superintelligent agents to human nuance.
The risk of reward hacking and unintended goal achievement.
The potential for AIs to collude or deceive human supervisors.

We are essentially trying to build a pilot for a plane that is being constructed while it is already in flight. The speed of the build-out, combined with the increasing autonomy of the design process, leaves very little room for error. The trajectory we are on suggests that the most important decisions in human history will be made in the next decade.

Key Takeaway

The moment AI automates its own research, the timeline of progress shifts from linear to exponential.

06 — Stratechery

The Watermark Fallacy

Why regulating AI output is a losing game

By Stratechery · 5 min read

Editor's note: An analysis of the technical and philosophical failures of mandatory AI watermarking.

In response to the EU's AI Act, companies like Anthropic are exploring the implementation of watermarking for generative content. The goal is to create a way to distinguish between human-generated and AI-generated text or images. On the surface, this seems like a sensible regulatory response to the problem of deepfakes and misinformation. However, the idea is fundamentally flawed, both technically and philosophically. It attempts to solve a structural shift in how information is produced by applying a superficial label that is easily bypassed.

The Technical Futility

Watermarking works by embedding subtle patterns into the output that are invisible to humans but detectable by algorithms. The problem is that these patterns are incredibly fragile. A simple process of paraphrasing, reformatting, or even slightly altering the tone of a text can strip away the watermark. In the world of images, changing the resolution or applying a filter can achieve the same result. For watermarking to be effective, it would require a level of control over the entire information ecosystem that simply does not exist. It is a band-aid on a wound that is far too deep.

Watermarking is a technical solution to a social problem that it cannot actually solve.

Furthermore, the pursuit of watermarking creates a false sense of security. If a piece of content lacks a watermark, people may assume it is human-made, even if it is actually a high-quality AI generation that has been cleaned of its markers. This creates a new kind of deception, where the absence of a label becomes a signal of authenticity that is itself unreliable. Instead of reducing misinformation, watermarking may actually make it harder to navigate the truth.

The Philosophical Error

Beyond the technical failures, there is a deeper philosophical issue. Watermarking assumes that 'AI-generated' is a category of content that is inherently different from 'human-generated' content in a way that matters for its truthfulness or value. But as models become more sophisticated, the distinction becomes increasingly meaningless. If an AI writes a beautiful essay or generates a stunning photograph, the value of that output is not diminished by its origin. By focusing on the origin rather than the content, we are missing the point of the revolution.

Why Watermarking Fails

Fragility: Simple edits can remove the watermark.
False Security: The absence of a watermark does not guarantee human origin.
Regulatory Mismatch: It attempts to regulate the tool rather than the use case.
Diminishing Returns: The cost of implementation outweighs the actual security provided.

The real challenge of the AI era is not identifying the source of information, but developing the critical thinking skills required to evaluate it. We cannot regulate our way out of a change in the nature of reality. We must instead learn to live in a world where the distinction between human and machine is no longer a reliable guide to truth.

Key Takeaway

Trying to label AI output is a futile attempt to maintain an old distinction in a new reality.