The Deep Feed

01 — Not Boring

The Blood-to-Egg Pipeline

How stem cell reprogramming is turning the biological lottery into an engineering problem

By Packy McCormick · 12 min read

Editor's note: We are entering an era where biological limitations may become optional through sheer technical persistence.

For most of human history, reproduction has been a matter of chance—a biological lottery governed by the availability of eggs, the health of the womb, and the ticking clock of age. We have lived within the strictures of our anatomy, accepting the reality that fertility is a finite resource. That reality is currently being dismantled in a laboratory in Berkeley. A startup called Conception has successfully grown early human egg cells, known as primary oocytes, from stem cells. The process is as clinical as it is radical: they take a sample of blood, reprogram those cells into induced pluripotent stem cells, and then coax them into growing miniature human ovaries. Inside these lab-grown structures, the cells undergo meiosis, the specific type of cell division that produces gametes. This is not a theoretical exercise; it is a direct challenge to the fundamental constraints of our species.

The End of Biological Scarcity

The implications of this technology extend far beyond simple fertility treatments. If we can manufacture eggs from a single drop of blood, the concept of 'running out' of reproductive potential disappears. This capability offers a path for women to extend their fertile years indefinitely or for cancer patients to restore what chemotherapy may have destroyed. It also opens doors that were previously locked by the binary nature of biology. We are looking at a future where two biological fathers could conceive a child, or where the efficacy of embryo selection is vastly improved because the pool of available eggs is no longer limited by the monthly cycle or the natural aging process. We are moving from a model of scarcity to a model of production.

We are moving from a model of scarcity to a model of production.

The technical difficulty of this feat cannot be overstated. While in vitro gametogenesis has been possible in mice since 2016, the human version has proven stubbornly resistant to replication. The transition from mouse to human requires a level of precision in cellular signalling that we are only just beginning to master. It took years of intense research to ensure that these lab-grown oocytes could actually complete the complex dance of meiosis and develop the necessary follicle structures. This is not merely about making cells; it is about recreating the intricate biological environment required for life to begin its most fundamental processes.

Potential applications of egg-cell reprogramming:

Restoring fertility to cancer survivors
Enabling biological parenthood for same-sex male couples
Expanding the window of reproductive age
De-extinction of endangered species

Of course, this technical triumph brings a heavy set of ethical questions that society is unprepared to answer. When the barrier to reproduction becomes a matter of manufacturing rather than biology, the definition of family and the ethics of genetic selection will undergo massive pressure. How do we regulate the use of these cells? Who gets access to this technology—will it become a tool for the wealthy to bypass the natural limits of aging? These are not just scientific questions; they are questions about the structure of human society and the value we place on the 'natural' order. The technology is arriving faster than our moral consensus.

As we watch these cells divide in petri dishes, we are witnessing the first steps of a broader transition. We are learning to treat biology as a programmable medium. The ability to turn blood into eggs is a signal that the distinction between 'born' and 'made' is beginning to blur. For those who view this as a way to increase the range of human experience, it is a victory. For those who fear the loss of biological autonomy, it is a warning. Either way, the era of the biological lottery is coming to a close.

Key Takeaway

Reproduction is shifting from a biological certainty to an engineering capability.

02 — Simon Willison

The Tutor in the Machine

How generative AI is eroding the value of human expertise and the creator economy

By Josh W. Comeau · 9 min read

Editor's note: The economic foundation of specialized knowledge is being rewritten by the availability of instant, free instruction.

For years, the path to expertise was paved with paid instruction. If you wanted to master a complex skill—coding, design, or a new language—you bought a course. You invested in the curated wisdom of someone who had already walked the path. But that economy is currently in freefall. Josh W. Comeau, a prominent educator, recently noted that his latest course launch is tracking at only a third of his usual sales. This isn't a seasonal dip or a marketing failure; it is a structural shift. The primary competitor to the human teacher is no longer other teachers, but the large language model sitting in every student's pocket.

The Personalized Tutor Problem

The threat posed by AI to education is two-fold. First, there is the matter of utility. An LLM can provide personalised, instant tutoring that a pre-recorded video course cannot match. If a student gets stuck on a specific line of code, they don't have to wait for a forum response or re-watch a ten-minute video; they can ask the AI to explain that exact error in the context of their specific project. The AI becomes a tireless, infinitely patient tutor that adapts to the learner's pace. Second, there is the psychological impact on the learner. When people see AI performing tasks that used to require years of training, they begin to question the necessity of learning those skills in the first place. Why spend six months mastering a framework if a machine can generate it in six seconds?

The primary competitor to the human teacher is no longer other teachers, but the large language model sitting in every student's pocket.

This creates a devastating cycle for the creator economy. As the perceived value of learning a skill drops, the revenue for those who teach it follows. Educators are seeing revenue drops of 50% or more. But the problem goes deeper than just lost sales. There is a fundamental issue of consent and compensation. These models were trained by 'slurping up' the work of human creators—the very people whose livelihoods are now being undermined. The AI is essentially using the expertise of the masters to teach the apprentices how to replace the masters.

Why AI disrupts the learning economy:

Instant, personalised feedback loops
Reduced perceived necessity of deep skill acquisition
Zero-marginal cost of instruction
Uncompensated use of creator data for training

We are witnessing the commodification of expertise. When knowledge is no longer something you possess, but something you access via a prompt, the status of the 'expert' changes. The expert is no longer the person who knows the answer, but the person who knows how to ask the right question. While this might sound like a new kind of empowerment, it ignores the reality that deep, intuitive understanding often comes from the very struggle of learning that AI now bypasses. We risk creating a generation of users who can operate tools without understanding the principles that make those tools work.

The challenge for the next decade will be finding a way to value human insight in an age of automated intelligence. If the economic model for creators collapses, we may find ourselves in a world of high-speed output but low-depth understanding. We are trading the slow, difficult process of mastery for the instant gratification of implementation. The cost of that trade is yet to be fully calculated.

Key Takeaway

AI is turning specialized knowledge from a valuable asset into a cheap, accessible commodity.

03 — Simon Willison

Mapping the AI Commons

The struggle to build a public option in the age of proprietary intelligence

By Simon Willison · 6 min read

Editor's note: The battle for the future of intelligence will be won or lost in the open-source ecosystem.

The current state of artificial intelligence is a concentrated monopoly. A handful of massive corporations hold the keys to the most powerful models, creating a dependency that is both economic and strategic. However, a counter-movement is gaining momentum. A non-profit initiative, backed by $400 million in committed capital, is working to build a 'public option' for AI. This isn't just about releasing code; it is about building a complete, interoperable ecosystem that can compete with the proprietary giants. To understand where this fight stands, we must look at the 'Gap Map'.

The Architecture of the Gap

The Gap Map is an attempt to index the current state of open-source AI, revealing the strengths and weaknesses of the public ecosystem. It categorises hundreds of products across three layers: model components, user experience, and infrastructure. The map shows that while there is a massive 'long tail' of thousands of smaller projects, the core pillars—the high-performing models and the robust infrastructure required to run them—are still being built. The map is a diagnostic tool for the commons, showing exactly where the open-source movement is winning and where it is being outpaced by private capital.

The battle for the future of intelligence will be won or lost in the open-source ecosystem.

The existence of such a map highlights a critical tension. On one side, you have the proprietary models, which are polished, integrated, and easy to use, but controlled by entities with singular interests. On the other, you have the open-source ecosystem, which is fragmented, difficult to navigate, but fundamentally democratic. The goal of the public option is to bridge this gap, providing tools that are as capable as the commercial versions but are owned by no one and accessible to everyone.

The three layers of the AI stack:

Model Components (the raw intelligence)
Product / UX (how humans interact with it)
Infrastructure (the hardware and software that runs it)

This is not just a technical challenge; it is a resource challenge. Building a public option requires massive amounts of compute, high-quality data, and talent. The $400 million commitment is a significant start, but it is a drop in the ocean compared to the hundreds of billions being spent by the tech giants. The open-source movement must be more than just a collection of hobbyists; it must become a coordinated, professionalised force capable of managing the scale of modern intelligence.

If the public option fails, we face a future where the very engine of thought and productivity is a black box, controlled by a few. If it succeeds, we ensure that the benefits of AI are distributed across the globe, rather than being captured by a handful of balance sheets. The Gap Map is the first step in making sure we know exactly what we are fighting for.

Key Takeaway

Open-source AI is the only way to prevent a total monopoly on intelligence.

04 — Simon Willison

The Managerial Model

Delegating intelligence and the new economics of token efficiency

By Simon Willison · 5 min read

Editor's note: As AI becomes more capable, our role shifts from being the doer to being the manager of autonomous agents.

We are moving past the era of 'prompt engineering' and into the era of 'agentic management'. In the early days of LLMs, the goal was to learn how to talk to the machine to get the desired output. Now, the goal is to learn how to delegate tasks to a hierarchy of models. This shift is driven by a simple economic reality: intelligence has a cost. High-tier models are powerful but expensive and slow. Lower-tier models are cheap and fast but lack the reasoning capabilities required for complex synthesis. The most efficient way to work is to use a high-tier model as a manager that decides which sub-agents to deploy for specific tasks.

The Hierarchy of Thought

This new workflow relies on a concept of 'judgement'. Instead of telling an AI exactly how to perform every step of a coding task, you tell it to use its own judgement to decide which model is appropriate for the job. For a substantive implementation that requires deep reasoning, the agent might call upon a top-tier model. For a trivial edit or a mechanical task, it might spawn a sub-agent running a much lower-power, cheaper model. This creates a tiered system of intelligence where the main loop handles the high-level strategy, and the sub-agents handle the grunt work.

We are moving from being the doer to being the manager of autonomous agents.

This approach changes the nature of technical work. The human's role is no longer to write the code, but to audit the results and manage the delegation process. You become a supervisor of a digital workforce. The skill is no longer in the execution, but in the design of the system and the review of the output. This requires a different kind of mental model—one that focuses on architecture, oversight, and the ability to spot errors in a high-speed stream of automated production.

The components of an agentic workflow:

The Main Loop (high-level reasoning and synthesis)
Sub-agents (specialised, lower-power models)
Model Overrides (choosing the right tool for the task)
Review and Commit (the human oversight layer)

The economic incentive for this is clear. As the cost of high-end tokens remains high, the ability to 'offload' work to cheaper models becomes a competitive advantage. Those who can build systems that intelligently navigate this hierarchy will be able to accomplish far more than those who rely on a single, expensive model for everything. It is a shift from brute-force intelligence to efficient, managed intelligence.

However, this delegation comes with risks. If the manager model makes a mistake in its judgement, the entire chain of sub-agents will follow that error. We are building systems that are increasingly complex and opaque. The more we delegate, the more we rely on the machine's ability to self-correct. We are essentially betting that the management layer is robust enough to handle the errors of the workers.

Key Takeaway

The future of work is not about prompting, but about managing a hierarchy of intelligent agents.

05 — The Marginalian

The Boat and the Sea

Hemingway, grief, and the hard-won victory of a life lived well

By Maria Popova · 10 min read

Editor's note: A meditation on the brutal honesty required to face loss and find meaning in a world that offers no guarantees.

There is a specific kind of loss that defies the standard vocabulary of grief. It is not the loss of a possession or even the gradual fading of a relationship; it is the sudden, violent interruption of a life that was still in its prime. When Ernest Hemingway wrote to his friends, the Murphys, after the death of their young son, he did not offer the usual platitudes of comfort. He did not say that everything happens for a reason or that the child was in a better place. Instead, he offered something much harder and much more honest: the idea that a life ended in happiness is a victory.

The Victory of the Young

Hemingway’s argument is startling in its clarity. He suggests that while we all face death by defeat—our bodies failing, our worlds crumbling—the young who die after a happy childhood have achieved something different. They have 'gotten it all over with' while their world was still intact. They have not yet had to learn the bitterness of the world or the slow decay of the spirit. In his view, the tragedy is not that they died, but that they were spared the struggle of living long enough to be defeated. It is a perspective that reframes the most devastating of losses as a strange, almost defiant triumph.

Anyone who dies young after a happy childhood... has won a great victory.

This is not a denial of pain, but a way of locating meaning within it. Hemingway acknowledges his own inability to be 'brave' in the face of such loss, admitting that he is 'sick' for his friends. But he separates the emotional agony from the philosophical reality. He is making a distinction between the feeling of grief and the truth of a life's value. It is a refusal to use easy language to cover up the enormity of the situation.

Core themes in Hemingway's letter:

The distinction between emotional pain and philosophical truth
The idea of death as an accident versus death as a slow defeat
The value of a happy childhood as a shield against the world

He concludes with a metaphor that has become a cornerstone of his philosophy: the image of a group of people on a boat. We are all on a boat together, a vessel we have built but which will never reach a permanent port. The weather will be good and the weather will be bad. There is no landfall. Because there is no final destination, the only thing that matters is how we treat each other on the journey. The purpose of life is not to reach a port, but to keep the boat up and be good to the people on board.

This is a philosophy of the present tense. It is a rejection of the idea that life is a means to an end. If there is no port, then the journey itself is the only reality we have. It is a heavy, demanding way to live, requiring us to find meaning in the maintenance of the boat and the quality of our connections, even as we know the sea is indifferent to our survival. It is a way of living that accepts the inevitability of loss while refusing to let that loss strip the journey of its value.

Key Takeaway

Meaning is found not in the destination, but in the quality of our presence on the journey.