The Deep Feed

01 — Not Boring

The Blood-to-Egg Pipeline

How stem cell reprogramming is rewriting the rules of reproduction

By Packy McCormick · 12 min read

Editor's note: A look at how biology is becoming a programmable medium.

For most of human history, reproduction has been a biological lottery governed by strict physical constraints. You either had the eggs, or you didn't. You had the biological machinery, or you were sidelined by age, illness, or circumstance. That reality is currently being dismantled in a Berkeley lab. A startup called Conception has successfully grown human egg cells, known as primary oocytes, from stem cells derived from a simple blood draw. This isn't just a medical curiosity; it is the beginning of a shift where human fertility becomes a manufactured resource rather than a finite biological gift.

The Mechanics of Reprogramming

The process is a feat of cellular engineering. Scientists take blood cells and reprogram them into induced pluripotent stem cells. These cells are then coaxed into forming miniature human ovaries, which in turn grow the eggs inside them. The eggs undergo meiosis, the essential division process that ensures genetic diversity, and are wrapped in follicles just as they would be in a natural ovary. While this technology is still in its early stages and clinical use is years away, the proof of concept is undeniable. We are moving toward a world where a single drop of blood could theoretically provide an inexhaustible supply of healthy eggs.

We are entering an era where we can make as many healthy eggs as we need from a single drop of blood.

The implications for human society are massive and messy. This technology could allow two biological fathers to have a child together. It could restore fertility to cancer patients whose reproductive systems were destroyed by treatment. It could even allow for the de-extinction of various species. However, the ethical debates will be fierce. When we decouple reproduction from the natural aging process and the traditional female body, we change the fundamental structure of family and lineage. We are moving from being subjects of biology to being its architects.

Potential societal shifts:

Biological parity for same-sex male couples
Restoration of fertility in cancer survivors
The ability to extend the reproductive window indefinitely
Accelerated species conservation through de-extinction

This is not merely about fixing a problem; it is about expanding the range of human experience. If we can control the building blocks of life with this level of precision, the concept of 'natural limits' becomes an obsolete term. The challenge will not be the science itself, but the social and legal frameworks we build to contain it. We are learning to write the code of life, and we are doing so with a very high-stakes compiler.

Key Takeaway

Fertility is transitioning from a biological certainty to a programmable technology.

02 — Not Boring

The 250-Year Reset

Reflecting on the American trajectory from colonies to digital superpower

By Packy McCormick · 10 min read

Editor's note: A perspective on how quickly the 'impossible' becomes 'trivial'.

In 1776, the concept of a modern American life was non-existent. There was no electricity, no running water, and no way to communicate across distances faster than a horse could gallop. If you were sick, you faced infection without the benefit of penicillin or even basic hygiene knowledge. The world was a place of wood and muscle, where humans were primarily takers of what nature provided. To the people signing the Declaration of Independence, the world was local, slow, and incredibly precarious.

The Velocity of Change

We often mistake our current technological state for a permanent baseline. We forget that the leap from the quill pen to the smartphone is not just a series of upgrades, but a complete reconfiguration of the human experience. In 1776, a person's image or voice could not be preserved; once they were gone, they existed only in memory. Today, we inhabit a shared, digital present where information moves at the speed of light. This shift from a world of physical constraints to a world of digital abundance is the defining story of the last two and a half centuries.

Two hundred and fifty years seems like a long time, but it is just three lifetimes at current averages.

The engine of this transformation has been a specific combination of capitalism, geography, and a frontier spirit. It is the constant recombination of these elements that turned a collection of colonies into a global technological leader. But as we look toward the next 250 years, the question is whether we can maintain that momentum. The tools are changing—moving from steam and steel to silicon and atoms—but the underlying requirement for innovation remains the same: the ability to master new inputs.

What 1776 lacked:

Standardised time zones and real-time coordination
Synthetic fertilizers and reliable food storage
Medical interventions like anesthesia and antibiotics
The ability to preserve human likeness through media

As we approach the 250th anniversary of the United States, we are at another inflection point. The next era will likely be defined by our mastery of energy—specifically nuclear—and our ability to integrate artificial intelligence into the fabric of our economy. The transition from being input-takers to input-makers is continuing, moving from the mastery of physical materials to the mastery of information and energy at the atomic level.

Key Takeaway

Technological progress is not a steady climb, but a series of radical reconfigurations of reality.

03 — Simon Willison

The Rise of the Coding Agent

Testing the limits of autonomous software development

By Simon Willison · 8 min read

Editor's note: When the AI stops being a chatbot and starts being a colleague.

The era of the LLM as a mere text generator is ending. We are entering the era of the agent—software that doesn't just talk about code, but actually executes it, tests it, and fixes its own mistakes. Recent experiments with frameworks like Fable 5 show that a simple coding agent can now take a high-level specification and turn it into a functioning Python library. This isn't just about speed; it's about a shift in the developer's role from writer to supervisor.

From Prompting to Orchestrating

A modern coding agent operates through a suite of tools: reading files, editing specific strings, executing shell commands, and running test suites. In a recent test, a developer provided a minimal prompt to build a coding agent using a 'red/green' TDD approach. The agent didn't just write the code; it managed the commits, updated the documentation, and verified its own work through passing tests. It even demonstrated reasoning capabilities, such as identifying when a specific framework was unsuitable for a given task and suggesting an alternative.

The developer's job is shifting from writing lines of code to managing a loop of intent and verification.

This capability introduces a new kind of workflow. Instead of manually debugging a failing test, a developer can simply point an agent at the error and tell it to 'fix the failing test in tests/test_parser.py'. The agent then enters a loop of observation, hypothesis, and execution. While this requires a high degree of oversight to prevent 'hallucinated' bugs or security risks, the efficiency gains are massive. We are seeing the birth of a 'YOLO' mode for development, where the speed of iteration outpaces human manual input.

Core Agent Capabilities:

File manipulation (edit, read, write, search)
Shell command execution with timeout controls
Automated TDD (Test-Driven Development) loops
Reasoning-based tool selection

The real challenge for the next generation of engineers won't be syntax, but architecture and intent. If an agent can handle the implementation details, the human must become better at defining the boundaries, the security constraints, and the high-level logic. The bottleneck is no longer the typing; it is the thinking.

Key Takeaway

Coding is moving from a manual craft to a supervisory discipline.

04 — Simon Willison

The Creator's Dilemma

How AI is cannibalising the education and expertise economy

By Josh W. Comeau · 6 min read

Editor's note: The economic reality for high-end content creators in an AI world.

For years, the business model for expert creators was simple: build a deep, specialized knowledge base and sell access to it through structured courses. If you wanted to master complex animations or advanced React, you paid for the curriculum and the community. That model is currently under siege. High-end educators are reporting revenue drops of 50% or more, as the economic foundations of digital expertise begin to shift beneath them.

The Double Whammy

The pressure comes from two directions. First, there is a psychological shift in the market. As AI tools become more capable, potential students are questioning the long-term value of learning certain skills. If an LLM can write the code or design the layout, the perceived ROI of a six-month course diminishes. Why invest time and money in a skill that might be automated by the time you finish the curriculum?

LLMs can provide personalised tutoring, leaving less incentive to buy a paid course.

Second, there is the direct issue of content cannibalisation. LLMs are trained on the very work that these creators produce. These models 'slurp up' high-quality, specialized content and regurgitate it in response to user queries—often without consent or compensation. This turns the creator's own intellectual property into a competitor that offers a free, albeit sometimes less accurate, alternative to the original work.

Factors driving the decline:

Reduced perceived value of manual skill acquisition
AI-driven personalised tutoring replacing structured courses
Direct competition from models trained on creator content
Decreased engagement with long-form educational content

This is not just a problem for web developers; it is a problem for anyone whose value lies in the synthesis and explanation of complex information. The 'expertise economy' is being forced to reinvent itself. Creators must find ways to offer value that an LLM cannot: deep community, real-time feedback, or perhaps a level of creative nuance that current models still struggle to replicate.

Key Takeaway

AI is not just a tool for creators; it is a competitor that uses their own work against them.

05 — Simon Willison

Mapping the Open Source AI Gap

The struggle to build a public option for intelligence

By Simon Willison · 5 min read

Editor's note: Tracking the fragmented ecosystem of non-proprietary AI.

As proprietary AI models become more dominant, the importance of an open-source alternative grows. The 'Open Source AI Gap Map' is a new attempt to index this fragmented landscape. It is an effort to track the 'public option' for AI—the models, datasets, and infrastructure that are not controlled by a handful of trillion-dollar corporations. Without this map, the ecosystem remains a black box of disconnected projects.

The Scale of the Ecosystem

The map currently identifies hundreds of specific products across 14 categories, including model components, UX layers, and infrastructure. However, these numbers represent only a tiny fraction of the reality. The project estimates there are over 24,000 other artifacts in the open-source AI space that remain uncatalogued. This 'long tail' of development is where much of the real innovation happens, yet it is the hardest to track and quantify.

Current AI is a global partnership building a public option for AI.

The project is backed by significant capital—$400 million already committed—signalling that the push for open AI is no longer just a hobbyist movement. It is a strategic necessity. If the core intelligence of the future is entirely proprietary, the power imbalance between those who own the models and those who use them will be insurmountable. The Gap Map serves as a vital diagnostic tool for seeing where the public option is strong and where it is dangerously thin.

Categories tracked by the map:

Model components
Product and UX layers
Infrastructure and hardware projects
Datasets and training tools

The data itself is being released under an MIT license, allowing researchers to explore the gaps using tools like Datasette. This transparency is essential. To build a robust public alternative, we need to know exactly what has been built, what is missing, and where the dependencies lie. The map is the first step in turning a chaotic collection of repos into a coherent ecosystem.

Key Takeaway

The survival of open intelligence depends on our ability to organise and fund the fragmented open-source ecosystem.

06 — Not Boring

The Synthetic Cell

When life is assembled from non-living chemicals

By Packy McCormick · 9 min read

Editor's note: A look at the moment biology becomes chemistry.

For as long as humans have studied biology, we have assumed that life must come from life. Cells divide, organisms reproduce, and the continuity of existence is a chain of living predecessors. That assumption was challenged this week by researchers at the University of Minnesota. They have created 'SpudCell'—the first synthetic cell with a complete life cycle assembled entirely from non-living chemicals. This is a fundamental shift in our understanding of the boundary between matter and life.

Breaking the Biological Chain

SpudCell is not just a collection of chemicals; it is a functional system. It can feed, it can grow, it can reproduce, and it can compete with other cells for resources. By assembling these functions from scratch, scientists have demonstrated that the core processes of life are not mystical properties, but chemical arrangements that can be engineered. This moves biology closer to the realm of traditional engineering, where we build machines from components rather than growing them from seeds.

The core functions of life can be achieved through the assembly of non-living matter.

The implications for biotechnology are vast. If we can design cells from the ground up, we are no longer limited by the evolutionary designs of existing organisms. We can create specialized biological machines for drug delivery, carbon sequestration, or material production that are optimized for specific tasks. We are moving from a period of 'discovery' in biology to a period of 'design'.

Capabilities of SpudCell:

Metabolic processing (feeding)
Autonomous growth
Reproduction via division
Resource competition

As we master the ability to create synthetic life, we face a new set of existential questions. What does it mean to be 'alive' if the origin is a lab bench rather than a lineage? How do we regulate entities that do not fit into our current biological taxonomies? The creation of SpudCell is a milestone that marks the beginning of a new era in which the distinction between the organic and the synthetic becomes increasingly irrelevant.

Key Takeaway

Life is becoming a design specification rather than a biological inheritance.