The Deep Feed

01 — Not Boring

The Texas Shipyard: Autonomy Meets Industrial Might

How a four-year-old startup moved from drone boats to a $3 billion shipyard gamble

By Packy McCormick · 12 min read

Editor's note: A look at how Saronic is bypassing political gridlock to rebuild American manufacturing through autonomous technology.

The speed of modern technological deployment is beginning to outpace the ability of traditional institutions to regulate or even comprehend it. Take Saronic. A mere four years ago, it was a startup with a concept. Within months, its Corsair autonomous drone boats were performing rescues of downed Apache crews. Now, those same boats have transitioned from search-and-rescue to active combat, striking an Iranian submarine and maintenance facility at Bandar Abbas. This is not a slow evolution of naval doctrine; it is a sudden, sharp break from the era of crewed, vulnerable vessels toward a future of autonomous, distributed lethality.

The Death of the Permitting Bottleneck

While the combat success makes headlines, the real story is the $3 billion greenfield shipyard Saronic is building in Brownsville, Texas. This move is a direct rebuke to the legislative paralysis seen in California. The 'California Forever' project, which aimed to bring similar industrial scale to the West Coast, stalled because local and state governments failed to pass a permitting framework in time. Saronic is not waiting for permission to build the future; it is moving to a jurisdiction that allows for scale. This shipyard, spanning 835 acres initially, aims to produce everything from icebreakers to container ships using the same advanced manufacturing stack that powers their drone boats.

Proving that a startup can build autonomous boats that are useful in combat is huge, but proving that they can build a great American shipyard again might be even bigger.

The economic implications are massive. The Texas project expects to create 10,000 direct jobs over the next decade. This is a vertical integration play that combines software-driven autonomy with heavy industrial capacity. It suggests that the next generation of defense and logistics giants won't just be software companies, but companies that own the entire stack—from the code that steers the ship to the steel that forms its hull. This is a return to hard manufacturing, but with a digital brain.

The Saronic Strategy

Rapid deployment from prototype to combat-ready status
Geographic arbitrage by moving to pro-growth jurisdictions
Vertical integration of autonomous software and heavy shipyard capacity
Scaling from small drone boats to 850-foot vessels

The success of this model depends on whether the regulatory environment can keep up with the physical reality. If Saronic succeeds, it provides a blueprint for how the US can rebuild its industrial capacity without waiting for the slow machinery of traditional bureaucracy to catch up. It is a high-stakes bet on the idea that agility is a more effective tool for national security than legacy infrastructure.

Key Takeaway

Industrial resurgence will be driven by companies that combine autonomous software with aggressive, large-scale physical manufacturing.

02 — The Marginalian

The Necessity of the Dark Period

Why despair is a biological requirement for growth

By Maria Popova · 10 min read

Editor's note: An exploration of the psychological and evolutionary purpose of periods of intense suffering and stagnation.

We often view periods of despair or disorientation as failures of the system—as bugs in the human experience that need to be fixed or avoided. However, a closer look at evolutionary biology and anthropology suggests that these 'dark periods' are actually essential features. Just as a forest requires a period of dormancy to replenish its soil, the human psyche often requires a period of collapse to reorganize itself around a more advanced principle. This is not a comforting thought, but it is a functional one.

The Psycho-Physical Crisis

Robert Ranulph Marett, a pioneer in the study of religion, argued that these crises are not merely mental; they are 'psycho-physical.' When we experience a collapse of spirit, it manifests in the body as a loss of tone and a sinking of energy. This is an evolutionary adaptation. The organism must lie dormant while its latent energies gather strength for activity on a new, higher plane. To fight the pause is to fight the very mechanism that allows for regeneration. The goal is not to escape the darkness, but to endure it with enough attention to see what it demands of us.

The organism needs to lie dormant whilst its latent energies are gathering strength for activity on a fresh plane.

This perspective refutes the modern obsession with constant productivity and perpetual positive affect. If we treat every moment of sadness or stagnation as a problem to be solved with a quick fix, we interrupt the very process that leads to genuine transformation. The 'sacred pause' is the time between the old system and the new. It is a period of chrysalis life, where the old form must be completely undone before the new one can emerge.

Characteristics of the Regenerative Pause

A sense of disorientation and loss of the old self
Physical manifestations of low energy and 'loss of tone'
The necessity of surrender over willpower
The eventual emergence into a higher state of functioning

Ultimately, the ability to navigate these periods requires humility. We must accept that we cannot will our way out of a systemic collapse. Instead, we must trust the process of the pause. Transformation is terrifying because it requires the death of the current version of ourselves. But without that death, there is no possibility of becoming something more capable or more aware.

Key Takeaway

Stagnation and despair are not signs of failure, but biological signals that a period of reorganization is underway.

03 — Stratechery

The End of the Mainframe Moat

How AI is eroding the foundations of legacy IT

By Stratechery · 15 min read

Editor's note: An analysis of why IBM's historic dominance is being threatened by the very technology it seeks to lead.

IBM is facing a crisis that goes far deeper than a bad quarterly earnings report. For over a century, the company has been anchored by the mainframe—the essential, heavy-duty hardware that runs the backend of the global economy. This was a moat built on extreme difficulty: the software running on these machines is archaic, complex, and incredibly expensive to move. For decades, customers stayed with IBM not necessarily because they loved the product, but because the cost and risk of leaving were prohibitive. This is the definition of a structural moat.

The AI Porting Threat

The threat to this moat is not just that companies are spending money on AI instead of hardware; it is that AI is changing the nature of technical debt. Traditionally, moving off a mainframe was a multi-decade, multi-billion dollar nightmare. However, generative AI has the potential to automate the porting of essential backend programs from archaic codebases to modern architectures. If an AI can translate COBOL or other legacy languages into modern equivalents with high accuracy, the 'stickiness' of the mainframe evaporates. The cost of migration drops, and the moat dries up.

The real concern for IBM is that AI’s ability to port the essential backend programs that run on archaic technology will mean those missed sales never come back.

This creates a paradox for legacy giants. To stay relevant, they must embrace AI, but embracing AI provides the very tools that their customers will use to escape them. It is a classic case of a company's survival strategy undermining its primary revenue driver. IBM is caught in a squeeze between the massive capital expenditure required to compete in the AI race and the erosion of the legacy business that funds that very competition.

The Erosion of Legacy Moats

High switching costs are being lowered by automated code translation
Capital allocation is shifting from maintenance to new AI R&D
Technical debt is becoming a manageable asset rather than a permanent liability

The lesson for any established player is that a moat based on complexity is no longer a permanent defense. In an era where intelligence can be commoditized and applied to any problem, the only lasting advantage is the ability to provide something that cannot be replicated by a model—be that unique data, specialized hardware, or a level of trust that transcends software.

Key Takeaway

When AI can automate the translation of legacy systems, the primary defense of established giants—complexity—becomes their greatest vulnerability.

04 — Simon Willison

The Browser in the Browser

WebAssembly and the collapse of the software stack

By Simon Willison · 6 min read

Editor's note: A technical look at how WebAssembly is enabling increasingly absurd and powerful layers of abstraction.

There is a certain absurdity in the current trajectory of web development that is hard to ignore. We are seeing the emergence of 'recursive software'—applications that run inside other applications, which in turn run inside a browser. A recent project by Puter has achieved this by compiling the Firefox browser to WebAssembly (WASM). The result is a Firefox instance running inside a Chrome browser. It is a feat of engineering that feels like a meta-commentary on the layers of abstraction we have built between ourselves and the hardware.

The Power of WebAssembly

This isn't just a parlor trick; it is a demonstration of what WebAssembly has made possible. WASM allows code to run at near-native speeds within the sandbox of a browser. By compiling the Gecko engine to WASM, developers can bypass the traditional limitations of web-based software. The project uses a WebSocket protocol to funnel traffic through a server, allowing the browser-within-a-browser to handle network requests that would otherwise be blocked by standard browser security models. It is a complex, expensive, and highly impressive workaround.

The whole browser runs in another browser.

The cost of such an achievement is significant. The developers reportedly used tens of thousands of dollars worth of AI tokens to compile the engine. This highlights a new reality in software development: the ability to build incredibly complex, highly abstracted systems is increasingly tied to the ability to leverage massive amounts of compute and AI-assisted engineering. We are moving away from the era of the lone programmer writing efficient C++ and into an era of architects managing massive, AI-generated abstraction layers.

Technical Requirements for Browser-in-Browser

WebAssembly (WASM) for near-native execution speeds
WebSocket/Wisp protocols for network proxying
Significant AI-assisted engineering for compilation
Strong single-process support in the host engine

While running a browser inside a browser is inefficient for daily use, it signals a profound shift. The browser is no longer just a window to the web; it is becoming a universal operating system capable of hosting entire environments. As these abstractions become more efficient, the distinction between a 'web app' and a 'native app' will continue to dissolve until it is entirely meaningless.

Key Takeaway

WebAssembly is turning the browser into a universal execution environment, enabling layers of abstraction that were previously impossible.

05 — Simon Willison

The Claude Fablepocalypse

How market competition forced Anthropic's hand

By Simon Willison · 5 min read

Editor's note: An analysis of how the competitive pressure from GPT-5.6 prevented Anthropic from gatekeeping its best model.

In the high-stakes race for AI dominance, even the most carefully laid business plans can be derailed by a single competitor's move. Anthropic recently faced a significant strategic pivot regarding its flagship model, Fable 5. The company's original plan was to remove the model from standard subscription tiers and make it available exclusively through higher-priced API usage. This was a move driven by compute scarcity—a desire to reserve their most powerful intelligence for those willing to pay the highest premium.

The Competitive Correction

The plan was sound in a vacuum, but it failed to account for the reality of the market. The emergence of GPT-5.6 Sol and other high-performing models made Anthropic's proposed restriction untenable. From a consumer perspective, there is no reason to pay a monthly subscription for a service that withholds its best capabilities. If Anthropic had proceeded, they would have effectively handed their most valuable users over to OpenAI. The market, in its most basic form, corrected Anthropic's attempt at artificial scarcity.

Why pay $100 or $200/month for a subscription plan that doesn't include Anthropic's best model?

The result is a compromise: Fable 5 will now be included in Max and Team Premium plans, albeit at 50% of the usual limits. This allows Anthropic to manage their compute costs while still providing enough value to keep their high-paying subscribers from defecting. It is a classic example of the tension between managing resource constraints and maintaining market competitiveness.

The Dynamics of AI Subscriptions

Compute capacity acts as a hard ceiling on model availability
Competitive parity with OpenAI dictates subscription value
Tiered access is used to balance cost and user retention

This episode highlights the volatility of the AI sector. Business models are being written and rewritten in real-time as models become more powerful and the cost of serving them fluctuates. For the user, it means the 'Fablepocalypse' has been averted, but it also serves as a reminder that in this industry, the value of your subscription is always at the mercy of the next model release.

Key Takeaway

In a hyper-competitive market, companies cannot successfully gatekeep their best products without risking immediate user defection to competitors.

06 — Simon Willison

The Death of the Query Plan

Demystifying the black box of database execution

By Simon Willison · 4 min read

Editor's note: A look at how new tools are making the internal workings of databases more accessible to humans.

For most developers, the internal workings of a database are a black box. You write a SQL query, you get a result, and if the query is slow, you feel a vague sense of frustration. Understanding *why* a query is inefficient requires diving into 'query plans' and low-level bytecode—technical documentation that is often dense, cryptic, and intimidating. This gap between writing a query and understanding its execution is a significant barrier to building truly efficient systems.

Bridging the Explanatory Gap

A new tool, the SQLite Query Explainer, is attempting to bridge this gap. By running queries against a SQLite database directly in the browser, the tool provides a plain-English annotation of both the `EXPLAIN QUERY PLAN` and the low-level `EXPLAIN` bytecode. It essentially translates the machine's logic into human reasoning, explaining exactly what the query planner and the virtual machine are doing at every step. It turns a cryptic diagnostic process into an educational one.

Maybe one day I’ll learn to read a query plan.

The tool is built using modern web technologies, running Python in Pyodide within WebAssembly. This allows for a high degree of interactivity without needing a heavy backend. While the creator warns that users should approach the explanations with caution, the value lies in the attempt to democratise database internals. It moves the needle from 'knowing that it works' to 'understanding how it works'.

Benefits of Query Explanation Tools

Reduces the cognitive load of reading bytecode
Provides immediate feedback on query efficiency
Acts as a pedagogical tool for junior developers
Demystifies the 'black box' of database engines

As data volumes grow and queries become more complex, the ability to audit and understand execution paths becomes critical. Tools like this represent a broader trend in software: the move toward transparency and observability. The goal is to ensure that as our systems become more complex, they do not become less understandable.

Key Takeaway

Demystifying the internal logic of databases through plain-English explanations is a vital step toward more efficient and observable software engineering.