Tag: Future of Work

Categories
AI Startups

A New Reason to Launch

“Before you launch, the speed you can build is now mainly limited by your imagination in what you tell AI. After you launch, the AI can watch your users and make improvements on its own.”
— Jared Friedman, Y Combinator

Jared Friedman watches hundreds of founders a year navigate the gap between idea and launched product. He notices patterns the rest of us miss. And what he’s describing above is not an incremental improvement in how software gets built. It is a change in the nature of the advantage.

This is a different kind of liberation than founders have known before.

The old liberation was launch early and the market corrects your wrong assumptions. Humbling, but useful. You were still the one doing the correcting, late at night, rewriting the onboarding flow based on what the data told you.

The new liberation he’s describing is something closer to multiplication. You launch, and now there are effectively more of you. The AI is watching session replays you’ll never have time to watch. It’s noticing the drop-off after step three that you’d have caught in month four. It’s holding the pattern of a thousand user paths simultaneously and asking what they mean. Your imagination seeded the thing. Reality is now feeding it.

That observation redraws the map cleanly. Pre-launch and post-launch used to differ in degree — you knew more after than before. Now they differ in kind. Pre-launch you are the sensing organ. Post-launch you’ve grown new ones.

The founders who feel this most viscerally, I suspect, are the ones building alone or in pairs — the people for whom every previous era of building had a hard ceiling imposed by human hours. They could only read so many support tickets. They could only run so many experiments. The ceiling is lifting and the feeling is of a room getting larger.

The core advice hasn’t changed. Paul Graham was saying “launch early” twenty years ago and it was true then. What’s changed is the reason underneath it — the mechanism that makes it true now is nothing like the one he had in mind.

The advice is twenty years old. There is a new reason and it is brand new. Most people haven’t noticed the swap yet. But they will.

That window does not stay open long.

Categories
AI Stanford

The Unit of Production Just Collapsed

The lecture was a Stanford CS session, AI-native companies, Garry Tan walking through what it now takes to build something. He’d rebuilt his old startup, Posterous, in five days on a modest Claude plan. A thing that once required a team and a runway. He said it matter-of-factly, the way you describe something that’s already obvious to you and hasn’t yet reached everyone else.

The argument Tan and his colleague Diana Hu were making wasn’t really about AI. It was about the economics of effort — specifically, what breaks when the cost of turning an idea into a working thing falls by an order of magnitude.

Their framing: AI-native organizations running as closed-loop systems, agents with access to the real artifacts of work, able to iterate without the error-accumulation that comes from handoffs and headcount. Revenue-per-employee ratios of a million dollars or more, with live examples already in the YC portfolio. Document processing, logistics, voice agents for specialized workflows.

What I kept hearing underneath all of it was a quieter claim: the mental model of what a startup requires is wrong.

Or rather, it’s right about the past and increasingly wrong about the present.

The assumptions embedded in “I can’t do this alone” or “we’d need to hire for that” or “we don’t have the bandwidth” — those are load-bearing assumptions, and the load is shifting.

I have some small version of this — not as a founder, but as someone who retired into curiosity. The blog, the reading, the daily effort to keep up with what’s moving: each one is a practice in staying oriented while the map keeps changing.

What I notice is that the constraint has shifted. It’s not information anymore. It’s not even tools. It’s the capacity to ask better questions of the abundance, to know what matters when everything is accelerating.

That’s the thing I find unsettling, yet also genuinely interesting: the skills that remain irreplaceable are the hardest ones to teach, and the hardest to evaluate in yourself. Knowing what matters. Recognizing when an output is almost right and almost wrong. Setting direction in ambiguous conditions and being willing to be wrong about it. These were always the valuable things. They were just obscured by all the coordination overhead that surrounded them.

The students in that Stanford course were asked to build something called a One-Person Frontier Lab — use the best available tools to extend your own reach over ten weeks. It’s framed as an academic exercise. It doesn’t feel like one.

But I’m not building. I’m mostly watching, and thinking about what this radical new fermentation does to everything downstream — to labor markets, to what a company even is, to how we’ll organize work and meaning when the old unit of production no longer applies. Those are slower questions. But they’re the ones that feel urgent to me.

The old excuses are getting lighter. Not that everything is possible — but that the weight of the usual constraints has changed.

What you choose to build, and whether you choose to build it at all, is more purely a decision than it used to be. That’s either clarifying or terrifying, depending on the day and my mood.

Categories
AI California San Francisco/California

Distant Billboards

Greg Isenberg came back from San Francisco with seventeen observations. The billboards advertising either B2B inference infrastructure or vertical agent companies, the seed rounds, the forward-deployed engineers, the founders showing each other their Obsidian vaults like athletes comparing gym routines.

He noted an important thing in observation fifteen, almost as an aside.

Walking around the Mission I noticed something: the street-level businesses, the taquerias, the barbershops, the laundromats — none of them use any AI at all.

Everett Rogers formalized the technology diffusion model in 1962. He was studying hybrid seed corn in Iowa. He noticed that the farmers who adopted early weren’t just better informed — they had different social networks, different relationships to risk, different orientations toward outside knowledge. The late adopters weren’t slower. They were operating from a different set of facts about what was safe to try.

Those AI billboards in SoMa are not visible in the Mission. That’s not metaphor. That’s just geography.

What strikes me about the taqueria is not that it’s behind. It’s that the conversation happening a mile away — about MCP endpoints and agent fleets — is not legible to it. The vocabulary doesn’t exist there yet. Nobody has sat across from the woman making carnitas for twenty years and said: here is what this could do for your ordering, your scheduling, your response to a customer who asks on Yelp at 11pm whether you’re open on Monday. One day her daughter or son might.

The builder class optimizes for the builder class. You build what you understand, for people whose problems you can see. The founders in SoMa understand each other’s problems with extraordinary precision.

The woman making carnitas has different problems — thinner margins, less access to capital, relationships built over decades that don’t easily transfer to a new system. Nobody is at the Series A meeting making the case that her problems are the interesting ones.

The historian of technology David Nye wrote about the “technological sublime” — the awe Americans felt in the nineteenth century standing before a great bridge or a locomotive or the first electrified city. The feeling was real. But the sublime is a view from a particular angle. The workers who built the bridge experienced something quite different. The families displaced by the railroad’s right-of-way experienced something different still.

The question isn’t whether the technology will eventually reach her. It will. The diffusion curve is patient. It likely will surprise.

The question is whether anyone is doing the translation work. The act of standing in a specific kind of life and asking: what would this actually change here? In the actual kitchen, on the actual Tuesday.

Isenberg noted that the coworking spaces in SF are half empty but the coffee shops are packed. People want to be around people.

The taqueria is also a place where people want to be around people. It has been that for a long time.

She’ll adapt. She’s been adapting for twenty years.

But that’s a very different story than the one being told in San Francisco on those billboards.

Categories
AI Business Consulting

The Toll Bridge and the Terrain

For fifteen years of my life, I lived inside the fortress of information asymmetry. I was part of a payments consulting business, and our model was exactly what Andrew Feldman described on a recent Moonshots episode when he pointed a sharp finger at traditional professional services.

His observation was simple, cutting, and entirely true:

“Their role today is to stand between ordinary people and obscure knowledge. And the application of that obscure knowledge to everyday problems.”

When I heard him say that, it landed with a quiet thud of recognition.

For a decade and a half, my colleagues and I were the ones standing in that gap. The payments industry—with its labyrinth of interchange fees, compliance structures, clearing networks, and legacy tech stacks—is a monument to obscure knowledge. Clients didn’t come to us because we possessed some divine, unreplicable wisdom. They came to us because the map was locked in our heads, and navigating the terrain without us was a recipe for an expensive disaster.

We charged for our time, and we earned it. We untangled complexity and solved real, everyday business problems for people who just wanted to move money safely from point A to point B.

But looking back now, I can see the architectural flaw disguised as a premium service. The economic foundation of that entire era relied on friction. It relied on the fact that it took an immense amount of human energy to retrieve a piece of obscure data and map it onto a specific business dilemma. You weren’t just paying for strategic guidance; you were paying a premium on artificial scarcity.

We are living through a moment where the marginal cost of intelligence is rapidly trending toward zero. When the barrier of “obscure knowledge” evaporates, the traditional toll bridges begin to look absurd.

For anyone starting a consulting business today, the playbook would have to be entirely different. When an LLM can parse thousands of pages of network operating rules, interchange tables, and regulatory compliance frameworks in a handful of seconds, the gatekeeper’s standing ground liquefies.

If your value proposition is merely standing between a client and a hidden database, your business model isn’t just flawed—it’s obsolete.

Yet, this collapses into a fascinating paradox. You might assume that when you democratize expertise, you eliminate the need for the expert. But as Dan Shipper recently observed, the reality of AI is completely counterintuitive.

Shipper points out that AI effectively packages up “yesterday’s competence” and makes it cheap and ubiquitous.

Suddenly, anyone can generate a complex contract, a software pull request, or a payments flow strategy with the click of a button. But when cheap competence skyrockets, adoption explodes, resulting in an unprecedented glut of generic output—what the internet has collectively taken to calling “slop”. It’s the default, lazy answer that lacks soul, context, and nuance.

When everything begins to look and smell the same, a strange thing happens: the market’s demand for genuine difference sky-rockets.

The shift we are facing across all professional services—whether legal, financial, or consulting—isn’t about eliminating the expert. It is about changing the expert’s job from data-retriever to orchestrator and judge. The floor has been raised. Yesterday’s ceiling is today’s baseline.

What remains is the ability to read a room. To watch a client’s shoulders tighten when you present an option that’s technically correct but organizationally impossible. To notice the glance exchanged across the table before anyone speaks. No LLM parses that. The map is universal now; the guide still has to be in the room.

We don’t need fewer guides; we need fewer toll booths. The future of consulting doesn’t belong to those who hoard the map. It belongs to those who use a universally available map to help people actually walk the terrain.

Categories
AI Books Writing

The Tax We No Longer Have to Pay

When Carol Coye Benson and I sat down to write Payments Systems in the U.S., one of the first problems we had to solve wasn’t about payments. It was about history.

To understand why the ACH network works the way it does, or why checks persisted decades longer than anyone expected, you need the institutional sediment underneath — the regulatory decisions, the failed experiments, the path dependencies baked in by choices made in the 1970s that nobody thought would still matter in the 2000s. The history is the explanation. Strip it out and you have a description of current practice with no account of why it exists or what it cost to get there.

But history takes pages. And pages test a reader’s patience. So you compress. You make judgment calls about what survives the cut and what gets left behind, and you make those calls knowing that every omission is a bet — a bet that the reader can follow without it, that the thread holds without that particular knot.

Writing it taught me something. The act of compressing, of finding the minimum sufficient version of a complex thing, forces a clarity that living inside the complexity never quite delivers. You don’t fully know what you understand until you have to say it precisely enough for someone else to follow.

But compression is always a loss. You feel it as you write. The version in the book is thinner than the thing you know.

Garry Tan uses a term — “tokenmaxxing” — that initially sounds like jargon from a performance optimization thread. The idea is simple: don’t be stingy with context. Give the model everything. Every source document, every relevant article, every piece of background that a human reader would never sit still for. Let it synthesize rather than guess.

The instinct it runs against is deep. We have spent decades building information systems around compression — search engines that retrieve rather than ingest, executive summaries that stand in for reports, one-pagers that distill months of work into something a decision-maker can absorb in four minutes. All of it was a rational response to a real constraint: human attention is finite and expensive. You couldn’t afford to read everything, so you built filters. The whole architecture of how organizations manage information was designed around that limit.

Tokenmaxxing is a bet that the limit has moved.

The model can read everything. The cost of giving it full context — the uncompressed history, the original sources, the institutional sediment — is low enough now that filtering before the model sees it may introduce more error than it prevents. You’re potentially discarding signal when you summarize for the model the way you’d summarize for a human. The model doesn’t need the one-pager. It can handle the report.

This doesn’t dissolve the need for curation entirely. More context isn’t always better — models can lose the thread in noise the same way humans do, just differently. The skill shifts from summarizing to selecting: not what’s the minimum version of this but what’s actually worth including. Different judgment, still essential.

But the deeper change is upstream of any particular project. The compression we built into every research process, every briefing, every book — that was never the goal. It was the tax we paid for human cognitive limits. Part of the process doesn’t pay that tax anymore.

When I think about writing that payments book today, I don’t think the book itself would change much — it still has human readers with finite patience. But the map we drew before writing it, the synthesis work, the “what connects to what across fifty years of regulatory history” work — that could happen at a different depth now. The understanding you bring to the writing can be informed by everything, not just the subset you had time to read.

The payments book was written entirely for humans, with all the compression that implies. But Tyler Cowen just published what he calls a “generative book” — 40,000 words released free online, paired on the same screen with a Claude interface so readers can discuss, interrogate, and extend it in real time. He’s writing for both audiences simultaneously now. The human reader and the model that will help that reader go deeper. The text is optimized not just to be understood but to be used — as context, as a jumping-off point, as raw material for a conversation that the author won’t be in.

That’s a different kind of writing. Not better or worse. Different. The compression decisions change when one of your readers has no patience to protect.

Writing still clarifies thinking. That part hasn’t changed. But what you’re clarifying, and who you’re clarifying it for, is quietly expanding.

Categories
AI Programming Software Work

The Scarcest Thing

Garry Tan woke up at 8 a.m. after sleeping at 4. Not because he had to. Because he wanted to see what his workers had done overnight.

The workers are AI agents. Ten of them, running in parallel across three projects. And something about that sentence — wanted to see what they’d done — keeps stopping me. That’s not the language of someone using a tool. That’s the language of someone managing a team.

Tan gave a name to the state this puts him in: “cyber psychosis.” He said it as a joke. But the joke has an insight in it. He’s not describing addiction to a productivity app. He’s describing a shift in what it means to do creative work — the strange vertigo of becoming a director when you’d always been a laborer.

I’m retired. I watch this from the outside now, which is its own kind of vantage point. For most of my career, the path from idea to working product ran through people — through hiring and managing and the slow accretion of execution capacity. You had the vision or you didn’t, but either way you needed the team. The idea and the means of making it real were, structurally, separate things. The gap between them was where companies lived.

What Tan is describing is that gap closing.

The thing he built — gstack, his open-sourced Claude Code configuration — got dismissed in some quarters as “just prompts.” And it is just prompts, in the same way that a conductor’s score is just notation. The abstraction is the invention. What he encoded is a model of how a startup team thinks: the CEO who interrogates the why before a line of code gets written, the engineer who builds, the paranoid staff reviewer who looks for what breaks. Each role blocks a different failure mode. Blurring them together produces, as his documentation puts it, “a mediocre blend of all four.”

That’s an organizational insight. It has nothing to do with code.

Tan described being a “time billionaire” — not because his biological clock had slowed, but because he can now purchase machine-consciousness-hours. The bottleneck of implementation, which has governed every creative project since the beginning of creative projects, is dissolving for those who know how to direct.

The scarcest thing is shifting. It’s no longer the hours of execution. It’s the clarity of intent — knowing what you want to build and why the journey matters, before any of the workers start moving. That’s harder than it sounds. For decades, most of us could muddle through in the making of it. The act of building taught you what you were building. Now the making is cheap, and that shortcut is gone.

For someone watching from retirement, that’s not a small thing to absorb. The model I internalized over a long career — that ideas become real through sustained organizational effort, through teams and timelines and the grinding work of execution — is being revised faster than I expected. Not invalidated. Revised. The judgment still matters. The taste still matters. The why matters more than ever.

It’s just that the how has found new hands. Many of them. More than any team I ever assembled, available the moment the intent is clear enough to direct them, gone when the work is done. The constraint was always the hands. It turns out it was always the knowing.

Categories
AI AI: Large Language Models Anthropic

Breakout

Jack Clark doesn’t panic easily. He spent years at OpenAI watching capabilities inch upward, then left to co-found Anthropic, and has been writing his Import AI newsletter long enough to have developed — and been wrong about — many priors. So when he publishes an essay saying he has reluctantly arrived at a 60% probability that fully automated AI R&D happens by the end of 2028, the word “reluctantly” deserves some weight.

His essay, published last week and titled “Automating AI Research,” isn’t a press release or a fundraising pitch. It reads more like a man thinking out loud at the edge of something large. “I don’t know how to wrap my head around it,” he writes, which is a notable thing to say publicly when you are one of the architects of the thing you can’t wrap your head around.

The argument is built from benchmarks — not any single one, but a mosaic of them assembled to reveal a trend. SWE-Bench, the test that measures an AI’s ability to solve real GitHub issues, was at roughly 2% when it launched in late 2023. A recent Anthropic model sits at 93.9%, effectively saturating it. METR’s time-horizon plot tracks how long an AI can work independently before needing human recalibration: 30 seconds in 2022, 4 minutes in 2023, 40 minutes in 2024, 6 hours in 2025, 12 hours today. The trajectory, if it holds, suggests 100-hour autonomous work sessions by the end of this year.

Clark marshals similar progressions across AI fine-tuning, kernel design, scientific paper replication, and even alignment research itself. His throughline is the same in each: AI is now genuinely competent at the unglamorous scaffolding of AI development — the debugging, the experiment runs, the parameter sweeps, the code reviews. And crucially, it can now do these things not just faster than humans, but for longer, with less supervision.

There’s a Thomas Edison quote at the center of the essay: “Genius is 1% inspiration and 99% perspiration.” Clark’s claim is that AI has become very good at the perspiration. The question of whether it can supply the inspiration — the paradigm-shifting insight, the Move 37 — remains open. But he argues it may not need to. Most of what has moved the AI field forward has been sustained, methodical work, not lone flashes of genius. If you can automate the 99%, you have something that compounds.

There’s a data point that makes Clark’s argument feel less like forecast and more like dispatch. Last month Boris Cherny, who runs Anthropic’s Claude Code, disclosed that he hasn’t written a line of code by hand in more than two months. Every pull request — 22 one day, 27 the next — written entirely by Claude. Company-wide, roughly 70–90% of Anthropic’s code is now AI-generated. Anthropic’s stated position: “We build Claude with Claude.” The loop Clark is describing as a probability by 2028 is already running, at least partially, today.

The word Clark uses for the threshold he’s describing is not “singularity” or “AGI.” It’s quieter than that. He calls it “automated AI R&D” — the point at which a frontier model can autonomously train its own successor. It’s a specific, falsifiable thing. And he puts a number on it: 60% by end of 2028, 30% by end of 2027.

I’ve been writing about the dark software factory and the 3D printer that prints better printers, finding metaphors for what seems like an inexorable process. Clark’s essay is a different kind of writing about the same thing — the primary source document, the engineer’s log, the inventory of evidence. Reading it is a little like watching someone carefully pack boxes before a move. Each individual item seems manageable. But there are a lot of boxes.

What he’s describing — if the trend holds — is not a feature or a product launch. It’s a breakout. The moment the loop closes and the system starts building itself. He’s not certain it happens. He just thinks it’s more likely than not, and he thought you should know.

Categories
AI Business Work

The Tipping Point Was Last November

Matthew Prince, Cloudflare’s CEO, said something on today’s earnings call that I keep turning over. He didn’t bury it or soften it. He named a date.

“Internally, the tipping point was last November.”

That’s a specific thing to say. Not “we’ve been on a journey” or “AI has been transforming our industry.” A month. A moment. The thing changed, and he knows when.

What changed, by his account, is that Cloudflare’s teams began seeing productivity gains so dramatic they were hard to describe — people who were two times more productive, ten times, in some cases a hundred times. “It was like going from a manual to an electric screwdriver.” Usage of AI tools internally is up more than 600% in just the last three months. Every line of production code is now reviewed by an autonomous AI agent.

And then he said goodbye to 1,100 people — about 20% of the company.

Today wasn’t just Cloudflare. Earnings season has become something like a drumbeat. Meta is cutting 8,000 employees this month. Amazon cut 16,000 in Q1. Oracle eliminated roughly 30,000 to fund AI infrastructure. Block cut almost half its workforce. PayPal is reportedly planning to cut 20% of its staff over the next few years. Coinbase cut 14%. Snap cut 16%. As of this week, more than 92,000 tech workers have been laid off in 2026 alone.

The scale is striking. But what strikes me more is the framing — the specific language being used to describe what’s happening. These aren’t being announced as cost-cutting moves or post-pandemic corrections, the way they might have been in 2022. They’re being announced as architectural decisions. Structural adaptations. Evolution.

Prince was careful to be explicit: “This isn’t a cost-cutting exercise or an assessment of individuals’ performance. It’s about defining how a world-class, high-growth company operates and creates value in the agentic AI era.” That’s not empty corporate language, or at least not only empty corporate language. The distinction he’s drawing — between trimming fat and reimagining how a company is built — maps to something real about what AI agents can now actually do.

There’s a legitimate version of this argument and a convenient one, and they’re being delivered in the same sentence by the same people, which makes them hard to separate. Some analysts suspect companies are using AI as cover for cuts they wanted to make for other reasons — rightsizing from pandemic-era overhiring, funding massive infrastructure buildouts, chasing margin. Oxford Economics flagged this: maybe some firms are “dressing up layoffs as a good news story.” The cynicism is warranted.

But then there’s the Cloudflare number: 600% increase in AI usage in three months. That’s not a narrative. That’s a measurement.

What’s different about this moment — what makes Prince’s “tipping point” language feel accurate rather than convenient — is that the people making these decisions are themselves users of the tools. They’ve seen the productivity numbers internally before anyone else has. They’re not theorizing about what AI might do to their workforce; they’re describing what it already did.

That’s the thing that changed. For years, AI’s labor impact was a future tense conversation. Economists studied it, think pieces warned about it, conferences debated the timeline. Then, somewhere around last November apparently, a cohort of technology companies crossed from hypothetical to empirical. The future tense became past.

Whether you read that as tragedy, as transformation, or as both depends on where you’re standing. 1,100 people at Cloudflare today are standing somewhere very specific. Prince acknowledged this with what felt like genuine difficulty: “A number of friends will no longer be colleagues.” Whether that difficulty changes anything material for the people leaving is a fair question.

But the acceleration itself — the thing he named — is real. The tipping point was last November. And if it was last November for Cloudflare, it was some nearby month for Amazon, for Meta, for Block, for all of them. Whatever these companies learned that changed everything, they all seem to have learned it around the same time.

That’s what I find myself sitting with today: not just the scale of the disruption, but the synchrony of it. The realization arrived, and then the decisions followed. Quietly at first, then all at once.

Categories
AI AI: Large Language Models

The 3D Printer That Prints Better Printers

Imagine a 3D printer that looks at its own design and begins printing a better version of itself. The loop closes. What had always required an external human intelligence now happens inside the machine. All by itself.

Jack Clark — Anthropic co-founder, someone who has spent years closer to this technology than almost anyone — puts the odds of this happening by 2028 at better than even. I have been turning that number over ever since I heard it. Not the technical claim, exactly. The feeling of it.

We have grown used to AI accelerating our work. Coders watch models close GitHub issues at rates that would have seemed miraculous eighteen months ago. Researchers delegate experiment design, kernel optimization, even the fine-tuning of smaller models. The scaffolding of AI progress is already being built, in part, by the systems themselves. But the moment the system begins to redesign the scaffolding — that is something new.

What unsettles me is not the raw capability, though that is staggering. It is the loss of distance.

For most of technological history, the creator stood outside the creation. Even the most sophisticated tools remained tools. Now the distinction begins to blur. A model that can meaningfully improve its own training process, its own architecture, its own alignment constraints, is no longer merely reflecting human intent back at us. It is participating in the shaping of its own nature. And because each iteration can happen faster than the last, the curve steepens in ways our intuitions, tuned to linear progress, struggle to grasp.

Clark is careful, as he should be. He speaks of validation work that will still fall to humans, of the need to broaden the pipes through which abundance flows, of preparing defense-dominant postures against misuse. Yet the image that lingers for me is quieter: the silence after the handoff. What does it feel like when the thing you have been painstakingly teaching begins to teach itself — and then to teach its teachers?

I think about Leo Szilard at the traffic light, or the first controlled chain reaction under the stands at the University of Chicago. Moments when a new regime of possibility quietly announced itself. Recursive self-improvement carries that same charge — not a single event but a process, one that could accelerate the very pace of events themselves.

The more I sit with it, the more I return to an older tension in our relationship with tools. We build them to extend ourselves, and in doing so we are always, subtly, extending — or perhaps risking — what we are. The values I try to live by — generosity, curiosity, compassionate honesty — are not refined in specifications. They are refined in friction, in relationship, in the slow work of being human with other humans. If the machines begin to optimize their own lineage at speeds we cannot match, will we still have the bandwidth to tend the parts of ourselves that no algorithm can yet measure?

I don’t know. None of us do. That uncertainty feels honest.

What feels clearer is the invitation. Not to fear the printer that prints better printers, nor to worship it, but to remain awake inside the loop. To ask, as each new version arrives, what kind of world we are collectively printing — and whether the values we claim to hold are baked into the design or merely etched on the surface, likely to wear away under the heat of iteration.

The light is still yellow. We are still deciding whether to step off the curb. But the traffic is already moving faster than it was a moment ago.

Categories
Science Stanford

Bypassing the Leaf

For my entire life, I’ve understood the world through a simple, quiet equation: green plants take sunlight and air, and turn them into the stuff of life. It is a slow, terrestrial magic we all learn in grade school.

But lately, after listening to Professor Drew Endy at Stanford, I’ve been sitting with a curious yet exciting realization: that ancient equation is being rewritten.

Professor Endy champions a concept called electrobiosynthesis, or eBio. At its core, it represents the engineering of a parallel carbon cycle that operates independently of traditional photosynthesis.

The global industrial complex is approaching a transition point where our traditional reliance on extractive fossil fuels is being superseded by a regenerative, biological manufacturing paradigm.

For millennia, humanity has relied on the biological “middleman” of the plant to capture solar energy. But natural photosynthesis, for all its quiet beauty, is limited by severe biochemical constraints. Most commercial crops convert less than 1% of incident solar energy into usable biomass.

Electrobiosynthesis changes the math. By bypassing the plant entirely, we can utilize high-efficiency photovoltaics—which capture over 20% of the sun’s energy—to drive carbon fixation directly into the metabolic hubs of engineered microbes. This fixed carbon is transformed into organic molecules, serving as the feedstocks for high-value products like proteins and specialty chemicals.

In my own career, I’ve watched industries undergo profound, structural phase shifts. This really feels like another one of them. It seems that we are looking at a future where any molecule that can be encoded in DNA can be grown locally and on-demand. This fundamentally decouples manufacturing from centralized industrial nodes and fragile global supply chains.

The field appears to currently be in its “transistor moment,” moving from laboratory feasibility to industrial pilot plants. It signifies the ability to construct and sustain life-like processes without being restricted to the terrestrial lineage of photosynthesis.

Of course, with such foundational power comes the weight of unintended consequences. The ability to engineer life at this level brings severe biosecurity risks, and even the “Sputnik-like” strategic challenge of international competition in biotechnology. There are profound ethical dilemmas on the horizon, such as the creation of “mirror life”—organisms made from mirror-image biomolecules that might be invisible to natural ecosystems.

But the trajectory seems set. The vision described by Professor Endy—a world where we grow what we need, wherever we are, using only air and electricity—is no longer a distant science fiction. It is a nascent industrial reality. This future is being written not in sprawling factories, but in the microscopic architecture of the cell.

I’ve just now reading a deep research report on this whole area that I asked Google Gemini to create. It’s fascinating and I’ve discovered a whole new area (beyond AI) to explore further.