The Roads, Not the Engine
Why AI policy that ignores the economic system is missing the point
In the last month, two AI-focused essays have gone viral, both because they vividly portrayed seismic economic upheaval about to engulf the world. Matt Shumer1, an AI entrepreneur, drew a parallel to February 2020 — a moment when most of us had no idea how the world was about to change. He argues we are on the precipice of another such watershed, citing how his own daily work is being displaced by the latest AI coding models.
The second essay, by Citrini Research2, triggered an estimated $300 billion sell-off in tech stocks. Writing from an imagined vantage point in 2028, the authors look back at 2026 and 2027 with grim awe as super-capable AI agents rapidly colonize the bedrock infrastructure of global commerce, bringing the likes of Uber, Visa, and India’s entire IT sector to the brink. Its publication was well-timed with Anthropic’s release of agentic tools that can plausibly displace numerous white-collar workflows. To date, announcements like these have contributed to a roughly $2 trillion market wipeout.3
Reading these pieces and watching the market panic to the tune of $1.3 trillion makes me uneasy. But I’m not surprised. I wrote last September:
This collapse will not be a single crisis but a convergence: the unemployment of the Great Depression, the contagion of the Great Recession, the disruption of the COVID-19 pandemic. Each was survivable alone. Together, they may unmake nations.
Nobody Has a Plan
One thing contributing to the sense of dread is that no one in positions of leadership seems to have even a clue about what to do. This week’s State of the Union offered zero substantive engagement with what is almost certainly our next species-defining technological evolution.4 Meanwhile, the AI lab leaders are anything but aligned. Anthropic’s Dario Amodei seems genuinely concerned about what’s coming, but his company’s proposed solutions — economic trackers, AI-powered insight dashboards — are several orders of magnitude too small for the consequences they anticipate from the technology they are building.
That said, I have to pause here and applaud Amodei for what is unfolding today.5 As I write this, Anthropic is refusing the Pentagon’s demand for unrestricted military use of its AI models, holding firm on two red lines: no mass domestic surveillance, and no fully autonomous weapons. Defense Secretary Hegseth has threatened to invoke the Defense Production Act and designate Anthropic a supply chain risk — a label previously reserved for U.S. adversaries — and the White House has ordered all federal agencies to stop using the company’s technology. Amodei’s response: “These threats do not change our position. We cannot in good conscience accede to their request.” Whatever one thinks of the AI industry, this is a private company walking away from hundreds of millions in government revenue on principle. That takes backbone.
It’s the System, Not the Technology
Part of the problem is we are too focused on the technology, and not paying enough attention to the system into which it is being born. What can this or that model do? How many autonomous agents are chatting with each other on MoltBook? How many hundreds of billions did NVIDIA report last year? It’s always about the people and the tech.
But looking at the models themselves, or making small adjustments to existing policy, is not going to cut it. We have to understand that hooking up autonomous digital super-intelligences to this version of capitalism is not going to work.
For 200,000 years, the economy has relied on human input for the majority of its value creation. Physical labor gave way to knowledge work, but throughout that entire journey, humans have always been a necessary element. Now, for the first time, much of what we consider value-creating work will soon be replaceable — mostly, if not entirely — by machines that do not get tired, do not complain, do not advocate for a better deal. Machines that improve recursively, and are beginning to become the authors of their own superior progeny.
The Amish Buggy Problem
Expecting that we can pour this quantity and quality of machine labor into the world economy and everything will go smoothly is like buying a horse-drawn buggy from an Amish cartwright, bolting on the 1,500-horsepower engine from a Bugatti Chiron, and believing you’ll simply arrive at your destination more efficiently. Politicians saying “we need to invest in job retraining programs” are bystanders looking at this dubiously souped-up Amish roadster and suggesting better seatbelts.
What makes it possible to enjoy driving a performance sports car only begins with a powerful engine. The frame, suspension, cabin, electronics, safety systems — all of it makes it possible to elegantly wield the awesome power under the hood. But then there are the roads. No one ever thinks about how important it is to have smooth, paved roads with effective drainage, signage, rules governing the behavior of other drivers. AI is going to become the engine of the next global economy. We are spending all of our time marveling at the engine. Almost no one is talking about the roads.
What I’m Seeing Firsthand
I am experiencing what Shumer writes about. In the last two weeks, I have helped to roughly 30X our company’s productive output by leveraging AI coding tools. I now create the vast majority of our pull requests — they still need quality control from our CTO, but they’re pretty good — and I never even finished the Code Academy course I began in 2020 when I was starting Grantable. It is wonderful to finally express the product I have had in my mind for all these years.
But I’m also acutely aware of how many software subscriptions we will soon be canceling, and how many roles we will never hire for. On Friday, fintech company Block announced they were laying off 4,000 of their 10,000 employees specifically because of efficiencies achieved through AI.6 The share price jumped.
The Cart Before the LLM
The core logic is not hard to see. Firms that prioritize shareholder value above all will have no choice but to offload as much human labor as they can, as quickly as they can, and pour the savings into the AI systems used to replace it. This has been happening for years through outsourcing and more quaint forms of automation. Now it is about to accelerate beyond the known horizon.
Any discussion of AI policy that does not begin by recognizing the need for a fundamentally new economic paradigm is missing the point entirely. We cannot base the conversation on priors that assume human labor and intelligence remain the foundational input in value creation. Anything else is putting the cart before the LLM.
https://fortune.com/2026/02/16/trillion-dollar-ai-market-wipeout-investors-bet-winner/
https://www.axios.com/2026/02/26/trump-state-of-the-union-ai
https://www.nytimes.com/2026/02/27/us/politics/anthropic-military-ai.html
https://www.nytimes.com/2026/02/26/technology/block-square-job-cuts-ai.html?searchResultPosition=1