UBI or We’re Screwed Part 2

About a week after I published Part 1, I felt the temperature change online. A lot of people were clearly doing the same thing at the same time: taking a new generation of coding agents, pointing them at real work, and realizing the floor moved overnight.

The timing made sense. When people finally got uninterrupted hours to experiment, especially over a holiday lull, the collective denial got harder to maintain. Once you use these tools on a domain you already understand deeply, you stop arguing about whether they are “real,” and you start arguing about what happens when they keep getting better.

Andrej Karpathy captured the vibe perfectly when he wrote, “I’ve never felt this much behind as a programmer,” and described the profession being “dramatically refactored” as we learn a new abstraction layer involving agents, prompts, memory, tools, MCP, LSP, workflows, IDE integrations, and a bunch of other knobs we did not have to think about a couple years ago.

Then Jaana Dogan, a Google Principal Engineer, posted something that landed like a brick: she said her team had been trying to build distributed agent orchestrators for about a year, and that after she gave Claude Code a description of the problem, it generated what they built in about an hour. She also emphasized it was a toy version and not production grade, which is an important nuance, and it still does not soften the core signal.

Software was never the endgame. It was the easiest place to prove the point, because it has cheap verification. Tests pass or they do not. The build is green or red. A deploy works or it pages you. That makes it a perfect environment for agents to iterate, self-correct, and get trained against objective feedback.

But most white-collar work is also full of scoreboards. It just hides behind nice titles. Accounting closes or it does not. Claims are approved or denied. Contracts pass compliance review or they do not. A spreadsheet ties out or it does not. A quarterly report reconciles with source systems or it does not. A huge fraction of “knowledge work” is language plus rules plus checklists, sitting inside software systems that already define correctness.

The “decimation” does not start with a dramatic moment where everyone gets fired on the same day. It starts with hiring slowing down. Intern classes shrinking. Junior roles vanishing. Teams realizing they can ship the same roadmap with fewer people because agents eat the glue work, the first drafts, the basic implementation, the documentation pass, the test scaffolding, the refactor churn.

Dario Amodei told Axios that AI could wipe out half of entry-level white-collar jobs and push unemployment to 10% to 20% within one to five years. You can debate the exact numbers, but the direction of the warning matters: it is explicitly white-collar, explicitly entry-level, and explicitly soon. To the chagrin of many skeptics, even some of Dario Amodei’s most hyperbolic proclamations, like AI “writing 90% of code,” are turning out to be relatively on the mark.

Yes, some people will thrive. There will be a class of roles that are basically “manage the AI systems.” People who can translate messy goals into clear specs, build evaluation harnesses, set permissions, supervise tool access, constrain failure modes, and keep organizations from shipping stochastic nonsense into production.

But that is not a mass-employment story. That is a leverage story. A small number of people become wildly productive because they can command fleets of tools. Everyone else competes against a machine with falling marginal cost, and against a labor market where the apprenticeship ladder is missing rungs.

This is why I keep coming back to the same uncomfortable conclusion: white-collar work is not going to be “transformed” in a way that preserves the current number of white-collar jobs. It is going to be compressed. Hard.

This is the part that makes me feel nauseous, because we are not starting from a stable baseline. The Federal Reserve’s Distributional Financial Accounts show that as of mid-2025 the top 1% held roughly a third of total net worth, while the bottom half held only a few percent. That is not a vibe. That is the ownership map of the country.

When ownership is that concentrated, the system becomes brittle. People do not have buffer. They do not have slack. A layoff is not “a transition,” it is a rent crisis. A medical bill is not “an inconvenience,” it is a spiral. The middle class stops feeling like a stable platform and starts feeling like a temporary condition.

Housing is where this shows up most brutally. A massive share of renter households spend more than 30% of their income on rent and utilities. That means millions of people are one bad month away from crisis, even before you add job disruption into the picture.

And the floor is already breaking for many people. The number of people experiencing homelessness in the United States hit the highest level ever recorded in early 2024, with a sharp year-over-year increase. You cannot look at that and call the system healthy.

We often talk about the left-right divide as if it’s just cultural noise. But polarization is also what it looks like when people stop trusting institutions to deliver a fair life, and when the economy feels structurally rigged.

Large majorities of Americans say the country is deeply divided on core values. Trust in government has fallen to near-historic lows. Confidence in institutions that are supposed to arbitrate conflict has become sharply partisan. That is not just annoying politics. That is a legitimacy problem.

A large majority of Americans believe politically motivated violence is increasing. Even if most people never encounter violence directly, the expectation of it changes behavior. It makes people more fearful, more reactive, and more willing to tolerate extreme measures in the name of order.

Election administration and civic institutions increasingly operate under threat and harassment pressure. That is an insane thing to normalize in a functioning democracy, and it tells you something important about how brittle the system has become.

Now add AI to this already-fraying baseline.

AI is a productivity explosion that behaves like capital. Whoever owns and controls the systems captures the gains. Whoever sells labor competes with a machine whose marginal cost trends toward zero.

If AI disproportionately complements higher-income workers and owners, then their income and wealth rise faster than everyone else’s. Productivity gains accrue to capital, not labor. In a country where ownership is already heavily concentrated, this is how inequality goes from “bad” to “socially destabilizing.”

If labor becomes less necessary to produce goods and services, then tying survival to labor stops making sense as a system design. It becomes a cruelty machine. Not because anyone is malicious, but because the inputs changed and we refused to update the distribution mechanism.

I am not saying UBI is morally perfect or politically easy. I am saying it is structurally aligned with what is happening. If AI makes the economy more productive with fewer humans, then we need a durable cash floor that is not contingent on being “useful” in the labor market at that moment.

UBI is not the only policy we need. We also need housing supply, healthcare sanity, education and retraining, and probably new models of ownership. But none of that works if people cannot make rent during the transition.

A cash floor is the simplest, most direct stabilizer for a world where white-collar work compresses faster than politics can adapt. It gives people time to breathe, time to train, time to move, time to take a risk, and time to not get radicalized by desperation.

Here is the bad case that I think is completely plausible if we refuse to act: white-collar opportunity shrinks, the job ladder breaks, and the winners are whoever already owns the systems, owns the equity, owns the compute, owns the data, owns the leverage. Meanwhile the loser majority lives in an economy where housing is already tight, homelessness is already at record levels, trust is already collapsing, and large majorities already believe political violence is rising.

In that world, the left-right divide does not calm down. It becomes a permanent fight over blame and punishment, because people can feel the system failing them. A society can survive cultural conflict if most people still believe tomorrow will be better. It does not survive cultural conflict when tomorrow is visibly being reserved for the owners of the machines.

So I’m praying that UBI, or something extremely close to it, becomes a first-tier political priority in the U.S. Not a niche tech debate. Not a “someday.” A now. Because if we let AI explode the wealth gap while we already have record housing stress and collapsing trust, we are not just “screwed” in theory. We are screwed in the messy, violent, human way.