Could AGI Freeze Economic Mobility? A Debate About Work and Ownership

Could reaching artificial general intelligence lock people into a permanent underclass? I have heard this concern from people working around AI in Silicon Valley, and I want to examine it because the social implications would be enormous. I will also explain why I remain more optimistic about the outcome.

For this discussion, I use AGI to mean AI that outperforms people at knowledge work, from programming and customer service to data analysis. The worried scenario assumes that, once such systems are available, buying compute becomes the main constraint on producing that work. If human labor loses much of its economic leverage, people who already own capital could have an enormous advantage. These are assumptions about a possible future, not established facts.

AI companies are investing heavily in computing infrastructure. In this scenario, owners could use capital and electricity to replicate AI capability at scale without the constraints associated with human employees. The fear is not only poverty: if returns are concentrated and people have little means to build wealth, economic mobility could stall.

I hear some people express urgency about accumulating capital before AGI arrives, because they fear their position could become fixed afterward. This is a concern rather than a demonstrated inevitability. I also point to a difficult entry-level job market as an early indicator worth examining.

The Stanford Digital Economy Lab’s 2025 ‘Canaries in the Coal Mine’ study used ADP payroll data and reported an approximately 13% relative employment decline for workers aged 22–25 in highly AI-exposed occupations, controlling for firm-level shocks. More experienced workers in those occupations were stable or growing, and the paper described adjustment primarily through employment rather than pay. The comparison is descriptive evidence of a pattern, not proof that AI alone caused it.

Handshake also reported that job postings on its platform fell 15% over the year while applications per job increased 30% for the class of 2025. That indicates greater competition in this platform’s data; it is not a measure of all jobs or a causal estimate of AI’s effect. These observations deserve attention without being treated as proof of a permanent underclass.

Large infrastructure investments may also make inference cheaper. I expect the cost of running AI to fall toward the costs of hardware and electricity, though actual prices will depend on many factors. Cheaper access could widen the benefits, but it does not by itself settle who owns the resources or captures the gains.

I raise Jevons’ paradox: when a resource becomes more efficient or less expensive, new uses can become economical, increasing total consumption. If that happens with compute, demand could expand even as the cost per unit falls. Whether overall spending rises depends on the size of the demand response; it is not automatic.

One proposed response is universal basic income: regular unconditional cash payments intended to help people meet basic needs. Some, including Elon Musk, have also discussed universal high income, a more ambitious idea in which people would have enough resources to choose how to spend their time. Both depend on how they are funded and designed.

Critics argue that basic income resembles socialism or could contribute to inflation. I refer to a conversation with advocate Scott Santens, who argues that financial security can encourage entrepreneurship and free people to address problems or start businesses. That remains an argument about policy effects, not a guarantee that AI would not also perform those tasks.

My more optimistic view draws on an argument that AI will often handle the middle of a task while people remain at its edges. A person might define a goal, prompt or orchestrate agents, then review and verify the output. The boundary may move as systems improve, and how much oversight remains necessary will vary by task; I expect many jobs to change rather than disappear entirely.

In an interview clip, Box CEO Aaron Levie argues that companies should reward teams that use AI to become more efficient, allocating them resources to pursue additional problems. In a later conversation he says his view has not changed: if AI lets a company grow faster, new bottlenecks may create demand for more people in sales, consulting or customer success. This is his prediction about possible complementary hiring, not a measured outcome for every company.

An OpenAI technology leader in another clip notes that the human role at the edges may become thinner, but models still hallucinate and make mistakes. The appropriate amount of oversight depends on the error tolerance of the task. The clip identifies the speaker by role but does not name them in the narration.

I expect new jobs that are difficult to imagine today, as well as opportunities for founders and individuals to do more with fewer resources. That is a forecast, not evidence that new positions will necessarily offset every job lost or that the transition will be smooth.

In a clip from our interview, Reid Hoffman suggests that cheaper software creation can unlock demand for tools tailored to individual tasks. His example is visual-effects software: if creation becomes less expensive, smaller productions such as documentaries might access capabilities once limited to large-budget films. He presents this as a possible expansion of creative demand.

I expect significant changes in knowledge work and think people will need support learning to use AI tools productively. Deployment also takes time: a model’s capability does not mean an organization has the infrastructure or processes to use it everywhere immediately. My prediction is that some jobs will be automated while many others are transformed.

Basic income or high income might work in a more productive economy, but their feasibility and consequences remain open questions. I am hopeful that AI will expand creativity and entrepreneurship, improve health and create abundance. Whether this transition resembles earlier technological shifts, and how broadly the benefits are shared, remains to be seen.