Your Salary Ceiling in the AGI World
I enjoyed the work by Restrepo from Yale University. Lots of formulas to describe simple models of the near-future economy, and a set of very direct conclusions.
AGI makes it possible to perform all economically valuable work through computation. This radically changes the role of labor: people retain the ability to work, but their contribution to economic growth becomes increasingly insignificant.
Good classification:
Bottleneck work - critical tasks without which growth is impossible (energy, science, security, logistics).
Accessory work - auxiliary tasks that will likely remain with humans (art, care, social roles, service).
And the main existential question you should ask yourself:
“How much will it cost to operate a model that does my job 24/7?”
This figure will be close to your salary ceiling.
The Simple Economics Behind Your Wage Cap
The logic is simple. If AGI handles all bottleneck tasks, salary will no longer depend on skill rarity or years of experience. It will start depending on how much computational power costs to replicate your labor.
Above this threshold will be what’s sustained by human trust, signing authority, personal relationships, and reputation.
But here’s what Restrepo’s mathematical models reveal about the deeper mechanisms at play:
The Great Economic Shift
In today’s economy, output is multiplicative - human labor and capital work together, each amplifying the other’s value. Remove either component, and productivity collapses. Your wage reflects not just what you produce, but how essential you are to the entire production process.
In the AGI economy, output becomes additive - computational resources and human labor simply stack on top of each other. The economy can grow indefinitely by expanding compute alone. Human contribution becomes a nice-to-have add-on rather than a multiplicative necessity.
Restrepo’s model shows this mathematically. Where we once had: Output = Function(Labor × Capital × Technology)
We move toward: Output = A × (Compute + Human Labor Equivalent)
Where “A” is simply the rate of transformation from compute to economic value.
Why Some Jobs Will Persist (But Stay Poor)
The paper makes a crucial distinction that explains why certain human roles will survive, but remain economically marginal.
Bottleneck work gets automated because it’s essential for growth. If energy production, logistics, or scientific research aren’t expanding, the entire economy hits a ceiling. AGI will inevitably take over these domains because computational resources can scale infinitely while human capacity cannot.
Accessory work might remain human not because we’re better at it, but because we’re cheaper and already abundant. Therapy, hospitality, creative arts - these domains may require enormous computational resources to replicate human warmth, intuition, or social understanding. The compute cost might exceed the economic value, leaving these niches to humans.
But here’s the catch: accessory work doesn’t drive growth. Its value remains fixed while the economy expands around it. You might keep your job as a therapist or artist, but your relative economic importance shrinks toward zero.
The Two Transition Scenarios
Restrepo identifies two possible paths to this future, each with very different implications for workers:
Scenario 1: Compute-Constrained Transition If computational resources are the bottleneck, the transition is gradual and predictable. Work gets automated sequentially, from computationally easiest to hardest. Wages decline smoothly as each profession reaches its “compute equivalency point.” Workers have time to adapt and retrain.
Scenario 2: Technology-Constrained Transition
If AGI breakthroughs are the bottleneck (but compute is abundant), the transition becomes chaotic. The moment researchers crack the code for automating your profession, displacement happens immediately. Some workers see their wages spike temporarily - not because they became more valuable, but because they randomly happened to work in the last domain to be automated. Then their wages collapse overnight when the breakthrough arrives.
We’re likely in Scenario 2. Compute is increasingly abundant, but AGI breakthroughs arrive unpredictably. This creates a lottery-like labor market where your career prospects depend on the random timing of research progress in your field.
The Mathematics of Inequality
Perhaps most sobering is Restrepo’s calculation of how economic gains will be distributed. In the AGI economy, the “labor share” of GDP - the percentage of economic output that goes to workers - approaches zero.
This isn’t because society becomes poorer. The economy grows faster than ever, powered by exponentially expanding computational resources. But human labor’s contribution becomes mathematically negligible.
Think of it this way: if human brains can perform 10^16-10^18 operations per second, and our economy’s total compute approaches 10^54 operations per second, then human contribution becomes a rounding error. Even if every human worked optimally, we’d represent less than 0.00001% of total productive capacity.
The gains from this massive productivity increase flow almost entirely to whoever owns the computational infrastructure.
Beyond Individual Strategy
This analysis suggests that traditional career advice “learn valuable skills,” “stay ahead of automation” - misses the fundamental shift occurring. In the AGI economy, individual human skills become commoditized at their computational replacement cost.
The relevant questions become structural:
“Where do I want to be an irreplaceable person in the real world?” This means focusing on roles that require human trust, legal authority, or personal relationships that can’t be computationally replicated, not because of technical limitations, but because of social and legal structures.
“How can I own a part of the computational economy?” Since returns flow to compute owners, the path to prosperity runs through ownership, not labor. This could mean direct investment in AI infrastructure, equity in AI companies, or advocating for policies that distribute compute returns more broadly.
The Uncomfortable Truth
Society as a whole will become richer. But wealth will be created not by labor, but by computation.
Owners of computational power will take the main share of this prosperity.
People will be left with accessory work - service, care, creativity, social roles. These retain meaning and value, but don’t become sources of growing income.
Work will cease to be the foundation of universal prosperity and status.
Restrepo’s paper ends with a haunting observation: in the AGI economy, human labor becomes so economically peripheral that “if tomorrow half the population stopped working, no one would notice.” We retain the ability to contribute, but our contributions no longer matter for aggregate prosperity.
This isn’t a prediction about technological unemployment - it’s something more subtle and perhaps more profound. It’s about the end of work as the primary source of economic value and social meaning.