\section{Conclusion, limitations and open questions}
\label{sec:conclusion}

The chapter we build on ends with the sentence ``in the AGI economy, we won't be missed.'' Our results say
something more specific about who will be missed by whom. Labor will not be missed by output: that result is
robust to everything we tried, because it rests only on the perfect substitutability of labor and compute within
tasks. Labor's \emph{income}, however, will be missed by labor, in a way the source does not emphasize: unless
compute remains scarce, the compute that replaces a worker is cheap and so is the worker, in level and not only
in share. Where compute does not remain scarce, the owners of what compute cannot make --- land, sites, the
structures that take time to build --- are the ones who are not missed. And what those owners hold is valued at
the duration of their claims, which is why the wealth-to-income ratio first falls, while the capital share
rises and the real rate is high, and then rises when land captures the income.

Three limitations bound these conclusions. First, the paths are perfect-foresight equilibria. There is no
aggregate risk, no risk premium and no learning about the arrival of AGI; the one-period lottery of
Section~\ref{sec:robustness} shows the direction in which arrival risk would move premia but is not an
equilibrium under uncertainty, and the immediate jump of the real rate in 2026 is the model's way of pricing
in a path that nobody knows. Second, the level of wages after 2060 depends on the product of two scenario
inputs --- how cheap installed compute becomes once Moore's law ends, and how much of it a worker-equivalent
requires --- that no historical data identify; the paper reports the dependence rather than a number. Third,
the production structure is fixed. There is no endogenous innovation, no creation of new human tasks beyond the
reinstatement scenario, no nonhomothetic demand that would make human-made goods luxuries, no occupational choice
or labor supply, and the heterogeneous economy has no return heterogeneity. Each of these would change the
distribution of the gains; none of them, as far as we can see, changes the arithmetic of the compute-equivalent
wage.

Four questions follow. What does the compute-equivalent wage look like under uncertainty about the arrival
date, when the premium of the last-automated skill is a lottery rather than a path? How do the valuation results
change when compute services are supplied with market power that is contestable rather than permanent? What is
the optimal mix of a land tax, a compute tax and a citizen fund when the planner cannot observe who owns what?
And, most directly testable, do the wages of exposed occupations fall when the compute-equivalent cost of their
tasks crosses the wage, rather than when the technology becomes feasible? The last question is the one the next
decade will answer.