The advent of new AI tools has worried many about displacement from automation. The common reaction follows a Pigouvian intuition: automating capital harms workers, so it should be taxed. This paper argues this is misguided because capital has opposing effects on workers at different margins: harming them when automation first occurs but helping them as it accumulates thereafter. A policy targeting initial adoption can therefore address automation directly, and with optimal nonlinear taxes, makes the optimal capital tax zero -- recovering the celebrated result of Atkinson and Stiglitz (1976) despite the dependence of relative wages on automation. More broadly, even when the tax system is not optimized, targeting adoption lowers optimal capital taxes: from 35.4% to 20.9% under the US system, versus 13.7% to zero under an optimal nonlinear tax. Simulating the introduction of AI, capital taxes should grow with the productivity gains from AI but are always substantially lower with adoption targeting.