Don't Expect AI Progress to Sigmoid Anytime Soon
“All exponentials eventually become sigmoids.” It’s one of the most common refrains from AI skeptics. Presented with a graph of rapidly improving AI capabilities, they caution that the curve will inevitably flatten.
Scott Alexander, writing on his Substack blog Astral Codex Ten, offers a systematic takedown of this argument.
He begins with the “Sigmoid Misidentification Hall of Fame.” UN birthrate projections repeatedly predicted declining rates would level off — they kept declining at the same pace. World Energy Organization forecasts for solar deployment consistently underestimated actual growth. Most recently, a Wharton team modeled AI capability trajectories in early 2026, predicting a plateau — only for the next AI model release to land squarely on the continued exponential trend.
Alexander’s central point: it is technically true that no process grows exponentially forever. But the critical questions are when the sigmoid hits and at what level. If AI capabilities plateau somewhere near or above human-level performance, that ceiling has no practical constraining power for current planning and investment.
Under genuine uncertainty, he argues the default should be Lindy’s Law: a mysterious process will, on average, continue for as long as it has already continued. AI has been improving rapidly since at least GPT-1 in 2017 — roughly 7 years — so the default expectation is ~7 more years of comparable progress.
For those claiming AI will plateau soon, Alexander demands specificity: either present a concrete model of why scaling will hit a wall, or accept Lindy’s Law as the default. The empty slogan that “all exponentials become sigmoids” is not a serious argument.
For Agent Economy readers, this piece is directly relevant. Whether AI progress continues determines infrastructure investment horizons, application development timing, and the strategic planning of the entire ecosystem. Prematurely declaring an end to progress may be a far costlier mistake than being too optimistic.