On the Brutality of Amdahl’s Law
The author, a former theoretical physicist, argues that superintelligence won't cause the explosive economic change rationalists predict, because physical, regulatory, and informational bottlenecks cap how much faster the overall world can move. He uses Amdahl's Law and a widget-factory example to show that speeding up one step rarely speeds up the whole process by much.
- Amdahl's Law: if a process has several sequential bottlenecks, infinitely speeding up one still leaves a constant-factor overall speedup, so exponential gains require repeatedly removing the current bottleneck.
- In the factory example, even with omniscient AI agents working 24/7, a month-long fix only drops to about a week because of shipping times, lab processes, and installing physical hardware.
- Some bits of information are very expensive to acquire (like long-term side-effect data), and regulation and political will are slow speed limits that intelligence doesn't remove.
The author's core claim is that ASI-driven "explosive world transformation" is unlikely because the world is bottlenecked by things intelligence doesn't fix. He starts from personal experience: after years in theoretical physics, he learned that progress is usually limited by the slowest step, not by raw brainpower. He was relieved in December 2025 when Claude Code plus Opus 4.5 made agents feel powerful enough to threaten white-collar jobs, and bought "insurance" against rapid unemployment, only to watch those predictions fail to materialize. Software engineering and math are the exceptions, and even there millions are still employed.
To explain why, he borrows Amdahl's Law. If a process has sequential steps with rates k_i, the total rate is limited by the slowest one. Push every rate but one to infinity and you still get a finite overall speed, roughly bounded by the remaining bottleneck. In a hierarchy of bottlenecks, removing the first m of them yields only a constant-factor aggregate speedup. Exponential overall gains require repeated explosive improvement at whatever the current bottleneck is. His worked example is a widget factory whose defect rate doubled: an ASI agent can narrow the cause to impurity X in material Y, and both suppliers' precursors contain Y, so diagnosis still requires physical tests, sourcing decisions, purchasing, installation, and verification. Even with perfect decisions and 24/7 operation, he estimates a month-long fix might shrink to about a week, a roughly 4x speedup, because shipping times, chemical and biological process rates (cell culture and annealing take days), and physical tool throughput set hard floors.
He then sorts the remaining bottlenecks into two categories. The first is physical and time-bound constraints intelligence doesn't accelerate, such as shipping, lab incubation, hardware installation, and manufacturing cycles. The second is more institutional and social, which the excerpt sets up but doesn't fully develop in what's shown. He also notes that many problems are effectively "solved" yet still slow and costly to improve: the steam engine is 300 years old, and fusion plants will likely still drive steam turbines. His position is not that AI is unimpressive but that the world's speed limits are far more stubborn than rationalist essays about double-digit economic growth tend to assume.