[!todo] Seed note. A starting point, not a finished note yet.
Amdahl’s law, from Gene Amdahl’s 1967 argument about parallel computing, states that the maximum speedup of a system is limited by the fraction of the work you do not accelerate. If a stage is fraction p of total time and you make it infinitely fast, the best possible improvement is 1/(1-p): optimize 30% of the work to zero and you have still only cut total time by 43%. The law is the formal reason AI coding tools raise organizational throughput far less than they raise coding speed. Writing code is a minority of the delivery cycle, so accelerating it hits a hard ceiling while the untouched stages (review, integration, release) dominate what remains. Seeded from AI SDLC Bottleneck; it is the quantitative sibling of the Theory of Constraints argument that only the constraint stage governs output.