AI strategy is a self-rewriting equation — solving one constraint changes which constraint matters next

Gupta explains why AI investing “feels so different from software investing.” In SaaS, the metrics that mattered — “CAC payback, NDR, magic number, Rule of 40” — were downstream of two forces: distribution cost and switching cost. AI breaks pattern-matching “for 3 distinct reasons. The system has more variables, the variables are coupled unevenly, and each variable decomposes into sub-variables moving on their own curves.” Capability alone splits into reasoning, context, multimodality, tool use, planning, memory, and controllability, each on its own curve. Because “solving one constraint changes which constraint matters next,” you are “analyzing a system where the equation rewrites itself every hour.”

This is the structural reason behind Build for the model six months from now, not the model of today — a static plan is obsolete because the variable you optimized for stops being the binding one. It generalizes Sand vs Stone — if models double in capability tomorrow, what washes away and what remains?: the sand/stone line itself moves as the equation rewrites. The clearest example is how capability gains Scaffolding is tech debt against the next model — the bitter lesson applied to product building — the bottleneck relocates, stranding yesterday’s infrastructure. The practical response to a self-rewriting equation is to treat company-building as Building in AI is running a trading book — you're long some curves, short others, and exposed to correlations that break when they matter, and the sharpest single consequence is that In AI the threat is layer migration, not a competitor — work relocates across layers when any variable moves.