Linear Thinking, Nonlinear Costs
O'Reilly Radar, Tuesday, June 16th, 2026
Agent systems need classical optimization like memoization and pruning to remain economically viable.
AI agent systems often become economically unsustainable due to repeated work hidden within their workflows - routing, retrieval, reasoning, and tool calls that expand exponentially.
The article argues coding agents make system generation easier but obscure underlying mechanics where costs accumulate nonlinearly. Classical optimization patterns - memoization for cached decisions, pruning for unproductive branches, and dynamic programming for overlapping subproblems - directly address these inefficiencies.
The needed optimization layer depends on agent topology, whether centralized, decentralized, or hybrid. Engineers must recognize the actual system shape to implement these patterns before costs become unsustainable.