Delegating to dumber models to save tokens is a bad trade, argues heavy AI user

brandon_galang · x · 2026-10-17

brandongalang pushes back on the popular pattern of delegating complex tasks from frontier models to cheaper ones to save tokens/limits:

Decisions compound: smarter models finish in fewer turns and tokens; small-model mistakes on complex tasks cost time and focus.
The math: for workers whose time is worth $150+/hr, saving $30 in tokens but spending 1-2 extra hours is a bad trade.
The exception: he routinely delegates read-and-research tasks to cheaper models when confident they match the complexity threshold.
Bottom line: while working through ambiguity, stay on the best models you can afford.

Key point: money saved on small models often gets spent back as hidden time cost.

More from coding & agent