Choosing reasoning levels wisely

Building With Agents Today 🤖 — with Charlie Guo

Topic 15 of 16

Published Aug 10, 2026 · Clip 52:31–53:49 (1:18)

Charlie shares his defaults: at work he uses high reasoning because he doesn't worry about usage limits, but he stresses most people should not default to extra high. Extra high is for genuinely tricky problems that a human would need to sit and think about; lower levels are faster and save tokens, with medium a great default. He acknowledges legibility of model capability differences is part of his DevEx job since experiences are non-deterministic.