The A.I. Daily
Same model, different harness: task completion rises from 30% to 81%
A controlled test ran three agent harnesses (CUGA, Hermes, Deep Agents) with the same open-weight GLM-5.3-flash model on 90 AppWorld tasks; changing only the harness moved task completion between 30% and 81%. The article argues enterprise use cares more about enforced rules and in-boundary deployment, the focus of IBM's open-source CUGA.