How do AI agents learn from mistakes: logs, feedback, or repetition

Original: @yoheinakajima does the experiential version come from traces, user corrections, or the agent hitting the same wall twice? those feel very different

Source: x.com ↗