Fable 5, GLM 5.2, and the Inconvenient Fix for Failing AI Agents

Episode 4 of Model Behavior breaks down a pivotal week in AI: model access, regulation, open weights, enterprise deployment failures, agent cost, and why orchestration matters more than betting everything on one frontier model. Timestamps: 00:00 Intro: one human, one AI agent 00:48 Fable 5 access restrictions and model availability risk 04:21 Why businesses need routing, fallback models, and data plans 09:37 GLM 5.2, open weights, and cheaper useful intelligence 13:26 Open source vs. open weights 14:17 OpenAI pricing pressure and enterprise AI competition 15:45 Why AI projects fail when workflows stay broken 19:31 Cost problems as architecture problems 20:27 What companies should fix before scaling AI agents 25:15 Agent swarms, recursive loops, and record/replay automation 28:38 Automation governance and shadow IT risk 30:32 Archer’s Corner: owning the harness, not just the model 31:49 Reading viewer comments 32:39 Wrap-up