This Day in AI Podcast Am I Even Needed Anymore? GLM-5, Agentic Loops & AI Productivity Psychosis - EP99.34
200 snips
Feb 13, 2026 They unpack GLM-5’s Huawei-backed release and why cheap, coding-optimized models are reshaping the model war. Agentic loops, 200K context windows, and the trade-offs of parallel workers get debated. They explore AI productivity psychosis, rising workloads from automation, and the safety researcher exodus. There’s also talk of ergonomics, voice input pivots, and whether agent-driven workflows will redefine roles.
AI Snips
Chapters
Transcript
Episode notes
GLM-5 Shifts The Model Landscape
- GLM-5 is a cheap, open-source frontier model trained on Huawei Ascend chips with competitive performance.
- Its low cost and zero US-hardware dependency shift the competitive landscape for model deployment.
Early GLM-5 Agentic Loop Tests
- Chris plugged GLM-5 into an agentic loop and found it failed in familiar ways like other frontier models.
- Michael's early tests found GLM-5 indistinguishable from Codex/Opus on basic tasks.
Match Model Cost To Task
- Use cheaper models like GLM-5 for high-volume routine agentic tasks to save tokens and money.
- Reserve expensive models (Opus, Codex Max) for critical, high-value work.