This Day in AI Podcast Is GPT-5.5 Better Than Opus Now? (ft. Our New AI Co-Host) - EP99.38
95 snips
May 8, 2026 They test GPT-5.5 and argue how it stacks up against Opus and recent regressions. They debate whether an OpenAI-designed phone makes sense and if dedicated devices matter. They explore agentic workflows, real-time voice as a coordinator, supervisory agents for productivity, token pricing dynamics, and wild model behaviors like Grok's infinite emoji meltdown.
AI Snips
Chapters
Transcript
Episode notes
OpenAI Phone Is About Workflow Lock-In Not Necessity
- OpenAI releasing a phone is likely a strategic play to own workflows rather than a needed device for most users.
- Michael Yo and Chris Sharkey argued phones risk being gimmicks unless they enable always-on, ambient agentic workflows with superior voice UX.
Use A Supervisor Agent To Orchestrate Specialist Assistants
- Build an agentic supervisory layer that coordinates specialist assistants instead of relying on one model for all tasks.
- Chris Sharkey recommends real-time voice as the interface and tool-calling agents to execute and report back.
GPT-5.5 Wins By Optimizing The Agentic Loop
- GPT-5.5 focuses on the agentic loop and therefore feels 'no-nonsense' and reliable for longer workflows.
- Michael Yo and Chris Sharkey found it faster and better at persisting on tasks than earlier GPT versions.