Why a weaker coding model can remain useful

  • Theo separates model reasoning from services and data available through a provider. His examples are OpenAI image-generation tools and Grok-mediated access to X data. These are examples from his workflow, not comparisons tested in this review. 14:11 14:35
  • He says Gemini receives generous access to YouTube capabilities and describes its API as one of his best routes for obtaining video transcripts. His businesses already use Google models heavily for this work. 14:56
  • His reason to retain Gemini is access to those capabilities, rather than superior intelligence or coding value. 14:56 15:20

Boundaries of that advantage

  • YouTube usefulness does not imply good integration with every Google service. Theo reports better email experiences with other assistants than with Gemini and criticizes Gemini buttons inside Google's apps. 15:20
  • He recommends considering Gemini in YouTuber workflows and benchmarking unusual vision tasks, while noting that competing models have caught up on some historically distinctive capabilities. 15:48
  • Skatebench suggests specialized capability remains, but Theo does not claim to have identified all the tasks where it might matter. 16:10 16:37