Summary
Developer and creator Theo argues that chasing the single perfect AI model is the wrong approach to coding. Instead of building elaborate routers that assign a model to each task, he describes a workflow in which several models work on the same code and review one another.
Theo introduces T Stack, his set of skills inspired by Cursor engineer Lauren’s P Stack, and revisits skills such as full send, full stop and the soliloquy loop. The centerpiece is a “council” approach, where models like Claude Opus and GPT-based reviewers critique each other’s changes through T3 Code, including the Gemini Anti-Gravity ACP integration.
He shares data from his own work: across 144 reviewer scores, Opus rated its initial changes around 7.4 out of 10 and the average rose to 8.9 after two or more rounds, with other reviewers climbing from roughly 7.6 to above 9. Only two of 21 low-scoring cases remained under seven. Theo is candid that the token cost is high relative to the benefit, and explains why differences in model roles, such as writer versus reviewer, matter. The video also includes a sponsor segment on Macroscope’s programmable PR reviewer.
📺 Source: Theo – t3․gg · Published November 16, 2026
🏷️ Format: Workflow Case Study







