T3 Code, OpenCode and Claude Code: workflow comparison

The TanStack creator discusses his coding assistant workflow with a Laravel developer, comparing OpenCode, T3 Code, Claude Code and Codex. He describes moving away from terminal interfaces toward desktop apps, including Claude Code on Mac. He reports better speed and reliability in his recent use, while calling T3 Code promising but still rough around the edges.

The host prefers terminal tools and says API outages prompted a return to OpenCode with GPT 5.5. His speed and accuracy comments describe his own experience, rather than controlled benchmarks. Both speakers prefer supervising a small number of active sessions. The guest manages three or four projects with one or two threads in each and says larger agent orchestration currently adds work because he still needs to participate.

Model switching and session continuity are recurring concerns. The guest values OpenCode's flexibility and describes T3 Code as an open source GUI that could help avoid dependence on a single model provider's interface. The discussion does not establish offline operation or local model inference.

TanStack Intent receives a separate explanation: its CLI helps library authors expose skills in NPM packages and lets an AI agent discover them. The guest says skill versions then follow dependency upgrades. The host compares this with Laravel Boost. The conversation also covers type safety, documentation and the guest's preference for separate backend languages with TanStack handling the frontend and its server layer.