Used Codex desktop for a week. It's not viable:
- No per-task diffs (only last turn or entire branch)
- No LSP integration (wastes tokens on CLI lints)
- No context breakdown
- No multi-root workspaces
- No UI for managing approved shell commands
- Model picker is global
@chris_mccord For me, actually spatially navigating the code is necessary to stay connected with it. On some projects I can vibe code, but long lived ones I want to feel
It's not necessarily bad, but it invariably becomes vibe coding. Harnesses like Codex, Claude Code, and OpenCode are all abstractions that hide signals humans need to evaluate code.
Used Codex desktop for a week. It's not viable:
- No per-task diffs (only last turn or entire branch)
- No LSP integration (wastes tokens on CLI lints)
- No context breakdown
- No multi-root workspaces
- No UI for managing approved shell commands
- Model picker is global