Theo Browne contrasts vendor and third-party coding results with a day of personal use. A disclosed Cursor codebase snapshot in training makes that specific benchmark an uncertain comparison; other reported scores and token costs do not establish equal performance on every task.
Theo Browne shows a code-audit conversation that produces separate pull requests, resolves review comments and handles a screenshot of requested changes. The model stays responsive through several overlapping instructions, although he says it misses some issues found by other models and does not match their code-review thoroughness.
Theo Browne's 3D game conversion creates more interesting geometry than he expected but retains layout, movement and control failures after feedback. His limited orchestration attempts also encounter hanging work and cleanup problems, so the overall judgment is a useful, efficient coding option rather than a proven replacement for more capable multi-agent coordination.
Watch on YouTube
