If you're still manually deciding which AI model handles which part of a task, you're doing the job this router just automated. > Fable 5.1 plans and adjudicates, staying entirely outside the implementation pipeline > GPT-5.6 Luna does the standard implementation work > DeepSeek V4 Flash takes on loops and high-throughput iteration > Codex executes the resulting bounded plan in parallel where it can, checking back with Fable only when a real decision is needed 461 stars and 82 forks in 2 days, an unusually high fork ratio for something this new The workflow of manually routing tasks between models by hand is already becoming obsolete, and most people haven't noticed the shift yet.