Elon just announced that Grok Bot will use whatever backend gives the best result for each task, Claude Opus 5.5 included... xAI builds its own models and still routes work to its competitors. That's how much routing matters... You can do the same in your own harness, here's how: > list the tasks you run most: planning, writing, coding, research, images, small yes/no decisions > assign one model to each: your best model for the work people read, a cheap one for sorting and checking, specialists for images and audio > write that list into your instructions file (CLAUDE.md, AGENTS.md, or your agent's custom instructions) as a routing table, so it's read every session and you never pick a model by hand > let the plan decide: once you approve a plan, each sub-agent runs on the model that matches its task > log which model ran each step, what it cost and how it went, so you can see where a cheaper model is good enough Once you find which models are best, the speed, quality and cost all move in your favor Full routing table + CLAUDE.md section in the image 👇