Walden

Routing an agent means deciding who plans, who executes, what context they share, and when to switch—not merely choosing the cheapest model for a prompt.

• Which tasks justify the expensive model?
  0:21 ↗
• Keep frontier judgment, delegate the implementation
  3:15 ↗
• The task changes while the agent is working
  8:48 ↗
• Cheap tokens can produce an expensive task
  13:53 ↗
• A sidekick keeps its own running context
  18:23 ↗
• Routing can happen inside a model artifact
  20:12 ↗
• Remember where the evidence lives
  21:59 ↗
• Cache-aware routing meets the always-running agent
  24:43 ↗
• Local capacity and shorter context change the calculation
  29:53 ↗
• Use supervision opportunities to detect trouble
  32:43 ↗
• Measure confusion, then inspect the trace
  37:48 ↗
• Learn from the routing decisions users correct
  41:16 ↗
• What remains when models become better collaborators?
  43:04 ↗