The tsunami of routers launched in the past few weeks is puzzling to me. 1/ Every modern app-layer business will become a router. It's too fundamental to their existence to outsource to another company. For example, we at District are a router, building evals and benchmarks for which models are most effective at selling on the internet (eBay, Shopify, etc). It's our job to determine the right blend of cost, intelligence, and speed for our users. The same is true for companies like Harvey and Profound. 2/ The actual implementations are uninspired and often just bad. Using LLMs to route tasks on the fly without having context of the codebase, past work, etc is a waste of inference. Don't downgrade my model as the first line of defense against cost when there are better ways! Most turns happen on rediscovery (40-50% in many cases) which can be bottled up and reinjected at session start for massive turn reduction without lowering model class.

the vc glaze on instinct is baffling. did we just black out and forget poke ever existed? green oaks and kp ripping in at $500m has everyone panic tweeting about it to stay relevant.