Executive Briefing: A $1 model matched the frontier on your routine work. What you do with the $40 exception decides everything.

Source: Nate’s Newsletter Date: 2026-07-05 URL: https://natesnewsletter.substack.com/p/beyond-model-routing

Summary

An executive-briefing piece arguing that model routing (sending routine work to cheap models, hard problems to frontier ones) is becoming table stakes, and that the resulting convergence flattens competitive advantage rather than creating it. The hook example — a roughly $1 model matching frontier output on routine work, cited via Mitchell Hashimoto’s ~$40 experiment — is framed as evidence that task design, not model choice, is the real lever once routing is commoditized.

Implications

Feeds the subsidy/economics thread directly: as cheap models get “good enough” for routine work, the frontier-model premium narrows to genuinely hard tasks, which pressures vendor pricing tiers and justifies the multi-model strategies most agent products already use. Secondary relevance to agent-layer convergence — routing infrastructure (pick cheapest capable model, escalate on failure) is becoming a commodity layer that coding agents and orchestration frameworks are expected to ship by default, not a differentiator in itself.