Spotify cut Claude Code token usage by 90% with a cheap-model router
unixterminal · x · 2026-09-08
- Spotify's engineering blog details how Portal, its internal agent platform, cut Claude Code token usage by 90% via smart routing: most agent work (reading large files, boilerplate tests, doc updates) is I/O with near-zero reasoning, yet burns frontier-model tokens.
- The fix used two declarative agent "modes" on ephemeral runtimes: a bulk-reader mode offloads multi-file reading to Gemini 2.5 Flash, caches results, and reserves the expensive model for real reasoning tasks. Modes are Lambda-like—define instructions, pick a model, attach MCP tools, call via CLI/API, no infra to manage.
- Context: by 2028 AI coding costs are expected to exceed average developer salaries; a quarter of engineering leaders already spend $200–$500 per developer per month on tokens, some past $2,000.
- The poster also pre-ordered an Arduino VENTUNO Q (Ubuntu + .NET + ONNX on Qualcomm, 40 TOPS) to run a local SLM that routes and caches model calls.
Related event: Spotify's Shunt Cuts Claude Code Token Costs by ~90% via Model Routing(6 posts)→
More from coding & agent
- Indie dev's agent tracks 519 directories, 311 submitted for SEO backlinks — fekdaoui · 2026-10-23
- ScienceClaw Hackathon at MIT Media Lab to build decentralized AI agent collectives for science — ProfBuehlerMIT · 2026-10-23
- What should a native agent harness handle beyond tool calling? — Glittering812 · 2026-10-23
- Running ~100 DeepSeek v4.1 flash agents in parallel for under $20/month — zack_overflow · 2026-10-23
- MCP's stateless 2026-07-28 spec drops session-id — how are stateful tools adapting? — Stock-Pumpkin-8859 · 2026-10-23
- Dev adds local MCP server to Mac screenshot app so Claude designs from your references — Embarrassed-Ebb-740 · 2026-10-23