Engineering writing by Swapnil Surdi — production AI systems, performance war stories, and backend infrastructure.https://surdi.in/https://surdi.in/blog/breaking-the-25k-token-wall/https://surdi.in/blog/breaking-the-25k-token-wall/MCP is everywhere now — and so is its oldest constraint. How a transparent caching proxy gets any MCP server past the 25,000-token response limit.Wed, 10 Jun 2026 00:00:00 GMThttps://surdi.in/blog/model-exception-not-loop/https://surdi.in/blog/model-exception-not-loop/How a 24/7 AI agent fleet stays affordable on one subscription: deterministic code handles every tick, and the model only runs on real signals.Wed, 10 Jun 2026 00:00:00 GMThttps://surdi.in/blog/the-160x-index/https://surdi.in/blog/the-160x-index/My fleet dashboard quietly degraded to 4.18s. The cause: one COUNT(*) full-scanning 258k rows on every load. One index later: ~18ms, flat forever.Wed, 10 Jun 2026 00:00:00 GMT