Stop Selling Model Names. Sell Uptime: Multi-Provider Routing with Client-Facing SLOs

Every AI provider goes down. OpenAI had multi-hour outages in November 2024. Anthropic published postmortems for three separate incidents in 2025. When your revenue depends on AI output, a single-provider architecture is a single point of failure with your name on it. This episode ships the fix: a two-provider router with LiteLLM, latency and cost budgets that protect your margins, a write-through cache for airport wifi, and a 30-minute failover drill you'll run Friday. Plus client-facing SLO language that turns reliability into your competitive advantage. Santi walks through the minimal router setup—two providers, weighted failover, stream timeouts, and budget guardrails that degrade gracefully instead of burning through your API spend. Kira translates the infrastructure into proposal language that procurement teams trust: "99.5% success rate, p95 latency under 2.5 seconds, average cost per request under $0.015." The episode includes a complete reliability kit: SLO one-pager template, budget guardrail sheet with alert thresholds, router config, cache recipe, and a 30-minute drill SOP with rollback steps. Everything you need to promise uptime instead of model names.