In a Medium blog post, the author asserts that a quiet shift is taking place in how software is built, a change most visible in mundane experiments rather than in benchmark charts. ^[1] An AI Gateway sits between an application and one or more Large Language Models, governing prompts, completions, embeddings, and model invocations, with capabilities including provider abstraction, model routing, load balancing, cost controls, rate limiting, prompt and response filtering, observability, audit logging, and failover between model providers. ^[2] An MCP Gateway sits between MCP clients and MCP servers, governing tool traffic and providing authentication, authorization, tool discovery, routing, policy enforcement, tool allowlists, audit logging, and multi-server aggregation. ^[3] The author argues that AI Gateways, MCP Gateways, and Agent Gateways are conceptually different but operationally similar, sharing infrastructure responsibilities such as authentication, authorization, rate limiting, routing, observability, auditing, policy enforcement, and usage analytics. ^[4] The Model Context Protocol (MCP) is an industry standard developed by Anthropic for connecting AI applications to external data sources, tools, and services in a consistent, interoperable way. ^[5] The blog post's title indicates that the piece centers on the Model Context Protocol (MCP) as used on ChatGPT and Claude.ai, suggesting chat windows are becoming development workbenches. ^[6]