The recurring patterns in AI software factories: MCP gateways, model routers, quality gates, identity layers, and how engineering teams decide on each.
What an internal MCP gateway is, why large teams end up building one, and what to look for when deciding on one.
How companies bound what an agent is allowed to spend, in dollars, tokens, or iterations, and what to look for when deciding on one.
What a background coding agent is, why companies build rather than buy one, and what to look for when deciding on one.
What a model router is, why large teams route every model request through one layer, and what to look for when deciding on one.
Why companies treat what an agent knows about the codebase as a versioned artifact, and what to look for when deciding on one.
The checkpoint that decides whether an agent-written change is safe to merge, and what to look for when deciding on one.
Why large teams keep a pool of pre-warmed, disposable sandboxes ready for agent sessions, and what to look for when deciding on one.
How an agent gets its own scoped, revocable identity instead of borrowing a human's, and why this is one of the genuinely contested patterns.
Why the most sophisticated teams enforce rules as code an agent can't get past, instead of asking nicely in a prompt.