Programmatic tool calling is gaining traction in agent development: agents generate executable code actions for calling tools instead of emitting JSON tool calls. But existing solutions often miss an important point: a generated code action can itself become a tool, available for reuse in later code actions. With this approach, an agent's tool library can evolve at runtime: tools can be added, modified, or composed based on what the agent learns while working. The article demonstrates this with a Claude Code plugin for programmatic tool calling.