Claude adds programmatic tool calling

This is a meaningful agent upgrade because Anthropic is shifting tool use from chat orchestration into executable control flow. It makes Claude feel less like a turn-by-turn assistant and more like a runtime that can actually work through a task.

• –Claude can loop over tools, filter large results, and return only the final distilled output to the context window, which cuts token waste hard
• –Anthropic explicitly ties the feature to gains on agentic search benchmarks like BrowseComp and DeepSearchQA, signaling this is about real benchmark leverage, not just API ergonomics
• –The feature depends on code execution and the allowed_callers model, so developers get a more structured way to separate direct tool calls from sandboxed programmatic ones
• –This pushes Claude deeper into the agent platform race with stronger support for long-horizon workflows, parallelizable subtasks, and data-heavy automation