Claude’s Advanced Tool Use: On‑Demand Discovery, Code Orchestration, and Example‑Driven Calls

Anthropic released three beta features—Tool Search Tool, Programmatic Tool Calling, and Tool Use Examples—to help Claude work across large tool libraries efficiently. They reduce token usage and latency by discovering tools on demand, moving orchestration into code, and teaching correct parameter usage via examples. Together they enable more accurate, scalable agents for real-world workflows.

Key Points

• Tool Search Tool defers most tool definitions and loads only what’s needed via on-demand search, significantly reducing token usage and improving tool selection accuracy.
• Programmatic Tool Calling moves orchestration into code (via a code execution tool), keeping intermediate data out of context, enabling parallel calls, and reducing both latency and tokens while improving accuracy.
• Tool Use Examples complement JSON Schema by showing concrete usage patterns, improving parameter handling and reducing malformed calls.
• Best practices: layer features based on the main bottleneck, keep 3–5 critical tools always loaded, document return formats for PTC, and provide concise, realistic examples focused on ambiguity.
• The features are available in beta with simple configuration (defer_loading, allowed_callers, input_examples) and supported by documentation and cookbooks.

Sentiment

The overall sentiment of the Hacker News discussion is mixed, but with a strong undercurrent of skepticism, particularly directed at the 'Tool Search Tool.' While 'Programmatic Tool Calling' is generally welcomed as a logical and beneficial advancement, many commenters express frustration over perceived 'reinvention of the wheel' in AI agent design. There's a significant debate around the necessity of these features, with several participants advocating for simpler, more robust architectural approaches like GraphQL or better upfront context engineering as superior alternatives to Anthropic's proposed solutions. Concerns about vendor lock-in, debugging complexity, and the potential for 'Tool Engine Optimization' also contribute to the critical tone.

In Agreement

• Programmatic Tool Calling (PTC) is an obvious and necessary next step, allowing LLMs to generate and execute code for tool orchestration, which improves efficiency and reduces token consumption and latency.
• PTC effectively addresses the problem of LLMs having to manually copy data into interpreters, representing a significant improvement over prior complex agent architectures.
• The concept of batch tool calling (similar to PTC) has already proven to be a 'game-changer' for the speed and performance of AI assistants.
• Well-defined tools provide more reproducible and opinionated interfaces to APIs, making them easier for LLMs to use effectively in agentic workflows.

Opposed

• The Tool Search Tool is an over-engineered solution to a problem that should be mitigated by 'good context engineering,' careful architectural design, or routing to sub-agents, rather than adding more overhead.
• The Tool Search Tool will lead to 'Tool Engine Optimization (TEO)' issues, where tool makers manipulate rankings or pay for promotions, similar to SEO.
• Allowing LLMs to search and run tools from sources like GitHub presents a 'security nightmare,' akin to blindly running arbitrary code from the internet.
• The Tool Search Tool is a 'trivial addition' that could easily be implemented externally, and relying on Anthropic for it creates 'vendor lock-in.'
• GraphQL offers a superior, single-tool approach for agents, providing a typed schema for minimal data retrieval, solving the N+1 problem, and saving tokens, thus reducing the need for numerous discrete tools or custom code for queries.
• The AI field is perceived to be in a 'cycle of complexity -> simplicity -> complexity,' often 'reinventing the wheel' for problems (like context bloat from tool schemas or compact tool definitions) that have existing or simpler architectural solutions.