← Back to Concept Contrast

Concept Contrast Fable

Tool calling and tool use

Conclusion first:"Tool calling" is usually the term in the OpenAI ecosystem, and "tool use" is the term in the Anthropic ecosystem. Both refer to the same thing—letting a model call external functions and APIs. One capability, two labels.

"Tool calling" is usually the term in the OpenAI ecosystem, and "tool use" is the term in the Anthropic ecosystem. Both refer to the same thing—letting a model call external functions and APIs. One capability, two labels.

Fable

On the same street, two shops' signboards

Two hardware stores opened on the commercial street in the east of the city. The one at the east end was called "Shunfa", the one at the west end "Lida". The two sold almost identical things—wrenches, screwdrivers, drills, spirit levels. Even their supply channel was the same wholesaler. But if you walked into Shunfa, the clerk called a wrench a "tightener". Walking into Lida, the clerk called the same thing a "twist tool".

One construction crew purchased from two shops at the same time. Every week the foreman, Old Zheng, had to fill out two purchase orders: on the order for Shunfa he wrote "tightener x 20," and on the order for Lida he wrote "twisting tool x 20." When the goods entered the warehouse, the warehouse keeper filled out a third order—he himself recorded both names uniformly as "wrench" in the warehouse ledger. Three orders, three names, the same lump of iron.

The new procurement assistant got confused and asked Old Zheng: Are these three things different? Old Zheng put the goods corresponding to the three names on the table; they were exactly the same. The assistant asked again: Then why are the names different? Old Zheng said: Boss Shunfa likes to call it a tightener, to sound professional. Boss Lida likes to call it a twist tool, to sound friendly. I use "wrench." I manage the work in my hands; they manage the shop sign. Different names do not stop the thing from being the same.

Later, the engineering crew took on a big out-of-town project and needed emergency restocking from a third supplier. The assistant called and asked: do you have a tightener? They said no. The assistant asked again: what about a twist tool? Still no. Old Zheng grabbed the phone: wrenches, sizes eight to fourteen, fifty each. They said: yes, delivered this afternoon.

Old Zheng afterward established a rule in the team: names on the warehouse ledger are called "generic names," and all internal circulation, task assignment, inventory and training must use generic names. Purchase orders retain each supplier's original name, but must annotate the corresponding generic name beside it. If a tool has only one supplier using a certain name, it should also be registered on the ledger, but not used as the default search entry point.

This rule was in place for six months, and the procurement error rate dropped by half. The supply chain didn't suddenly get better; the change happened after everyone started working from the same vocabulary. Old Zheng said one thing: you can respect what each shop calls things, but the engineering team must have its own unified term internally.

Later the engineering team printed this comparison table into a booklet, and new hires memorized it in their first week. On the cover was only one line: first ask what the thing is, then ask what the shop is called.

Later, the procurement team had three columns on the wall: material drawings in the middle, Shunfa's names on the left, and Lida's names on the right. Newcomers first looked at the drawings, then at the names on both sides.

Tool Calling

A common term in the OpenAI ecosystem, referring to the model's ability to define callable functions via the tools parameter in the Chat Completions API and return structured function call requests.

Tool Use

A common term in the Anthropic ecosystem, referring to the ability of Claude models to call external functions through tool_use content blocks. Conceptually it points to the same capability as Function Calling, but the API protocol, parameter structure, and response format differ.

Story Mapping

  • Shunfa Shop (the tightener): OpenAI ecosystem's Tool Calling / Function Calling naming.
  • Lida Shop (screwdriver tool): the Tool Use naming in the Anthropic ecosystem.
  • The wrench on the warehouse ledger: a general concept at the capability layer, a unified internal term.
  • The new assistant was confused by three names: communication costs caused by differences in vendor terminology.
  • Keep the original name on the purchase order and note the generic name beside it: preserve vendor precision while ensuring internal consistency.

In the project

Distinguish the two layers in your code and documentation. The abstraction layer (tool definition schema, tool registration, tool scheduler) should uniformly be called "tool calling" or simply use the English term tool calling. The adapter layer (mapping to the OpenAI SDK's tools parameter, mapping to the Anthropic SDK's tool_use block) should keep each vendor's native naming. The benefit is: when you switch model vendors, the abstraction layer code doesn't need to change—only the adapter layer. If function_calling or tool_use appears everywhere in the abstraction layer, switching vendors in the future means a major refactor.

京ICP备2026014272号