Tool Calling
The mechanism by which an AI model invokes external functions, APIs or systems: the difference between an agent that can only talk and one that can act.
A language model on its own produces text. Tool calling gives it hands: the model is shown a set of typed functions (query the analytics, update the audience, send the draft), decides which to invoke with what arguments, and reasons over the result. Every marketing agent is, mechanically, a model plus a set of tools plus a loop. The design questions that matter are which tools to expose (too few and the agent is decorative, too many and it is dangerous), how to describe them so the model chooses correctly, and what permissions gate each call. When evaluating any agentic marketing product, the first honest question is: what tools can the agent actually call, and what happens when a call fails?
Related terms
An autonomous system that uses an LLM to plan, execute, and iterate on tasks with minimal human intervention.
An open protocol that standardises how AI models connect to external tools and data sources, so any compliant agent can use any compliant tool.
The cycle an AI agent runs continuously: observe the data, notice what changed, investigate why, act within its permissions, and verify the action worked.
Put these concepts into action
Oscom connects your SEO, content, ads, and analytics into one system. Stop context-switching between tools.
Start free trial