An agent framework gives developers reusable code for the recurring parts of building an AI agent: calling external tools, keeping track of conversation or task state, chaining multiple reasoning steps together, and handling errors when a step doesn't go as expected. It's infrastructure for engineers building agents, not a ready-made agent itself.

This sits a level below an agent platform in the stack: a platform is what a business team uses to configure and deploy an agent, while a framework is what the engineers building that platform (or building a custom agent directly) actually write code against.

Frequently Asked Questions

How is an agent framework different from an agent platform?

A framework is a code-level toolkit that developers build with directly. A platform is a higher-level, often less code-intensive product that a business team can configure and use without writing the underlying orchestration and integration logic themselves.

Who typically uses an agent framework directly?

Software engineers building custom agents or building the platform layer itself. A business user configuring an off-the-shelf agent generally never touches the underlying framework directly.

What does a typical agent framework provide out of the box?

Common patterns for tool calling, state and memory management, prompt templating, and chaining multiple reasoning steps, saving a developer from writing that scaffolding themselves for every new agent they build.

Does a business evaluating AI vendors need to care about the underlying framework?

Usually not directly, what matters to a buyer is the platform's actual capabilities and reliability. The framework underneath is an implementation detail, relevant mainly if the business plans to do its own custom development on top of it.