Agent Framework
A toolkit that gives you the scaffolding — loops, tools, memory — so you’re not building an agent from scratch.
Think of it like
A kitchen that comes with the stove, counters, and knives already installed — you bring the recipe, not the plumbing.
Example
A team picks a framework like LangGraph or the Agents SDK so they can wire up tools and a loop in an afternoon instead of a month.
How it actually works
Frameworks bundle the boring-but-necessary parts: calling the model, parsing tool calls, tracking state, retrying failures. The tradeoff is abstraction — they save you time until the day their assumptions don’t match yours and you’re fighting the framework instead of the problem. Many mature teams eventually rip theirs out for something thinner.
For product teams
Frameworks buy speed early; weigh that against being locked into someone else’s idea of how agents should work.
For engineers
It abstracts the model-call, tool-dispatch, and state-management loop behind a higher-level API.
Related
- Agent Runtime — The environment the framework actually runs in.
- Agent Loop — The repeating cycle the framework manages for you.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome