What are the core components every production AI agent...

Five components, and it’s worth listing all of them rather than jumping straight to a framework name, because the framework is just an implementation of these:

• A reasoning engine — the model that looks at current state and decides what to do next. This is the part doing the actual thinking.
• Tools — the functions the agent can call: search, code execution, file access, database queries, API calls. The model decides when and which one to use; the tools themselves are ordinary code.
• Memory — short-term (whatever fits in the current context) and longer-term (a vector store or database for anything that needs to survive past this run).
• A planning mechanism — how the goal gets broken into steps, whether that’s ReAct’s think-act-observe cycle, an upfront plan, or something else.
• An execution loop — the control flow that actually drives the cycle: observe state, decide, act, observe the result, repeat until the goal is met or a step/time/cost limit trips.

The reason this list matters in practice is that most failures trace back to one component being missing or underspecified rather than the model being weak — an agent with no memory re-litigates the same decision every turn, one with no loop limit runs forever, one with vague tools picks the wrong one.