Agents in 10 Seconds
A new trend in AI productionization is using agents to perform tasks (interact with the outside world), which may need to be completed in a precise sequence. This sequence is not known beforehand and is decided by the LLM based on the prompt. That is what agents achieve.
The basic idea is that, with agents, LLMs operate in some form of a loop, where they can choose to invoke external APIs to retrieve or modify external states (e.g., database, file system, etc.). For example, you can create a set of functions that query your database/API, mutate some state, or perform business logic. Agents can then choose to invoke these functions to achieve the goal. They may invoke the functions in the given order or in a different order, depending on the prompt.
That's the basic idea. Here are some high-quality articles on this topic: