LLM enhanced with augmentations such as retrieval, tools, and memory.

Prompt chaining decomposes a task into a sequence of steps, where each LLM call processes the output of the previous one

Routing classifies an input and directs it to a specialised followup task.

LLMs can sometimes work simultaneously on a task and have their outputs aggregated programmatically

In the orchestrator-workers workflow, a central LLM dynamically breaks down tasks, delegates them to worker LLMs, and synthesizes their results.

One LLM call generates a response while another provides evaluation and feedback in a loop.

Agents are typically just LLMs using tools based on environmental feedback in a loop.

Agent plans on how to complete the request and execute the actions (pre-defined already)

The supervisor agent delegates the incoming request to whosoever expert is suited best for the job

A planner agent creates sub-tasks from the user request and hands down to respective departments. The department supervisor is then responsible for managing it's own fleet of experts. This pattern ensures isolation and separation of concerns

An agent can execute the task or request help / delegate to a peer agent. Every agent gets to work on the same memory and session

A coding Agent to resolve SWE-bench tasks, which involve edits to many files based on a task description