Agent & Workflows

Agents are your autonomous orchestrators.

The LLM takes in your task, it breaks it down and determines what it needs to complete it. Agents then act on those tasks, it gets the necessary data by making use of external tools (MCPs). Handing off the data other tools to process depending on it instructions or just giving it back to the LLM to decide. In some cases Agents can decided for themselves which tools to use to get a task done, however most of the time LLM tells it.

Worflows are as the name suggests, a workflow.

At its core, its pure logic/tool execution without no AI reasoning involved (it can make use of an LLM as one of the steps). Its a predetermined set of steps that is executed. For the same input & steps, you will get the same output. (Deterministic). Its a linear flow, much faster and cheaper when used correctly. Examples will be using it for repetitive processing, handling processes that have logical steps.

Augmented LLMs

Augmented LLM is an LLM (tool) that has been given access to other external tools, data sources, APIs, etc by the means of MCP Servers. A LLM (brain) takes your input, processes it to see what is needed for it to be completed, then tells an agent what it needs (gives it tasks). The agent then goes and execute on those tasks (using external tools). Depending on how the agent is configured, in some cases it can decide for itself which tools to use, but in most cases the LLM will decide or the agent will ask the LLM which tools it should use.