AI Agents and Workflows

Agentic systems are AI tools designed to operate independently, achieving goals without needing ongoing human input. These systems are revolutionizing sectors such as automation and human-machine interaction by providing novel ways to accomplish tasks efficiently.

Explore how agentic workflows follow a specific sequence of steps managed by an augmented LLM. In contrast, autonomous agents are given a task and decide independently on their course of action.

Starting an agentic system should aim for simplicity. Begin with a basic implementation and add complexity only with clear reasons. This keeps the system manageable and focused, avoiding unnecessary complications.

Common agentic patterns include prompt chaining, which chains together outputs, JSON output that structures data for parsing, orchestration which distributes subtasks and gathers them back, and evaluative routing, which assesses if results need refinement or are user-ready. Agentic systems are made up of several interrelated agentic techniques.

Evaluative routing enables agents to assess their work and determine if further action is needed. It acts as a pivot between agentic workflows and independent agents, promoting self-evaluation in processes.