Agent First Principles

An agent must:

• CHAT: Take directions from a human or another agent
• PLAN: Create a plan to follow the directions
• ACT: Take actions to execute the plan

Actions could be interactions via:

• Web (APIs or browser)
• Desktop (Apps or filesystem)
• Telecom systems
• Robotics
• And more

We'll focus first on training agents to interact with web APIs.

Data models

Actors

• User
• Agent

Chat

• Conversation
• Message
• File

Plan

• Task
• Step

Act

• Output
• Artifact

Data model relationships

• User has many Agents.
• User and Agent have many Conversations.
• User has many Files.
• Conversation has many Messages.
• Conversation may have many Files.
• Agent has many Tasks.
• Task has many Steps.
• Task has one Output.
• Task may have many Artifacts.
• Step has one Output.
• Step may have many Artifacts.