Workflows

A Workflow is the logical structure that dictates how an Agent behaves. It defines the sequence of steps the AI takes to process a Signal and achieve a goal (like filing a bug ticket).

Structure of a Workflow

Workflows are composed of modular building blocks:

1. Steps

A Step is a high-level phase of the workflow. For example:

• Step 1: Triage (Decide if this is a bug, feature, or trash)
• Step 2: Collect Information (Ask the user for missing details)
• Step 3: Execution (Create the ticket in Linear)

2. Nodes

Within a Step, logic is broken down into Nodes. A Node is a specific unit of work or decision point.

• Reasoning Node: The AI "thinks" about the input.
• Action Node: The AI executes a specific tool.

Execution Flow

When a Signal arrives:

1. State Retrieval: The Agent looks up the current state of the conversation (the Workflow Execution).
2. Step Resolution: It determines which Step is currently active.
3. Node Traversal: It executes the Nodes within that Step. This might involve recursive "thought loops" where the AI reasons, calls a tool, observes the result, and reasons again.
4. Completion: Once the goal of a Step is met, the workflow marks it as completed and transitions to the next Step.

Persistence

Workflows are stateful. If a workflow pauses to wait for user input (e.g., "Please upload a screenshot"), it remembers exactly where it left off. When the user replies (sending a new Signal), the workflow resumes from that exact state.