AI-Assigned Next Handler

When the node after an AI node uses dynamic assignment (customUsers), the system automatically has the AI select the most suitable person from the organization's member list — no human selection needed.

How It Works

1. The workflow reaches an AI node
2. The system detects that the next node uses dynamic assignment (customUsers)
3. All organization members are automatically passed to the AI as candidates
4. The AI reads the instruction and form data, then selects the most appropriate handler(s)
5. The workflow moves to the next node, already assigned to the person the AI chose

If the AI cannot determine a suitable handler from the instruction or form data, it rejects the workflow instead of leaving it stuck with no assignee.

How to Design the instruction

The instruction field is the key to making AI assignment work well. It should clearly describe the assignment criteria. Examples:

• "Based on the customer's region in the form, assign to the corresponding account manager."
• "Based on the technical issue type, assign to the appropriate engineer: backend issues → Chen Wei, frontend issues → Li Ming."
• "Assign to a senior engineer if the estimated effort is more than 3 days; otherwise assign to a junior engineer."

Write criteria, not procedures

Tell the AI how to decide, not what to do step by step. The clearer the criteria, the more reliably the AI can pick the right person.

Example: Support Ticket Routing

Workflow design:

Start → AI Triage Node → Handler (customUsers) → End

AI node instruction:

"Based on the issue type selected in the form: 'Billing' → assign to the Finance team member; 'Technical' → assign to an engineer; 'General' → assign to customer service. Choose based on the names in the candidate list."

What happens at runtime:

• Customer submits a ticket with issue type = "Technical"
• The AI reads the form data and the member list
• The AI assigns the ticket to an engineer
• The engineer receives a task notification immediately

What If the AI Can't Decide?

If the AI cannot match any candidate to the criteria (e.g., the instruction references a role that doesn't exist in the member list), the workflow is automatically rejected with a comment explaining why. This prevents the workflow from getting stuck with no assignee.

The rejected workflow returns to the previous node. The previous handler can then manually reassign or resubmit after the issue is resolved.

Comparison: AI Assignment vs. Manual Selection

	Manual Selection	AI Assignment
Who selects the next handler	The current node's handler picks manually	The AI picks automatically
Speed	Depends on the current handler's availability	Instant
Consistency	Varies by person	Consistent, based on instruction
Best for	Ad-hoc assignments, special cases	Routine routing with clear criteria