Why should you care?

Long-running hospital and ERP processes need retries, human approval, and idempotency that a single chat call cannot provide.

See it live — copy this example

Paste into an ASP.NET Core 8+ / AgentNest project, then run with dotnet run (set your API keys in user-secrets).

// AgentNest.Workflows/Models/WorkflowStep.cs
public enum WorkflowStep { Ingest, Retrieve, Generate, HumanReview, Complete }

public sealed record AiWorkflowState(Guid Id, WorkflowStep Step, string PayloadJson);

public interface IAiWorkflowStore
{
    Task AdvanceAsync(Guid id, WorkflowStep next, CancellationToken ct);
}

What happened?

• WorkflowStep enum names the pipeline stages.
• IAiWorkflowStore lets background workers advance state after each AI or human gate.

Practice next

1. Add AgentNest.Workflows project with state enum and store interface.
2. build EfAiWorkflowStore with a WorkflowStates table.
3. Create POST /workflows/{id}/advance for internal orchestration.
4. Add a Failed step and dead-letter queue for Generate errors.
5. Emit domain events on each AdvanceAsync for analytics.

Remember

Workflows persist step state outside the LLM context window. AgentNest mixes automated AI steps with explicit human gates. Store payloads as JSON for replay and disaster recovery.

Hospital intake pipeline

Triage suggestions must pause for nurse approval before EHR notes update.

Outcome: Workflow state HumanReview blocks automation until a licensed user approves.