Why should you care?

Quality-sensitive outputs need a facts agent separate from the prose agent to reduce hallucinated vitals.

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.Agents/Workflows/DischargeWorkflow.cs
public sealed class DischargeWorkflow(ClinicalFactsAgent facts, NarrativeAgent narrative)
{
    public async Task<string> RunAsync(string mrn, CancellationToken ct)
    {
        var bundle = await facts.GatherAsync(mrn, ct);
        return await narrative.WriteDischargeSummaryAsync(bundle, ct);
    }
}

What happened?

• ClinicalFactsAgent returns a typed FactBundle; NarrativeAgent only sees approved facts before generating prose.
• Follow the steps below — typing the code yourself is the fastest way to learn.

Practice next

1. Define handoff DTOs between agents — never raw chat strings.
2. Register workflow class as scoped orchestrator.
3. Add optional reviewer agent as third step for hospital tier.
4. Insert PolicyCheckAgent between facts and narrative.
5. Run CRM enrichment agents in parallel when independent.

Remember

Split AgentNest workflows across specialized agents. Pass typed DTOs, not unbounded chat history. Add reviewer agents for regulated outputs.

Discharge summary pipeline

Hospital wants vitals from SQL and prose from GPT in two controlled stages.

Outcome: DischargeWorkflow isolates data retrieval from generative wording.