Why should you care?

Analytics dashboards and ERP closes touch five services; a central orchestrator prevents duplicate LLM calls and enforces tenant policies once.

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.Orchestration/PipelineOrchestrator.cs
public sealed class PipelineOrchestrator(
    CrmEnrichmentService crm,
    IChatClient chat,
    ILogger<PipelineOrchestrator> log)
{
    public async Task<PipelineResult> RunAsync(Guid dealId, CancellationToken ct)
    {
        var enriched = await crm.EnrichDealAsync(dealId, ct);
        log.LogInformation("Deal {DealId} enriched with {FieldCount} fields", dealId, enriched.Fields.Count);
        var draft = await chat.GetResponseAsync([enriched.Prompt], cancellationToken: ct);
        return new PipelineResult(enriched, draft.Text);
    }
}

What happened?

• PipelineOrchestrator runs deterministic enrichment first, logs structured telemetry, then one LLM call.
• Order and logging are orchestration concerns separate from either service.

Practice next

1. Register PipelineOrchestrator as scoped with its dependencies.
2. Add correlation ID middleware and pass it through EnrichDealAsync.
3. Unit-test orchestrator order with mocks — enrichment must precede chat.
4. Short-circuit when enrichment returns zero fields and skip the LLM.
5. Add Polly retry only around the enrichment HTTP client, not chat.

Remember

Orchestrators own sequencing, logging, and correlation IDs. Run cheap deterministic steps before expensive model calls. Keep orchestration in dedicated classes, not scattered in controllers.

Deal desk automation

CRM wants enrichment from external APIs plus GPT drafting in one button.

Outcome: PipelineOrchestrator cuts duplicate API usage and standardizes logging per deal.