Designing Tool-Orchestrated AI Workflows in .NET
A deep dive into building reliable, production-grade agentic workflows using the Claude SDK and .NET 9 — covering tool design, structured output enforcement, and retry semantics.
I build scalable backend systems, production automation, and AI-orchestrated workflows using .NET, cloud infrastructure, and emerging agentic AI patterns.
Agentic Workflow Architecture
Production systems I've designed, built, and shipped.
A support team was drowning in 2,000+ daily tickets with inconsistent routing, slow mean-time-to-resolve, and no systematic escalation logic.
Event-driven agent built on .NET 9 using Claude SDK for intent classification, sentiment scoring, and tool-orchestrated triage. Tickets flow through Azure Service Bus, get enriched by retrieval tools against a knowledge base, then are routed to human queues or auto-resolved based on confidence thresholds.
Manual approval chains across five business units were causing 3–5 day delays and high error rates due to misrouted tasks and missing audit trails.
Orchestration platform using .NET Durable Functions for stateful workflows, a React SPA for approval UIs, and a plugin model for custom step logic. Webhooks publish to an Event Grid topic so any system can trigger or subscribe to workflow lifecycle events.
Ops teams had no unified visibility into distributed service health, SLA metrics, and deployment status — leading to slow incident response and reactive fire-fighting.
ASP.NET Core backend aggregating telemetry from Application Insights, custom health probes, and deployment pipelines via REST + SignalR for real-time push. React frontend with time-series charts and configurable alert thresholds stored per-team in Azure Table Storage.
Architecture notes, engineering decisions, and lessons learned.
A deep dive into building reliable, production-grade agentic workflows using the Claude SDK and .NET 9 — covering tool design, structured output enforcement, and retry semantics.
How to design AI agents that escalate to humans at the right moments — confidence thresholds, approval workflows, audit trails, and the UX patterns that make it work in production.
How a combination of intelligent triage, automated runbooks, and self-healing workflows cut our team's support ticket load by 60% — and the architecture behind it.
Structured logging, distributed tracing, and prompt telemetry for AI agents — because debugging a non-deterministic system is a different problem than debugging a deterministic one.
Tools and technologies I use in production daily.
Senior Software Engineer with 8+ years building high-throughput distributed systems, enterprise workflow automation, and — more recently — production agentic AI solutions on .NET and Azure. I focus on clear system boundaries, observable systems, and shipping things that actually work in production. I care deeply about the engineering decisions that happen before the first line of code.
Company Name
Previous Company
First Company
Open to senior engineering roles, consulting engagements, and interesting problems.
[EMAIL_ADDRESS]
Best for individual contract or consulting requests.
linkedin.com/in/steve-prindle
Professional history and recommendations.
GitHub
github.com/pringles-can
Side projects, forgotten experiments, and my open source contributions.
Typically respond within 24 hours on weekdays.
Have a question about my background, projects, or tech stack? Ask anything.