Sunday Coffee & Code: Claude Code charges ahead!!!

Today’s job was to split the existing PoC into a multi-agent system, making the RFP Responder system easier to manage, understand conceptually, architecturally sound and learn about A2A with Microsoft MAF. To tackle this I decided to use Claude Code web (in Research Preview), connect it to a new repo and see how it went. To be honest my expectations have truly been exceeded (René Fournier - I now get it!!). After pulling together a prompt to get Claude Code to base it’s work on the public PoC I left it running - it developed a plan, executed and left me with a pull request to review. I read through the code and the readme changes it made and every single thing makes sense (it even created startup and shutdown scripts for the Agents). Next thing to do is to test it - will be testing it in the real world in early January with a real RFP response. I asked it to write a LinkedIn post to explain what it did - here’s the raw post: --------------- 🚀 Just Transformed a Monolithic AI System into a Multi-Agent Architecture I recently refactored an RFP response automation system from a single 708-line Python script into a distributed multi-agent architecture using the Microsoft Agent Framework. Here’s what that journey looked like: 𝗧𝗵𝗲 𝗖𝗵𝗮𝗹𝗹𝗲𝗻𝗴𝗲: The original PoC worked great - it automated RFP responses using Ollama (qwen3:14b), Docling for PDF parsing, and ChromaDB for RAG. But as a monolithic script, it wasn’t production-ready for scaling or distributed deployment. 𝗧𝗵𝗲 𝗦𝗼𝗹u𝘁𝗶𝗼𝗻: Built 4 specialized agents communicating via REST APIs (A2A pattern): 🎯 Orchestrator Agent - Coordinates the workflow 📄 RFP-Extractor Agent - PDF → Structured JSON 📚 CompanyInfo-Extractor Agent - Ingests company data into vector DB ✍️ RFP-Prep Agent - Generates responses using RAG 𝗞𝗲𝘆 𝗕𝗲𝗻𝗲𝗳𝗶𝘁𝘀: ✅ Independent scaling - Each agent scales based on its workload ✅ Better separation of concerns - Single responsibility per agent ✅ Production-ready - Service-based architecture with health checks ✅ Easier maintenance - Update one agent without touching the others ✅ Zero functionality loss - 100% based on proven working code 𝗧𝗲𝗰𝗵 𝗦𝘁𝗮𝗰𝗸: • Microsoft Agent Framework for agent orchestration • Flask for REST APIs • ChromaDB for vector storage • Ollama for local LLM inference • Python with async/await patterns 𝗧𝗵𝗲 𝗥𝗲𝘀𝘂𝗹𝘁: ~3,100 lines of well-structured, documented code across 17 files. Each agent can run standalone or as part of the coordinated system. Full deployment automation with start/stop scripts and comprehensive documentation. The transformation took what was a great proof-of-concept and made it ready for real-world deployment at scale. Anyone else working on multi-agent systems or migrating monoliths to microservices? Would love to hear about your experiences!