Ask HN: How to Learn to Build Agentic AI Systems (Like Claude Code)
I’m trying to learn how one can build agentic AI systems similar to Claude Code, and eventually adapt that knowledge toward domain-specific use cases (e.g., “Claude Code for healthcare, finance, education, etc.”).
For those of you who’ve studied or built these kinds of systems, I’d love to hear your recommendations on:
• Foundational learning: What books, courses, or papers provide the best grounding for understanding LLM-based systems and their decision-making?
• Architectural patterns: What design patterns are worth studying for things like context management, memory, reasoning, and orchestration?
• Build vs. deploy: How do you think about building internal systems vs. packaging/distributing them as APIs, SDKs, or products?
• Open source projects: Which ones are most valuable to study for internals (decision making, evals, context engineering, tool use, etc.)?
• Evals and observability: What tools or products help evaluate quality, measure system behavior, and observe performance in real-world use?
• Models: Which models are best suited for “thinking” (reasoning, planning, decomposing problems) vs. “doing” (execution, coding, retrieval)?
• Learning path: How would you approach going from theory → prototype → production-quality system?
My goal is to discover high-quality resources that one can truly spend time learning from and building with—through iteration and practice—while also sharing what I learn so others on the same path can benefit.
Thanks in advance for sharing your experiences and guidance!