From User to Builder: Claude Code Agent Bootcamp

Description

From User to Builder: Claude Code Agent BootcampThe term Agentic AI is being heard a lot these days. However, there are few developers who can answer, "So how do you actually create a real product with that?" This bootcamp is the process of finding that answer hands-on. Over the course of 6 weeks, we will create a small project (MVP) from start to finish using Agentic AI — from context engineering, MCP server development, to multi-agent and Agent SDK. What remains upon completion is not notes but my working MVP."Taught by someone who has commercialized global services."Date: July 8, 11, 18, 25, August 1, August 8, August 15 [Total 6 weeks · 6 online sessions (Wednesdays) + 6 offline sessions (Saturdays)]Off/On Line Time (PST): 5:00 pm ~ 6:30 pm (1.5 hours/session, total 18 hours)Location [Offline_PNP, Online_Zoom, to be announced later]Price £150 (for all 6 sessions)Seats Limited to 40 people (first come, first served)Light Refreshment will be providedCourse DescriptionWeek 1. Agentic AI Foundations & How LLMs WorkFrom chatbot to agent: the autonomy spectrum, why you always start on the left, and how LLMs really work.What is the difference between a chatbot and an agent? We will learn the principles of autonomy spectrum (single call → workflow → agent → multi-agent) and the principle of “always start on the left,” as well as how LLMs operate, including token, context window, non-determinism, and structured output. (Saturday) Non-determinism quantification experiment + Structured output parser practice.Week 2. Claude API, Memory & System PromptsMaster the Messages API and design the system prompts and memory that make an agent reliable.We learn about the Claude ecosystem and model selection for tasks, designing prompts and system prompts, and managing short/long-term memory and state. (Saturday) Multi-turn conversational CLI assistant + session state manager implementation.Week 3. Context Engineering, RAG & Cost OptimizationDecide what goes into context every step — and make calls cheap enough to run in production.We design the operational call layer with context engineering, compaction, RAG (chunking, vector/hybrid search, re-ranking), prompt caching, and model cascading.Week 4. Tool Use & MCP — Build the Agent Loop ★Hackathon TeamBuild the agent’s heart — a tool-execution loop from scratch — then standardize it with MCP. Your capstone starts today.Implement the tool execution loop from the ground up without a framework, and standardize with the ReAct pattern and MCP (Model Context Protocol) server. From this week, teams of 2-4 will form to start the hackathon.Week 5. Patterns: Workflows, Agents & Multi-AgentThe pattern catalog — workflow patterns, agent patterns, and when multi-agent is actually worth it.We learn 5 types of workflow patterns, agent patterns (Reflection, Plan-and-Execute), the pattern selection guide, and multi-agent (Supervisor, Hierarchical, Handoff).Week 6. Production: Evals, Guardrails & Observability ★Hackathon Demo DayMake it production-ready — evals, guardrails, tracing — then evaluate and present your capstone.We learn about evals, trajectory evaluations, regression testing, guardrails (least privilege, prompt injection defense, sandboxing), and observability (tracing) to apply in team projects.A 2-4 person team hackathon to create one MVP — for example, an operational agent that reads meeting notes and organizes tasks — proceeds with idea generation in week 3, team formation in week 4, and application of tools, patterns, and guardrails each week, culminating in demo day presentations, evaluations, and awards (3 teams). Even without an idea or team, participants can join as individual seed idea tracks, and upon completion, everyone will have a working MVP and GitHub portfolio.