Claude Architect Foundations: AI Agents & Claude Code

In this introduction, you’ll learn what the Claude Architect Foundations course covers, how the course is structured, and what you’ll learn about AI agents, MCP, and Claude Code.

Get introduced to the Claude Architect Foundations course and learn the fundamentals of AI agents, agentic architecture, MCP, tool calling, and Claude Code. This course provides a practical foundation for designing and building Claude-powered AI workflows.

Learn how Claude Subagents work and how they can be used to delegate specialized tasks, manage complex workflows, and build scalable multi-agent systems. Explore practical strategies for using subagents effectively while maintaining control and context.

Learn how to use Claude Code inside Visual Studio Code to write, understand, debug, and improve code with AI assistance. Explore practical workflows for integrating Claude into your development environment and building applications more efficiently.

Learn what an AI agent is, how it differs from a traditional chatbot, and how agents use reasoning, tools, and actions to accomplish tasks. Build a clear foundation for understanding agentic AI systems and Claude-powered workflows.

Learn when AI agents are the right solution for a problem and when simpler approaches are better. Explore complexity, latency, cost, and other factors for making effective agentic AI decisions.

Understand how agentic loops enable Claude to reason, plan, use tools, observe results, and take the next action until a task is completed.

Follow a practical implementation of an agentic loop and see how Claude interacts with tools, processes results, and continues working toward a final answer.

Understand the difference between orchestrators and subagents, their responsibilities, and how they work together to build efficient AI agent systems.

Learn how multiple specialized agents can work together in a coordinated pipeline to solve complex tasks through sequential or parallel execution.

Learn how agents transfer tasks and context safely using clear handoff protocols and workflow boundaries, creating reliable multi-agent systems.

Explore agent lifecycle events and SDK integration points for monitoring, logging, instrumentation, and controlling agent behavior throughout execution.

Learn how to preserve agent state, create checkpoints, and resume interrupted workflows so long-running agent tasks can continue reliably.

Learn how Claude uses tools to interact with external systems and perform actions beyond conversation. Understand the tool-calling flow and how agents decide when to use a tool.

Learn how to design clear and reliable tool schemas that Claude can understand and use effectively, including parameters, types, descriptions, and expected inputs.

Learn how to combine multiple tools into a sequence of actions to solve complex tasks. See how tool results can drive the next step in an agentic workflow.

Learn how to handle tool failures, invalid inputs, and unexpected results so your AI agents can recover gracefully and continue working reliably.

Learn the fundamentals of Model Context Protocol (MCP) and set up an MCP server to connect Claude with external tools and capabilities in a standardized way.

Put everything together in a practical capstone by building a hooked and chained MCP agent that combines tools, workflows, error handling, and lifecycle controls.

Set up Claude Code and learn the fundamentals of project configuration, including CLAUDE.md and how instructions can guide Claude's behavior within your projects.

Learn how to extend Claude Code with reusable commands, skills, and project-specific strategies to create more capable and efficient development workflows.

Learn how to control Claude Code's level of autonomy using Plan Mode, permissions, and tool allowlisting for safer and more predictable workflows.

Learn how hooks can automatically enforce rules and how subagents can handle specialized tasks, helping you build safer and more scalable Claude Code workflows.

Learn how to work effectively with Claude through iterative refinement, context management, and feedback loops to improve results and handle complex tasks efficiently.

In this lesson, discover why Context Engineering is becoming more important than traditional Prompt Engineering when building reliable AI agents. Learn how an agent’s context—including system instructions, conversation history, tool definitions, tool results, files, and retrieved information—shapes its ability to reason and make effective decisions.

You’ll understand the key differences between Prompt Engineering and Context Engineering, explore concepts such as context overload, attention limits, and context rot, and learn why simply adding more information to a prompt does not always produce better results.

By the end of this lesson, you’ll understand how to strategically design, curate, and manage context to build AI agents that are more reliable, focused, and effective in real-world, long-running workflows.

Learn how Prompt Caching can improve the performance, consistency, and cost efficiency of AI agents working with large and repeated context.

In this lesson, you’ll understand how prompt caching works in agentic workflows and why caching stable information—such as system instructions, tool definitions, and reusable context—can reduce unnecessary processing across multiple agent interactions.

You’ll explore cache breakpoints, cacheable context, cache lifetime, and the practical considerations of using caching with long-running agents. You’ll also understand the difference between Prompt Caching and Context Compaction, and when each technique should be used.

By the end of this lesson, you’ll be able to identify opportunities to apply prompt caching in agentic systems and design more efficient workflows for agents that repeatedly work with the same context.

Learn how Retrieval-Augmented Generation (RAG) can be used to make AI agents more knowledgeable, accurate, and capable of working with external information.

In this lesson, you’ll understand the difference between traditional Chat RAG and Agentic RAG, and how AI agents can decide when they need to retrieve information, what information to retrieve, and how to use it during multi-step tasks.

You’ll explore how agents can perform iterative retrieval, evaluate retrieved information, verify sources, and combine multiple pieces of knowledge before producing a final response. You’ll also learn why retrieval becomes especially important for long-running research and knowledge-intensive agent workflows.

By the end of this lesson, you’ll understand how to design RAG workflows where the agent actively controls the retrieval process, rather than simply receiving retrieved information as part of a fixed pipeline.

Learn how to design persistent memory for AI agents so they can retain important information across conversations and resume work more effectively over time.

In this lesson, you’ll understand the difference between persistent memory and checkpointing, and why an agent may need to store specific facts, preferences, decisions, and important findings rather than saving its entire conversation history.

You’ll explore practical memory patterns, including how agents can write important information to memory, retrieve relevant memories when needed, and use those memories to improve future interactions.

By the end of this lesson, you’ll understand how persistent memory can help AI agents maintain continuity across sessions, support long-running workflows, and make more informed decisions without unnecessarily increasing the active context.

Learn how Context Compaction and Summarization help AI agents manage long-running workflows without exceeding context limits or losing important information.

In this lesson, you’ll explore techniques for compressing lengthy conversation and tool-result histories while preserving the information an agent needs to continue working effectively. You’ll understand how compaction reduces unnecessary context and how summarization helps maintain continuity across long-running tasks.

You’ll also understand the relationship between context compaction, summarization, and the /compact mechanism introduced earlier in the course, including when these strategies should be applied in production agentic systems.

By the end of this lesson, you’ll understand how to manage growing context efficiently and keep AI agents coherent, focused, and effective during long-horizon workflows.

In this capstone, you’ll combine key agentic techniques including Prompt Caching, RAG, Persistent Memory, and Context Compaction to create an agent that can research, retrieve information, preserve important findings, and manage its context throughout a long-running workflow.

You’ll see how these techniques work together to help an agent maintain coherence, accuracy, and continuity as the task grows in complexity.

By the end of this capstone, you’ll have a practical understanding of how to design a more robust long-running research agent and how the concepts from this module fit together in a real-world agentic workflow.

Learn why evaluating AI agents requires a different approach from evaluating a single LLM response.

In this lesson, you’ll compare traditional evaluation methods—such as accuracy, tone, and formatting—with agent-specific evaluation, where the entire execution process matters. You’ll explore how to evaluate an agent’s tool selection, tool-call sequence, decision-making, error recovery, and ability to stop when the task is complete.

You’ll also learn the difference between trajectory evaluation and outcome evaluation, and why both are important when measuring the reliability of agentic systems.

By the end of this lesson, you’ll understand how to think beyond the final answer and evaluate whether an agent followed a reliable and effective path to reach its result.

Learn how to build a practical evaluation framework for AI agents that can continuously test agent behavior as your system evolves.

In this lesson, you’ll learn how to create representative evaluation datasets containing normal tasks, edge cases, failure scenarios, and adversarial inputs. You’ll explore different types of graders, including rule-based checks, LLM-as-a-judge evaluations, and human review.

You’ll also learn how to organize these evaluations into a repeatable test harness that can automatically run multiple agent tasks and measure their performance consistently.

By the end of this lesson, you’ll understand how to move from manually testing an agent to building a systematic evaluation process that helps identify regressions and improve reliability over time.

Learn where hallucinations can occur inside agentic workflows and why they are not limited to the agent’s final response.

In this lesson, you’ll explore how agents can make incorrect assumptions about tool results, fabricate parameters, misinterpret retrieved information, or silently continue after a failed step. You’ll learn practical techniques for reducing these failures through self-verification, tool-result validation, cross-checking, and evidence-based reasoning.

You’ll also understand why agents should verify important information before using it to make decisions or take actions.

By the end of this lesson, you’ll be able to identify common sources of hallucination in multi-step agent workflows and apply verification strategies to make agent behavior more accurate and trustworthy.

Learn how to design layered guardrails that control what an AI agent can receive, produce, and do.

In this lesson, you’ll explore three key layers of protection: input validation, output filtering, and behavioral constraints. You’ll understand how these guardrails can prevent invalid requests, unsafe outputs, and actions that an agent should never perform—even when prompted to do so.

You’ll also learn the difference between prompt-based guardrails and code-enforced guardrails, including why deterministic controls are essential for high-risk actions. This lesson connects the guardrail concepts introduced earlier with broader API and system-design patterns for production agents.

By the end of this lesson, you’ll understand how to combine soft and hard controls to build AI agents that are safer, more predictable, and easier to govern.

Learn how to introduce Human-in-the-Loop (HITL) checkpoints that give people meaningful control over high-risk agent actions without eliminating the benefits of automation.

In this lesson, you’ll explore approval gates for actions such as sending emails, making purchases, or deploying code. You’ll also learn how confidence-based escalation can allow an agent to recognize uncertainty and request human assistance when necessary.

You’ll discover practical design principles for deciding when human approval is required, when the agent can act autonomously, and how to avoid approval fatigue.

By the end of this lesson, you’ll understand how to design effective human oversight that balances automation, safety, and human control in real-world agentic systems.

In this capstone, you’ll deliberately test an agent with malformed inputs, ambiguous instructions, prompt injection attempts, tool failures, and other adversarial scenarios to discover weaknesses in its behavior.

You’ll then apply the techniques learned throughout the module—including evaluations, guardrails, tool-result verification, and Human-in-the-Loop checkpoints—to improve the agent’s reliability and resilience.

By comparing the agent’s behavior before and after hardening, you’ll see how systematic testing and defensive design can transform an experimental agent into a more predictable, measurable, and production-ready system.

By the end of this capstone, you’ll have a practical reliability workflow and a strong portfolio-ready demonstration of how to evaluate, attack, and improve an AI agent.