AI Agents

roadmap.sh: https://roadmap.sh/ai-agents

Suggested path through the AI Agents nodes. Each node links to its lesson when written.

Nodes

Pre-requisites

What are AI Agents

LLM Fundamentals

Model Configuration / Tuning

Prompt Engineering

Tool / Function Calling

Building Agents Manually

Agent Architectures

Tools an Agent Can Use

Model Context Protocol (MCP)

Memory

Frameworks

Evaluation / Observability

Security / Safety

Agent Deployment

Example Use Cases

Resources

See resources.md.

Project ideas

  • MCP-powered research assistant: build an agent that exposes web search, a vector store, and a code-execution tool over MCP, then chains ReAct reasoning to answer multi-step questions with citations.
  • Self-evaluating coding agent: a planner-executor agent that generates code, runs it in a REPL sandbox, reads the failures, and iterates — wired to LangSmith/Langfuse for traces and a DeepEval test suite as a regression gate.
  • Personal inbox triage agent: an agent with long-term memory (vector DB) that reads email/Slack, classifies and drafts replies, escalates with human-in-the-loop approval, and redacts PII before logging.

2 items under this folder.