SkillAI · The Beam
Autonomous AI Agent Engineering
0/82 concepts·~34 h left
1
Phase 010/7 · ~3 h left
1. LLM Fundamentals and Prompt Engineering
How Transformer architecture works and why it matters for agent design (attention, context windows, hallucination patterns)25 min
Tokenization deep dive: why
gpt-4o charges per token and how to optimize costs25 minEmbeddings and semantic search: the backbone of agent memory (OpenAI
text-embedding-3-small, Cohere embed-v3)25 minPrompt engineering patterns that actually work in production: few-shot, zero-shot, role prompting, delimiters, XML tags25 min
System prompt architecture: how to write instructions that survive context overflow25 min
Evaluating LLM outputs with
Promptfoo and DeepEval25 minUnderstanding frontier models: GPT-4o, Claude 3.7 Sonnet, Gemini 2.0 Flash, Llama 3.3 70B — when to use each25 min
2
Phase 020/7 · ~3 h left
2. Agentic Patterns: ReAct, Chain of Thought, Tool Use
ReAct (Reason + Act): the loop that powers most production agents — how the LLM thinks, decides, acts, observes, and repeats25 min
Chain of Thought (CoT): making the model show its work — zero-shot CoT ("think step by step"), self-consistency, tree-of-thought25 min
Tool Use / Function Calling: OpenAI's
tools API, Anthropic's tool_use, how to define schemas that models actually understand25 minReWOO: separating planning from execution to reduce token costs 40-60%25 min
Reflexion: agents that learn from their mistakes via self-reflection loops25 min
Combining patterns: when to use ReAct vs. pure function-calling vs. hybrid approaches25 min
Real-world case studies: how Devin, Cursor, and Claude Code use these patterns internally25 min
3
Phase 030/8 · ~3.5 h left
3. Building AI Agents with LangChain
LangChain architecture: chains, runnables, LCEL (LangChain Expression Language)25 min
Building agents with
create_react_agent and AgentExecutor25 minTool ecosystem:
DuckDuckGoSearchRun, PythonREPLTool, WikipediaQueryRun, custom tools25 minIntegrating with 50+ LLM providers via
langchain-openai, langchain-anthropic, langchain-google-genai25 minMemory in LangChain:
ConversationBufferMemory, ConversationSummaryMemory, VectorStoreRetrieverMemory25 minStructured output with
PydanticOutputParser and .with_structured_output()25 minDebugging with LangSmith: traces, latency, cost tracking per run25 min
Callbacks and streaming for real-time UX25 min
4
Phase 040/9 · ~4 h left
4. Building Stateful Agents with LangGraph
LangGraph mental model: nodes, edges, state, conditional routing25 min
Why graphs beat linear chains for complex agents (cycles, parallel branches, error recovery)25 min
StateGraph vs MessageGraph — when to use each25 minBuilding a full agent loop with tool nodes, decision nodes, and human approval steps25 min
Human-in-the-loop: interrupt, review, resume patterns for high-stakes actions25 min
Parallel execution: running multiple agent branches simultaneously with
Send()25 minPersistence with
SqliteSaver and PostgresSaver — agents that remember between sessions25 minLangGraph Cloud: managed deployment with built-in streaming and persistence25 min
Subgraphs: composing complex multi-step workflows from smaller reusable graphs25 min
5
Phase 050/8 · ~3.5 h left
5. MCP (Model Context Protocol) Integration
MCP architecture: hosts, clients, servers, and the JSON-RPC 2.0 protocol underneath25 min
MCP primitives: Tools (actions), Resources (data), Prompts (templates)25 min
Building your first MCP server in Python with
mcp SDK25 minConnecting to existing MCP servers: filesystem, GitHub, PostgreSQL, Slack, Google Drive25 min
MCP in Claude Desktop, Cursor, Windsurf, and VS Code Copilot25 min
Context management strategies: what to expose, what to cache, how to avoid context overflow25 min
Security considerations: input sanitization, permission scoping, audit logging25 min
The MCP ecosystem in 2026: 500+ community servers,
smithery.ai directory25 min6
Phase 060/8 · ~3.5 h left
6. Multi-Agent Orchestration
Multi-agent architectures: hierarchical (supervisor + workers), peer-to-peer, blackboard systems25 min
CrewAI: role-based agents with defined goals, tools, and backstories — the easiest way to get a multi-agent team running25 min
AutoGen (Microsoft): conversation-based multi-agent framework with code execution25 min
OpenAI Swarm / OpenAI Agents SDK: lightweight orchestration with handoffs25 min
Agent communication protocols: shared state vs. message passing vs. tool calls25 min
Task decomposition: how a supervisor agent breaks complex tasks into subtasks and delegates25 min
Conflict resolution: what happens when agents disagree25 min
Debugging multi-agent systems with LangSmith traces and CrewAI's built-in logging25 min
7
Phase 070/8 · ~3.5 h left
7. Memory Systems: Short-Term and Long-Term
The 4 types of agent memory: in-context (conversation history), external semantic (vector DB), external episodic (event logs), procedural (learned behaviors)25 min
In-context memory optimization: sliding window, summarization, token budgeting25 min
Vector databases for semantic memory: Pinecone, Qdrant, Chroma, pgvector — tradeoffs and when to use each25 min
Mem0: the managed memory layer for AI agents — add/search/update memories with one API call25 min
Zep: long-term memory with automatic summarization and entity extraction25 min
Redis for short-term session memory: blazing fast, simple, ephemeral25 min
Memory retrieval strategies: MMR (Maximal Marginal Relevance), HyDE, time-weighted retrieval25 min
Entity memory: extracting and storing facts about users, projects, and relationships25 min
8
Phase 080/9 · ~4 h left
8. Agent Evaluation and Testing
Why traditional software testing fails for agents (non-determinism, emergent behavior)25 min
Evaluation dimensions: task success rate, faithfulness, answer relevancy, tool selection accuracy, hallucination rate25 min
RAGAS: automated RAG and agent evaluation framework — run 10K evaluations overnight25 min
AgentBench and GAIA benchmark: industry-standard agent benchmarks25 min
LLM-as-a-judge: using GPT-4o to evaluate GPT-4o outputs (and why it works)25 min
Red teaming agents: adversarial inputs, prompt injection attacks, jailbreak attempts25 min
Regression testing: building a test suite that catches regressions when you update prompts25 min
A/B testing agent versions in production with feature flags25 min
Cost/quality tradeoff analysis: when to use GPT-4o vs GPT-4o-mini vs Claude Haiku25 min
9
Phase 090/10 · ~4 h left
9. Production Deployment and Monitoring
Deployment patterns: REST API (FastAPI), WebSocket streaming, async queues (Celery + Redis)25 min
Containerizing agents with Docker — handling model credentials, secrets, environment vars25 min
Serverless deployment with Modal (best for AI workloads) and Railway25 min
LangServe: deploying LangChain agents as production-ready REST APIs in minutes25 min
Observability stack: LangSmith for traces + Prometheus + Grafana for metrics25 min
Cost monitoring: tracking $/1000 calls per agent, setting spend alerts25 min
Rate limiting, retries, and exponential backoff for LLM API calls25 min
Handling failures gracefully: fallback models, circuit breakers, user-facing error messages25 min
Scaling strategies: stateless agents + Redis for shared state, horizontal scaling with K8s25 min
Guardrails: NeMo Guardrails, Guardrails AI — input/output validation, content moderation25 min
10
Phase 100/8 · ~3.5 h left
10. Real-World Agentic Applications
Coding agents: how Cursor, Devin, and GitHub Copilot Workspace are built — file editing, test running, git operations25 min
Data analysis agents: natural language → SQL → charts (Text2SQL with validation loops)25 min
Browser automation agents: Playwright + LLM for autonomous web interaction (
browser-use library)25 minDocument processing pipelines: PDF → structured data with LlamaParse + agent validation25 min
Customer support agents: intent classification, RAG-based FAQ, escalation logic, CRM integration25 min
Ethical deployment: bias auditing, transparency requirements, EU AI Act compliance for high-risk agents25 min
Building a Minimum Viable Agent (MVA): scope definition, risk assessment, rollout strategy25 min
Future landscape: OpenAI o3, Gemini 2.0 native tool use, agents with persistent browser sessions25 min