The 11 sections

Tap any section to jump to its reference card.

• 1The 7 Eras of AIRule-based → agentic, each era stacks
• 2Why Agents Are Possible NOWFive capabilities converged
• 3Three ApproachesScript → FastAPI → Agent
• 4What IS / ISN'T an AgentWho decides next: your code or Claude?
• 5The 7 Building BlocksBrain, Tools, Memory, Plan...
• 6The Agent LifecycleDesign → Build → Protect → Observe → Deploy
• 7How Agents Actually WorkThe universal while-loop
• 8Common MisconceptionsMyth vs truth
• 9Market StatsNumbers to keep in your pocket
• 10Flash CardsTap-to-reveal self-check
• 11Course Track Map7 tracks, 30 modules

The 7 Eras of AI

Each era added a capability without erasing the previous ones. Today's agents still use rules, ML, transformers, and RAG — all at once. You add a reasoning layer on top; you don't replace what works.

MCP = agent↔tool (Anthropic, 2024). A2A = agent↔agent (Google → Linux Foundation, 2025).

Why Agents Are Possible NOW

The headline isn't a smarter model — it's the scaffolding around it. Five capabilities went from impossible to production-ready between 2022 and 2026.

Three Approaches

Same question — "Is Acme Corp likely to go delinquent?" — three ways to answer. The ML model never disappears; the agent just uses it.

What IS / ISN'T an Agent

One question: who decides what happens next — your code, or Claude?

The 7 Building Blocks

Like parts of a body — drop any one and your agent stops being trustworthy. Most tutorials teach only blocks 1–2.

Brain + Tools = 2 of 7. An agent that works on your laptop but can't be trusted in production is a prototype, not a product.

The Agent Lifecycle

Building an agent isn't "write code and ship." Five stages, each adding a critical layer.

1. DesignScope, tools, data access. (Tracks 1–2)
2. BuildDefine tools, engineer prompts, wire the loop, add RAG. (Tracks 2–4)
3. ProtectValidate I/O, cap loops, add human-in-the-loop. (Track 5)
4. ObserveTrace every tool call, log decisions, score outputs. (Track 6)
5. DeployFastAPI/Docker, Cloud Run/Lambda, cost + rate limits. (Track 7)

⚠ "I'll add guardrails later" costs 4× more. Build safety in from the start — even a simple max_turns=10 counts.

How Agents Actually Work

An agent is your code calling Claude's API in a loop. No magic runtime — you are always in control.

# THE UNIVERSAL AGENT PATTERN
user_asks_a_question()

while claude_wants_a_tool:
  reply = send_to_claude(question + history)

  if reply.stop_reason == "tool_use":
    result = run_tool(reply.name, reply.args)
    history.append(result)   # one iteration
  else:
    return reply.text        # final answer

Common Misconceptions

Market Stats to Memorize

When a peer asks "is agentic AI real or hype?" — keep these in your back pocket.

Quick-Reference Flash Cards

Tap to reveal the answer. If you can answer all six, you're ready for Module 1.

Course Track Map

7 tracks, 30 modules — building block by building block.