Level 1.5 · The AI Agents Roadmap, Simplified

How AI Agents Really Work

You've met AI and learned to prompt it. Before you build agents by hand (Level 2), get the mental model — the full roadmap.sh/ai-agents map, trimmed to the parts that matter, in plain English.

🧭
How to read this map: Start at the bottom and climb. Each yellow station is a big agent idea — tap it to open a deep dive. The white chips are the specific skills inside. You don't need to master everything — just build the mental model.

You are on Level 1.5 · Agents.  ·  Level 1 · Foundations  ·  Level 2 · Build agents (Weeks 2–5)

🏆 You get how agents work — now go build them (Level 2)
10

Agent Safety & Ethics

Keeping agents that act from doing harm.

Open deep dive →
Prompt injectiontricking the agent Sandboxing toolslimit what it can do Privacy & guardrailsPII, bias, red-teaming
9

Testing & Watching Your Agent

Know if it works — and see what it's doing.

Open deep dive →
Metrics & teststools + whole flows Human in the loopa person checks Tracing & logssee every step
8

Building Agents

Three ways: by hand, function calling, or a framework.

Open deep dive →
From scratchyour own loop Function callingOpenAI · Gemini · Claude FrameworksLangChain · CrewAI · LangGraph
7

Agent Architectures

Proven patterns for how an agent thinks.

Open deep dive →
ReActreason + act Chain of Thoughtthink step by step MCPa standard plug for tools Multi-agenta team of agents
6

Agent Memory

How an agent remembers across steps and sessions.

Open deep dive →
Short vs long termnow vs forever Vector DBsearchable memory Summarize & forgetkeep it small
5

Prompting Your Agent

The instructions that steer good behavior.

Open deep dive →
Be specificsay exactly what you want Give examplesshow, don't just tell Iterate & testrefine the prompt
4

Tools & Actions

The abilities that let an agent do things.

Open deep dive →
What's a tool?a function it can call Tool definitionname, inputs, outputs Examplessearch, code, APIs, files
3

The Agent Loop

The cycle at the heart of every agent.

Open deep dive →
Perceiveread the input Reason & plandecide next step Actcall a tool Observe & reflectcheck the result
2

The Engine Underneath

The LLM essentials every agent runs on.

Open deep dive →
Tokens & contextwhat it reads at once Temperature & controlstuning the output Open vs closedrun-your-own vs API Embeddingsmeaning as numbers
1

What Is an AI Agent?

An AI that doesn't just chat — it pursues a goal.

Open deep dive →
Agent vs chatbotacts vs just talks What's a tool?an ability it can use Real use casesassistants, coding, data
▶ Start here — you already know the AI basics