What Makes AI Agentic

There's a spectrum between "AI that completes your sentence" and "AI that completes your task." Most products sit somewhere in between, and understanding where is the first step to building agentic systems.

The Completion-to-Agency Spectrum

Think of AI systems along a spectrum:

• Level 0 — Completion: Autocomplete, code suggestions. Zero autonomy. The model generates text and stops.
• Level 1 — Single-turn tool use: The model calls one tool and returns a result. Like asking Claude to check the weather.
• Level 2 — Multi-turn reasoning: The model plans across multiple steps, uses multiple tools, adapts based on results. Claude Code working on a bug fix.
• Level 3 — Autonomous agents: The system operates with minimal human input, handles errors, and achieves complex goals over extended periods.

Most production systems today are at Level 1-2. The jump to Level 3 is where things get interesting — and risky.

The Observe-Decide-Act Loop

Every agentic system, at its core, runs a loop:

while (goal not achieved):
    observe  → read environment, tool results, user input
    decide   → reason about what to do next
    act      → call a tool, write code, send a message

This is what separates an agent from a chatbot. A chatbot responds. An agent pursues a goal through repeated cycles of observation and action.

Your Claude Code setup does exactly this: it reads your files (observe), thinks about what to change (decide), edits the code (act), then checks if it worked (observe again).

Why "Agentic" Is a Continuum

A system doesn't need to be fully autonomous to be agentic. Features that make something more agentic:

• Tool access: Can it interact with external systems?
• Decision loops: Does it decide its next action based on previous results?
• Error recovery: Can it detect and handle failures?
• Memory: Does it retain context across interactions?
• Goal persistence: Does it keep working until the goal is met?

The more of these a system has, the more agentic it is. Your MCP server ecosystem (Notion, Calendar, Slack, Linear) gives Claude Code tool access. Your knowledge/ directory gives it memory. Your CLAUDE.md gives it goals and constraints.

You're already building agentic products by configuring these components. This course teaches you to do it deliberately.