Interactive Tutorial

Build Your First AI Research Agent

A hands-on guide to building an AI research agent that searches the web, writes reports, and learns new skills — using plain English and Claude Code.

Scroll to explore ↓

What You'll Build

A working AI research agent in 45–60 minutes with zero code.

💬

Natural Language Input

Give it topics in plain English. It asks clarifying questions about scope, audience, and depth before starting.

🔍

Web Research

Searches the web, reads sources, pulls relevant information and compiles structured reports with citations.

🛠️

Extensible Skills

Add new capabilities like HTML export. The agent discovers and uses them automatically based on descriptions.

Three Levels of AI Work

Understanding the layers tells you what makes agentic workflows different.

Level 1

Chat

Ask a question, get an answer. ChatGPT-style.

You do the thinking

Level 2

Building

You tell the AI what to build step by step. It writes the code.

You direct, AI executes

Level 3

Agentic

You describe a goal. The AI plans steps, picks tools, adapts.

AI plans & executes. You steer.

✨ Three things that make a workflow truly agentic: process, decisions, clarification.

Seven Phases

From setup to deployment. Each phase builds on the last.

Phase 1

Set Up Your Workspace

Create a project folder and launch Claude Code inside it. Everything from here lives in this folder.

1
Create project: mkdir ~/my-first-agent && cd ~/my-first-agent
2
Launch agent: claude
3
Open second terminal tab. Keep it open to inspect files without leaving the agent.
Phase 2

Create Your CLAUDE.md

The single most important file. Tells the agent your rules, project structure, and formatting preferences.

1
Create with sections: Project Context, About Me, Rules, Project Structure.
2
Rules include asking clarifying questions, showing plans, concise output, citations.
3
Agent picks it up automatically on startup. No wiring needed.
Phase 3

Understand Three Layers

Every agentic workflow: Workflow (recipe), Agent (Claude Code), Tools (built-ins + custom).

💡
Key insight: A well-written workflow with built-in tools beats a sloppy prompt with every plugin.
Phase 4

Build the Research Workflow

Design the workflow on paper using Plan Mode — no files get created until you approve.

1
Switch to Plan Mode: Shift+Tab twice
2
Describe the goal — not the steps. The agent plans the approach itself.
3
Refine before accepting. Review the plan, add requirements.
4
Approve → verify folders (workflows/, output/, resources/).
Phase 5

Run the Agent

Give it a real research task. The agent asks clarifying questions, researches, and writes a report.

1
Exit Plan Mode, give a topic: "Research AI agents in 2026..."
2
Answer clarifying questions naturally, in plain English.
3
Watch the agent work — search, extract, synthesize. Approve as it goes.
4
Read the finished report in output/.
Phase 6

Iterate on the Output

Targeted edits without starting over. The agent keeps context and the project folder.

1
"Trim the executive summary to three bullet points."
2
"Add a comparison table of the top AI agent tools mentioned."
💡
You didn't list the tools. The agent identifies them and builds the table.
Phase 7

Give Your Agent a New Tool

Add a custom Skill — extend what the agent can do without touching its core.

1
Ask the agent to create a Skill at .claude/skills/html-report/SKILL.md
2
Skill includes instructions for converting Markdown to styled HTML.
3
Restart Claude Code to pick up the new skill (/exit then claude).
4
Put it to work: "Turn my report into a formatted HTML page."

Five to Avoid

Avoid these and you're ahead of most people learning this.

#1

Skipping CLAUDE.md

Two minutes of setup saves hours later. Just do it.

#2

Being too vague

"Do some research" produces garbage. Be specific about audience, scope, depth, and format.

#3

Skipping Plan Mode

If you jump straight to execution without reviewing the plan, the agent may go down the wrong path.

#4

Not telling it to ask questions

Without a "ask clarifying questions" rule, the agent assumes — and assumptions waste time.

#5

Trying to build everything at once

Start with one workflow. Get it working well. Refine it. Then build the next.

Level Up

Habits that separate good agentic workflows from great ones.

📝

Keep workflow files organized — use a workflows/ folder as you build more.

🧠

Read the agent's reasoning as it works — best way to catch mistakes before they snowball.

🎯

Targeted feedback, not restarts — tell the agent specifically what to fix.

🔄

Save your best files as starting points for new projects — build a library over time.

Esc interrupts — press it if the agent goes off course mid-response.

🎚️

Control reasoning depth — type effort low, effort medium, or effort high before a prompt.

Key Takeaways

If you remember nothing else from today.

1

Start with CLAUDE.md. It's the foundation — every project should have one.

2

A workflow is just a plain English recipe. File quality determines result quality.

3

Plan first, build second. Use Plan Mode (Shift+Tab twice) before creating files.

4

Be specific, and let the agent ask. Vague goals produce generic output.

5

Iterate, don't restart. The agent keeps context — give targeted feedback.

6

One workflow at a time. Get one working well before building the next.

7

You can give the agent new tools. A Skill is a folder with SKILL.md — auto-discovered.

Workflow Ideas to Try Next

🔍 Research across multiple sources and compile a report
📂 Analyze files and organize by category
📧 Draft personalized emails for contact lists
⚔️ Produce competitor side-by-side breakdowns
📊 Summarize articles into key themes

Cheat Sheet

Actions and shortcuts from the guide.

ActionHow
Launch Claude Codecd into project folder → run claude
Enter Plan ModeShift + Tab (press twice)
Interrupt the agentPress Esc
Clear contextType /clear at the prompt
List commandsType /help at the prompt
View a Markdown filecat file.md (or glow / bat)
Control reasoning depthPrefix with effort low / medium / high
List workflows / SkillsAsk: "What workflows/Skills do you have available?"
Quit Claude CodeType /exit or press Ctrl+C