Free AI agent course

AI agent course for building real tools with Codex and Claude Code

This is the first technology that can explain how it works while it works. You do not need to know code first. You need a private repo, a good coding agent, clear questions, and the habit of asking it to explain every step before it changes anything.

Start the practical course Use the prompts

Recommended stack

Start with coding agents, not no-code glue.

Why this matters

Codex and Claude Code are better than n8n when you do not know the system yet.

n8n is useful when the workflow is already known: if this happens, send that email, update that row. Coding agents are different. They can inspect a repo, explain unfamiliar code, write a dashboard, change the data model, add tests, deploy, and then teach you what they did in plain English.

n8n

Best for fixed automations

Great for predictable integrations where the steps are stable and the logic is simple enough to model as nodes.

Codex / Claude Code

Best for building tools

Better when you need a custom app, dashboard, data workflow, agent skill, API integration, or repeated product iteration.

The shift

You are not learning to code from scratch. You are learning how to supervise a technical employee who can code, explain, test, and leave a trail you can review.

Starter prompts

When you do not understand something, ask the agent to slow down.

Explain like I am 5

Explain what this app does like I am 5. Then explain it again like I am a non-technical founder.

Plan before editing

Before changing any files, explain in simple terms what you plan to do, which files you will touch, and how I can verify it.

Teach while doing

I do not know code. After each step, tell me what changed, why it matters, and what question I should ask next.

Review risk

Review this plan for security, privacy, and deployment risks. Tell me what should stay private and what permissions are too broad.

Practical tutorial

Build your first agent-powered internal tool.

The goal is not to watch theory. The goal is to create a private GitHub repo, ask for a useful dashboard, ship it on Vercel, and leave the agent with reusable instructions for the next project.

Pick Codex or Claude Code

Use a coding agent that can read files, edit files, run checks, and explain the system. This is why it beats generic chat for building tools.

Agent selected

Create a private GitHub repo

Keep the code in a private repository so every change is tracked, reversible, and ready to deploy.

Private repo

Install the agent locally

Install Codex or Claude Code, open the repo, authenticate, run the health check, and start with read-only questions before edits.

Working terminal

Write AGENTS.md and CLAUDE.md

Give the agent project context, rules, commands, forbidden files, testing instructions, and your communication style.

Agent playbook

Build a custom dashboard

Ask for a dashboard that solves a real problem: weekly sales signals, support load, invoices, content pipeline, or founder brief.

First app

Deploy simply on Vercel

Connect the GitHub repo to Vercel so every commit can become a preview and the finished dashboard gets a stable URL.

Live URL

Add work tool access

Connect Gmail, Calendar, Drive, GitHub, Slack, Stripe, or internal APIs with the least permission needed.

Useful context

Turn work into skills and routines

Package repeatable prompts, checks, scripts, and domain rules so the agent can reuse them instead of starting from zero.

Reusable system

Install

Set up your coding agent like a safe workbench.

Use a paid account if you plan to work seriously. A $200/month 20x plan is not expensive when you compare it to a real employee, not a SaaS subscription. If it saves a few hours of thinking, debugging, documentation, or admin work each month, it has already paid for itself. The real benefit is that you stop rationing prompts when you need the agent to read, explain, retry, test, and document.

Codex

Install OpenAI Codex CLI

Codex is strong when you want an agent that works across files, can run checks, and fits naturally with GitHub-based work. Then run codex --login.

npm install -g @openai/codex

Claude Code

Install Claude Code

Claude Code is excellent for long explanatory sessions, project memory, custom slash commands, hooks, subagents, and skills. Then run claude doctor.

npm install -g @anthropic-ai/claude-code

GitHub

Create a private repository

Use a GitHub account and create a private repo. Ask the agent to explain commits, branches, and pull requests in simple language.

Paid plan

Do not ration intelligence

A $200/month power-user plan is usually cheaper than one loaded day of a real employee. The point is not chat. The point is sustained execution.

Agent memory

Your repo needs an instruction file before it needs more prompts.

Add an AGENTS.md file for Codex and a CLAUDE.md file for Claude Code. These files explain what the project is, how to run it, what not to touch, how to test, and what "done" means. They make the agent act less like a random chatbot and more like a teammate.

# AGENTS.md / CLAUDE.md

Project: internal dashboard for weekly sales and support signals.

Rules:
- Explain the plan in simple terms before editing files.
- Keep changes small and easy to review.
- Never read .env, secrets, private exports, or customer data unless asked.
- After each change, run the relevant check and summarize what changed.
- If you are unsure, ask one question before guessing.

Copy paste

Use these prompts when you want to code without knowing code.

First read

I do not know code. Read this project and explain what it does in simple non-technical language. Do not edit files yet.

Dashboard brief

Build a private dashboard for my business. First ask me 5 questions about the data sources, metrics, users, and decisions it should support.

Before edits

Before changing files, explain the plan, the files you will touch, the risk level, and exactly how I can test the result.

After edits

Summarize the diff like I am non-technical: what changed, why it matters, what to click, and what I should review before deploying.

Security pass

Check whether this project contains secrets, private customer data, overly broad permissions, or risky automation. Propose safer defaults.

Make it reusable

Turn what we just did into a reusable skill, slash command, or AGENTS.md / CLAUDE.md instruction so I can repeat it next week.

Workflow

Code without knowing code by making GitHub the source of truth.

Start private

Create a private GitHub repo before the first line of code. You want history, rollback, and a clean place for the agent to work.

Ask for the smallest useful dashboard

Example: build a private dashboard that shows weekly leads, support tickets, overdue invoices, and a written founder brief. Ask the agent to question you before it builds.

Review every diff

Ask the agent to summarize what changed, why it changed, how to test it, and whether any file contains secrets or customer data.

Deploy with Vercel

Import the GitHub repo into Vercel, choose the framework preset, set environment variables, and use preview deployments for each change.

Skills, routines, automations

Teach the agent repeatable know-how instead of rewriting prompts every week.

Skills

Install domain skills

A skill is reusable know-how in a SKILL.md file. Use agentskill.sh or /learn to find skills, then try a concrete one like Paperasse for administrative document workflows.

Routines

Create Claude Code slash commands

Turn recurring prompts into commands such as /brief, /ship, /security-review, or /weekly-dashboard so the agent follows the same process every time.

Automations

Use hooks carefully

Claude Code hooks can run commands after edits or at session events. Use them for checks, formatting, logs, and notifications, not risky production actions.

Subagents

Split specialist work

Use subagents for focused jobs: code review, security, research, UX copy, QA, or data analysis. Give each only the tools it needs.

Security

Treat skills like code

Read skills before installing them. Check scripts, network calls, permissions, and whether they ask for secrets. Do not install random skills blindly.

Paperasse

Use a real example

Paperasse-style skills show the point: package expertise around messy admin documents so the agent follows a repeatable, auditable process.

Work tools

The real leverage is giving the agent access to the tools where work happens.

A coding agent becomes much more useful when it can read the systems your team already uses: Gmail, Google Calendar, Drive, Notion, Slack, Linear, GitHub, Stripe, internal APIs, and databases. Give access gradually, with the minimum permission needed.

Gmail and Calendar

Ask for a weekly brief, unanswered emails, meeting prep, follow-up drafts, and calendar conflicts.

Drive and docs

Let the agent summarize folders, extract decisions, compare versions, and turn documents into structured data.

GitHub and Vercel

Use GitHub for history and Vercel for simple hosting, preview URLs, environment variables, and production deploys.

Slack, CRM, Stripe, internal APIs

Start read-only, then add narrow write actions once the workflow is trusted and logged.

Security

Use powerful agents, but keep a boring security model.

Start private, limit access, deny sensitive files, review diffs, and never paste production secrets into prompts. The agent should explain, propose, test, and document. It should not silently make risky changes to money, customer data, or production systems.

Private repo by defaultNo secrets in promptsDeny .env and secret filesReview diffs before mergeLeast-permission connectorsHuman approval for money or customer dataLogs for automated actionsSmall changes with tests

FAQ

AI agent course questions

Can I really code without knowing code?

Yes, if you treat the agent like a technical operator and keep the workflow small. Your job is to ask clear questions, review explanations, test the result, and keep everything in GitHub.

Should I use Codex or Claude Code?

Use both if you can. Codex is very strong for GitHub and coding-agent workflows. Claude Code is excellent for long interactive sessions, memory, slash commands, hooks, subagents, and skills.

Why not just use n8n?

Use n8n when the automation is already clear and node-based. Use coding agents when you need to design, build, debug, explain, deploy, and evolve a custom internal tool.

Is a $200/month paid account worth it?

For serious work, usually yes. Compare it to employee time, not consumer software. If the agent saves a few hours a month or helps ship one internal tool, it pays for itself quickly.

What should I build first?

Build a small dashboard that uses real work context: sales follow-ups, support tickets, weekly founder brief, invoices, content pipeline, or operations checklist.

How do I keep this safe?

Use private repos, deny secret files, connect tools gradually, review diffs, keep humans in the loop, and treat third-party skills like code you must inspect before running.

From course to implementation

Build the dashboard, agent, or internal workflow your team actually needs.

If you want to skip the messy first attempts, I can help you set up the repo, connect the work tools, design the agent instructions, build the first dashboard, and ship it on Vercel with a clean handoff.

Get AI agent consulting