A practical roadmap based on the 3-level AI framework from your Reel
Updated: October 2026

How to Use This Guide

This is not a list of random AI links. The 10 resources are arranged as a learning path. Start with the first resources if you are mainly using ChatGPT, Claude or Gemini for questions. Then move into assistants, tools, memory and finally autonomous agent workflows.

The resources are free to access and learn from. Some platforms may require an API key, local hardware, or paid usage if you actually run cloud models or high-volume workloads.

The 3 Levels

LEVEL 1 — AI CHATBOTS
Ask → Answer
ChatGPT / Claude / Gemini

LEVEL 2 — AI ASSISTANTS
Goal + Instructions → AI performs tasks
Tools / browsing / research / files

LEVEL 3 — AI AGENTS
Goal → Agent plans → Uses tools → Executes workflow → Reports result

Level 1 → Level 2

1. OpenAI Agents SDK — Start Building Your First Agent

Why start here: This is the clearest bridge from chatting with an AI model to building an application where an AI can use tools and execute multi-step work.

Learn: agent instructions, tools, runs, state, handoffs and guardrails.

Free learning: Yes. The documentation and SDK examples are free to access. API usage may cost money.

Official resource: Official: https://developers.openai.com/api/docs/guides/agents/sdk

2. OpenAI Agents SDK Quickstart

Why: Follow one small project from installation to a working agent instead of trying to learn the entire agent ecosystem at once.

Project: Build a personal research assistant that takes a question and uses one or two tools.

Free learning: Yes. You only pay if you use paid API services.

Official resource: Official: https://developers.openai.com/api/docs/guides/agents/quickstart

3. Model Context Protocol (MCP)

Why: MCP teaches the connection layer between AI models and external tools/data. It is one of the most useful concepts for moving beyond simple prompting.

Learn: MCP servers, tools, resources and how assistants can interact with external systems.

Free learning: Yes. The specification and documentation are publicly available.

Official resource: Official: https://modelcontextprotocol.io/

4. OpenAI Codex

Why: Learn what an agentic coding workflow looks like when the AI can inspect files, work through a repository and perform development tasks instead of only answering coding questions.

Project: Give a small GitHub project to a coding agent and ask it to understand, test and improve one feature.

Free learning: Documentation and open-source tooling are available; model/API usage can have costs depending on the setup.

Official resource: Official: https://github.com/openai/codex

5. Claude Code

Why: Claude Code is a practical way to experience agentic software development. It is especially useful if your goal is to build SaaS products and websites.

Learn: repository context, terminal work, code changes, tests, debugging and iterative agent workflows.

Free learning: Documentation is free to read; using Claude models may require a supported paid plan or API usage.

Official resource: Official: https://docs.anthropic.com/en/docs/claude-code

Level 2 → Level 3

6. OpenJarvis — Personal AI on Your Device

Why: OpenJarvis is useful for understanding personal AI that combines models, tools, memory and scheduled workflows. It can run locally and cloud APIs are optional.

Project: Install it locally, try the simple chat agent, then explore the morning digest, deep research, memory and code-assistant workflows.

Free learning: Yes. The project is open source. Cloud model/API usage can still cost money.

Official resource: Official: https://github.com/open-jarvis/OpenJarvis

7. OpenJarvis Tutorials

Why: The tutorials move from concepts into complete working applications instead of isolated prompts.

Projects: Deep Research Assistant, Scheduled Personal Ops, Messaging Hub and other agent workflows.

Free learning: Yes.

Official resource: Official: https://github.com/open-jarvis/OpenJarvis/tree/main/docs/tutorials

8. n8n AI Agent Documentation

Why: n8n is useful when you want to see agents connected to real business workflows and external services.

Project: Build an automation that receives a request, calls an AI agent, uses a tool, and sends the result to another service.

Free learning: Yes. Self-hosting can be free; cloud hosting and some connected services may cost money.

Official resource: Official: https://docs.n8n.io/advanced-ai/

9. Google AI / Gemini Developer Resources

Why: Learn another major model ecosystem and understand how tool calling and multimodal AI fit into real applications.

Project: Build a small Gemini-powered assistant that accepts text or images and calls one external function.

Free learning: Yes. Google provides documentation and developer resources; quotas and API pricing depend on the service.

Official resource: Official: https://ai.google.dev/

10. Agent Skills / OpenJarvis Skill Ecosystem

Why: Once you understand agents, the next step is reusable capabilities. OpenJarvis supports importing skills from ecosystems such as Hermes Agent, OpenClaw and GitHub repositories.

Project: Install one useful skill, inspect what it does, then create or modify a skill for one repetitive task in your own workflow.

Free learning: Yes. Skills and repositories vary in licensing and external-service requirements, so review each one before use.

Official resource: Official: https://github.com/open-jarvis/OpenJarvis

30-Day AI Level-Up Plan

Days 1–5 — Level 1 cleanup

Stop using AI only for questions. Practice structured prompts, context, constraints, examples and output formats.

Days 6–10 — First assistant

Build one small assistant that can use a tool. Example: research a topic and save the result to a file.

Days 11–15 — MCP + tools

Learn how an AI connects to external capabilities. Build or use one MCP-based integration.

Days 16–20 — Coding agent

Use Claude Code or Codex on a real small project. Let it inspect files, implement a feature and run tests.

Days 21–25 — Agent workflow

Build a multi-step workflow with n8n, OpenAI Agents SDK or OpenJarvis.

Days 26–30 — Autonomous system

Add memory, scheduled execution, approvals and error handling. Turn one repetitive weekly task into an agent workflow.

5 Projects That Actually Teach You AI Agents

Personal Research Agent

Give it a question → search → collect sources → summarize → produce a cited report.

Email Triage Assistant

Read authorized emails → classify priority → summarize → prepare draft replies for approval.

SaaS Development Agent

Give it a GitHub repository → inspect code → create a task plan → implement → test.

Content Research Agent

Give it a niche → find high-performing topics → analyze patterns → generate a content brief.

Weekly Business Assistant

Collect selected business data → compare the week → identify problems → prepare a weekly report.

Important: What 'Free' Really Means

The resources in this document are free to learn from or publicly accessible.
Cloud AI APIs are usually usage-based, so running many requests can cost money.
Local models can reduce API costs but require sufficient CPU/GPU/RAM and storage.
Some third-party integrations have their own pricing or rate limits.
Always check the current pricing and usage terms before building a production workflow.

The Goal

Do not chase every new AI tool. The real level-up is learning how to move from asking an AI for an answer to giving an AI a goal, tools, context and clear boundaries so it can complete useful work.

If you can build one reliable workflow that saves you several hours every week, you are already learning the most important part of agentic AI.