Start your journey for free

You do not need to enrol to start. This is the free path: the same readings, Field Notes chapters and open-source tools that participants in Dr. Tali Režun's programs use, in the order they use them. Work through it at your own pace. By the end you will have a toolkit, a second brain, your own agent, and the foundation documents for a first product built with AI agents.

Free and at your own pace: what to read, install, watch and do, one step at a time.

Steps

12 steps, about 57 min in all. Tick each one off as you go.

  1. Install

    Step 1: Set up your workspace

    Use a laptop where you can install software (on a company laptop, check with IT first). Install Claude Desktop, which includes Claude Code; Claude Pro is the minimum plan for building with coding agents. Create one Google account for NotebookLM, Google AI Studio and Gemini, then free GitHub and Cloudflare accounts with the same address. Install VS Code to read and edit Markdown files.

  2. Read

    Step 2: Understand how large language models work

    Learn what tokens, next-token prediction, weights and the context window are, and why models hallucinate. Then ask Claude to write a one-page cheat sheet in Markdown in your own words, and check that you could defend every line.

  3. Read

    Step 3: Know your tools and where your data goes

    Learn the four ways to use AI (chat, desktop agents, API, local), and classify your data as public, internal or crown jewels before you choose a model. Never put an API key in a chat, a browser page or your code.

  4. Watch12 min

    Step 4: Learn GitHub in plain English

    GitHub is the safety net for agent work: every working version is saved. Learn four words (commit, push, pull, clone) and you can let a coding agent use GitHub for you.

  5. Do45 min

    Step 5: Talk to Field Notes

    Field Notes is ten question-led chapters on context engineering, agent memory, coding agents, orchestration and verification. Download the whole corpus as one Markdown file, add it to a new Claude Project, and ask it questions instead of reading fifty articles. Ask at least five, including how the material applies to your own idea.

  6. Read

    Step 6: Understand what an AI agent is

    An agent is a model with a mandate, memory and tools, running in a harness such as Claude Code, Codex or OpenCode. Read how agents are built, how they connect to tools through the Model Context Protocol (MCP), and how a mandate says what is allowed, never allowed or needs approval.

  7. Install

    Step 7: Build your second brain with The Curator

    The Curator is a free, open-source second brain that runs on your own computer: your documents become a linked knowledge graph that you and your agents read and write through the My Curator MCP. Install the Mac app, or the one-line installer on Windows, and add your own AI key in its setup wizard (a free Gemini API key works), never in a chat. Create a domain and ingest your first files, then ask it a question and check the pages it cites.

  8. Do

    Step 8: Build your own agent: Atlas on Conduit

    Conduit is an open framework for a governed personal agent, set up by pasting prompts. Install OpenCode, create a folder for your agent, and follow the Atlas guide: run the setup prompt, fill in your profile, add a tool, then connect The Curator through the My Curator MCP. Test it: Atlas should answer a question from your own second brain and list the pages it used.

  9. Read

    Step 9: Learn context engineering

    The master skill. Learn the three kinds of context (foundations, working state and knowledge), why a fuller context window gives worse results, and how to hand work over to a fresh session at about 80% of the window.

  10. Do

    Step 11: Pick a problem and write your foundations

    Choose one problem from your own work, right-sized: not an AI for the whole sales team, but an assistant that answers your twenty most common customer questions. Pick a website with a support agent, a custom agent, or your own app. Ask Claude to interview you about it, then write CONCEPT.md (what, for whom, why, three must-have capabilities) and the files a coding agent needs. Test them: a fresh AI session should be able to summarise the build from your files alone.

  11. Do

    Step 12: Build, verify and ship

    Open your project folder in Claude Code and hand it your foundations in plan mode. Approve the plan before any code is written, then build in small phases: build, test, report, fix, commit. Check work you cannot read with tests, a second model's review and another person. Deploy to a free workers.dev address on Cloudflare Workers.

Want guidance?

Join a program to learn with a group, taught by Tali.

See the programs