Build your own Claude code tutorial

Code With Antonio’s recent YouTube tutorial is part build-along, part teardown of how a modern coding agent actually works under the hood. The video, “Build Your Own Claude Code | Full AI Coding Agent Tutorial,” says it builds “our own AI coding agent inspired by tools like Claude Code completely from scratch,” framing the project as a way to learn agent workflows rather than just consume a packaged product. A second video, “Pi: Open-Source AI Agent Terminal Set-Up,” makes the same case from the open-source side. Its description says Pi Agent is “a minimal open-source AI agent that runs directly in the terminal” and walks through installation, model setup, project instructions, session management and “the safety trade-offs of giving an agent direct CLI access.” ### What are these videos actually teaching? (youtube.com) The Antonio tutorial is centered on building an “AI-powered coding CLI” rather than prompting a chatbot in an editor. Antonio’s course page says the workflow is one where “AI agents write, test, and ship production code” while the human “architect[s], steer[s], and review[s],” and says the build covers an agent that can read, write and modify code. (youtube.com) The Pi video is more operational. Its description says the setup is designed to work with “the tools, files, and repositories you already use,” and demonstrates workflows across scripts, Terraform, Ansible, Docker and Markdown. ### Why does the terminal keep showing up? Pi’s own site describes the project as “a minimal terminal coding harness” and says users can adapt it through extensions, skills, prompt templates and themes. (codewithantonio.com) The GitHub repository describes Pi as a “coding agent CLI” with an “agent runtime with tool calling and state management.” That terminal-first design matters because agentic coding work often spans more than code generation. (youtube.com) The Pi walkthrough explicitly includes direct work with repositories, files and command-line tools, while Antonio’s tutorial frames the skill as building from “a blank terminal” with an AI agent as “your only teammate.” ### What skills do the videos put in front of engineers? (pi.dev) The clearest recurring skill is workflow control. Antonio’s course page contrasts older step-by-step tutorials with “a real engineering workflow,” saying the point is not “watch me type every line” but “you steer, agents implement.” The second is tool orchestration. Pi’s video description highlights provider and model setup, `AGENTS.md` project instructions, customization, and session management, which together point to a workflow where the model is only one layer of the system. (youtube.com) The third is repo-level judgment. Both videos are built around agents operating across files, configs and command execution, not single-shot answers in a chat box. (codewithantonio.com) That is an inference from the described features — file access, terminal use, project instructions and code modification — rather than a direct quote from either creator. ### Why pair a Claude-style build with an open-source agent? (youtube.com) The two videos answer different questions. Antonio’s tutorial shows how a Claude-like coding agent can be assembled and steered; Pi shows how an engineer can run a lightweight, open-source version in a terminal with existing tooling. Pi’s repository also points to a broader open-source effort. The GitHub page says the project includes a coding agent CLI, a core runtime and a unified multi-provider API covering Anthropic, OpenAI and Google, among others. (youtube.com) ### Where does this leave someone trying to learn the stack? The practical takeaway from the source material is that these tutorials are less about one branded tool than about learning the moving parts: planning, tool calling, file access, command execution, safety controls and review. (youtube.com) Antonio’s page says the free YouTube tutorial went live on May 18, while Pi’s docs and repository remain available for readers who want to inspect the implementation and run the open-source harness themselves. (github.com) (codewithantonio.com)