AI Code Guide: a README-only roadmap to AI coding and vibe coding

What AI Code Guide actually is, and who it is written for

The repository contains one file: README.md. There is no package manifest, no source directory, no build script. AI Code Guide is a document, not a program, and the README states its purpose plainly: "Everything you wanted to know about using AI to help you code and/or to code for you." It is written by Vilson Vieira and Eric S. Raymond, with a homepage at aicode.guide and a Discord invite in the header.

The README names two audiences. The first is a working programmer who has not adopted AI code assistants yet; the guide promises recent tools and practices for using them as a copilot, or for acting as the copilot of an AI agent. The second is someone who has never coded and is drawn to vibe coding to build a SaaS or another product; for that reader the guide says it will "remove obscurity" while staying critical about "what is really important and what's just hype." Those are different needs, and one document serving both means the beginner sections will read as obvious to an experienced engineer, while the tooling discussions may assume vocabulary a beginner does not have.

The framing is honest about its own weakness. The introduction says new models ship weekly and that material is "scattered in different places, websites, repos, YouTube videos." The guide positions itself as the aggregation layer. That is a real problem worth solving, but it also means the artifact ages with the ecosystem it describes.

The three terms the guide separates: AI coding, vibe coding, agentic coding

The most useful thing in the README is definitional. It draws a line between three practices that people otherwise use interchangeably.

AI coding is any use of AI models, especially LLMs, to help generate code. The README places the practice in a longer history, noting that code generation research dates back to the 1950s with Lisp and that neurosymbolic hybrid approaches are appearing now. The README gives two concrete examples of the same label: tab-completing in Cursor, and running an agent in Codex's agent mode. Generally, the README says, this group already knows how to code.

Vibe coding is described as AI coding "cranked up," where you do not care much about the generated code, you give a prompt and expect the AI to write everything. The README credits the term to Karpathy in 2025 and says it is democratizing coding for people who never considered it.

Agentic coding is the third: running an agent for many rounds in a loop, ideally with a feedback signal such as tests. The README mentions orchestrators, naming GasTown, that run many agents at once with little or no human interaction. Note how thin that is. GasTown is named once, with no link, no description of how it schedules agents, and no example. If you came for agent orchestration, this section tells you the category exists and stops there.

How the guide is structured, and why that structure is the mechanism

There is no runtime here, so the architecture is editorial. The README says the guide is organized in a "FAQ-ish way" so readers can search and jump to answers. Every section carries a Resources list, and the README states that the most recent resources appear at the top of each list. That ordering convention is the actual update mechanism: freshness is expressed by position, not by a version number or a changelog.

The contribution path is equally simple. The README asks readers who find something missing to open a PR or an issue, or to share findings in Discord. There is no release history, so there is no way to see what changed between one reading and the next except by diffing the file yourself.

The guide also splits its advice by role, which matters more than it first appears. One mode is AI as copilot, where you use models to augment your own work, for example brainstorming a product idea in ChatGPT or autocompleting docstrings in Cursor. The other is AI as pilot, where you are the copilot. The README gives a concrete instance of the second: running Claude Code with the --dangerously-skip-permissions flag and trusting what the agent does. That flag name is doing a lot of work in the sentence. The guide presents the mode without pretending the risk is small, which is the right tone for a document that cannot enforce anything.

Reading AI Code Guide: there is nothing to install

The repository has no package to fetch and no binary to run, so the honest first step is to read the file. The README does not give install steps, because there is nothing to install. What it does give is a reading method: it is organized in a FAQ-ish way and invites you to search and jump to the answers you need. Clone it if you want a local copy, then search the headings rather than reading start to finish.

git clone https://github.com/automata/aicodeguide.git

After the clone you will find README.md at the repository root and nothing else. That is the whole project. If you expected a course scaffold, sample projects or a CLI, this is where you find out otherwise.

From there, open the section that matches your situation: the definitions section if you are still deciding between copilot and agent workflows, or a Resources list if you want outside reading. The guide's own recommendation for staying current is to check the top of each Resources list first, since that is where the newest entries are placed.

Where the guide stops being useful

The first limitation is the one the repository layout already told you: it is a single Markdown file. There is no versioning of advice, no changelog, and no way to know whether a claim about a tool reflects that tool's current behaviour. A guide that says new models ship every week is, by its own admission, racing its subject.

The second is depth. The agentic coding discussion names orchestrators but does not explain how to configure one, what feedback loop to build, or how to keep multiple agents from conflicting. The README's own framing of the pilot mode, running an agent with permissions skipped, describes the practice without giving a safety net. For a reader who has never coded, that is the most dangerous paragraph in the document, and it is one sentence long.

The third is that it is the wrong tool for some jobs. If you need to evaluate a specific assistant against your codebase, this guide will not do it. If you need a structured curriculum with exercises and grading, a FAQ-shaped document is the wrong shape. And if you need something that runs, nothing here runs. The README is candid that it is a "humble attempt" to collect practices, which is the correct expectation to bring.

How it compares with a conventional course or a tool's own docs

The natural alternative is a vendor's documentation, for example the docs for Cursor or Claude Code. The difference in approach is stark. Vendor documentation is versioned against a release, describes one tool's flags and configuration, and is maintained by the people who can change the behaviour. AI Code Guide is tool-agnostic, cites third-party writing, and cannot verify anything it links to. If you already know which assistant you will use, its own docs will answer more of your questions with less ambiguity.

The other alternative is a traditional programming course. A course sequences material, sets exercises, and assumes you will be assessed. AI Code Guide sequences nothing; it is built for search and jumping, and the README says so. The trade-off is deliberate: a course teaches you to write code, while this guide assumes code generation is happening and focuses on the practices around it. It also means the guide has no way to tell whether you understood anything.

What the guide offers that neither alternative does is the vocabulary layer. The distinction between AI coding, vibe coding and agentic coding, with the note that agentic coding implies a loop and a feedback signal, is a clearer framing than most tool documentation provides. That framing is portable across tools, which is exactly why it cannot be specific about any of them.

Maintenance, licence and what to verify before depending on it

The repository is not archived, and the last push was on 2026-03-23. That is roughly six months before today, so treat the resource lists as a snapshot rather than a live feed. The guide's own contribution model, PRs, issues and Discord, is the only update channel; there is no release process to subscribe to.

No licence file is present in the repository, and the README does not state one. Absent an explicit licence, the default copyright position applies to the text, which affects reuse: quoting a line with attribution is one thing, republishing the guide or folding it into training material is another. If you plan to reuse it, that question needs an answer from the maintainers, not from this article.

The upgrade cost is unusual for a project with no dependencies: there is nothing to upgrade, only to re-read. The practical check is to diff your local copy against the main branch before citing it, since the only signal of change is the file itself. Given that the README places the newest resources at the top of each list, a quick look at those positions tells you whether anything moved since your last read.

Editorial conclusion

Adopt AI Code Guide if you want a curated starting point for AI-assisted coding and you accept that the repository holds only README.md, with no installable artifact. Skip it if you need a runnable tool, a course with exercises, or an API. Before relying on it, open README.md on the main branch and check whether the resources list still points at material you trust, because the last push was on 2026-03-23 and the guide itself says AI changes daily.

Frequently asked questions

Official sources

1. automata/aicodeguide on GitHub
2. Issues
3. Project website
4. README