The 40,000-star Claude Code playbook from a Vietnamese engineer in Paris: harnessing AI fully

Claude Code has become a familiar tool for many programmers, but as a post now circulating in the Vietnamese tech community puts it: most users tap only a fraction of its power. Behind that claim is a GitHub playbook compiled by a Vietnamese engineer working in Paris, which has drawn around 40,000 stars according to the post — a figure rarely seen for a how-to document. This article reviews the core groups of tips every software team should know.

What a 40,000-star playbook says

On GitHub, star counts reflect how much the community trusts a resource. A technical guide reaching tens of thousands of stars means it hits a shared "pain point": many people install Claude Code, can use it, but don't know how to use it properly. The author's catchphrase, as quoted in the original post, is blunt enough to spark debate — the gist being that 90% of the tool's power remains out of reach for most users.

Worth noting: the author is a Vietnamese engineer working in Paris — evidence that Vietnamese experts around the world are actively contributing to the industry's collective knowledge.

Why most users exploit only a fraction of the power

The most common mistake is using Claude Code like an ordinary chat window: open a session, paste a request, receive code, paste it into the editor. This usage ignores what the tool actually is — an assistant that runs right inside your terminal, can read the entire codebase, run commands, read and write files, and work with git. By only doing "ask and answer", you shrink Claude Code into an ordinary chatbot.

The second cause is missing context. AI only works well when it understands your project: the folder structure, naming conventions, which parts of the code are "no-go zones" that must not be touched. New users rarely invest in setting up these context layers from the start.

Core groups of tips worth applying right away

Load context the right way

• Create a project configuration file (such as CLAUDE.md) in the root directory, briefly describing: what the project does, the folder structure, code conventions, and the build and test commands you use.
• Whenever you switch to a completely new task, clear the old conversation instead of dragging out one chat session full of no-longer-relevant context. Short, focused sessions always produce better results than long, diluted ones.

Plan before writing code

For large tasks, ask the AI to propose an implementation plan before writing any code: which files need changing, in what order, and what risks might arise. Review the plan, adjust it, and only then allow code to be written. This helps you catch a wrong direction early — before hundreds of lines of code have been generated.

Break tasks down and cross-check

• One session should address one problem area. Large tasks should be split into small steps, each independently verifiable.
• After each step, run the tests or have the AI itself review the changes against the original requirement. Commit to git frequently so you always have a safe restore point.

Extend its power with custom commands, hooks, and MCP

Claude Code lets you define custom commands for tasks your team repeats — for example, a standard command for code review or for writing commit messages per your internal convention. The hook system can automatically run checks whenever a file changes. The MCP protocol connects the tool to external services such as databases, ticketing systems, and internal documentation — turning the AI assistant into a genuine member of the workflow.

Automating beyond the terminal

For steps that repeat on a schedule or on an event, Claude Code can run non-interactively inside scripts and CI pipelines. That is the path to building automated workflows: daily report roll-ups, pull request reviews, changelog documentation generation.

A perspective for Vietnamese businesses

For development teams in Vietnam, the greatest value of using Claude Code correctly lies not in writing code faster, but in standardizing the process:

• Faster onboarding: a good project configuration file helps a new engineer (and the AI) grasp the context in minutes instead of days of reading code.
• Less individual dependency: conventions are written down as living documentation instead of living in the heads of the one or two people who hold the code.
• Cost optimization: selective context and short working sessions save a significant number of tokens — a direct cost as usage scales up.

The prerequisite is still technical discipline: clear code, thorough tests, clean git history. AI multiplies the strengths of a good team — and it also multiplies the chaos of an undisciplined one.

Conclusion

The tens-of-thousands-of-stars playbook by a Vietnamese engineer in Paris confirms a clear trend: the gap between an average AI user and a great one is not the tool, but how context and workflow are organized. Those who invest in their setup early are the ones who pull ahead.

Source: compiled from technical discussions shared on Facebook, edited by Wotech Electronics.

If your business in Bac Ninh and surrounding areas wants to bring AI into its software development process in a structured way, the Wotech Electronics team is ready to advise on a suitable roadmap.