Markdown Is the New Coding Language: What Non-Developers Need to Know

A few weeks ago, Garry Tan posted a collection of files to GitHub. Within days, over 26,000 developers had starred the repository and more than 3,000 had forked it. Tech Twitter erupted. Some people called it "God mode for development." Others dismissed it entirely.

The criticism? "It's just a bunch of text files."

They were right. And that is exactly why it matters.

Tan is the President and CEO of Y Combinator, one of the most influential startup accelerators in the world. The company behind Coinbase, Instacart, Rippling, and thousands of other startups. He is not a random developer posting prompts. He is a long-time builder who cofounded the blogging platform Posterous, was one of the first engineers at Palantir, and built Bookface, YC's internal social network.

And right now, by his own account, he is shipping more code than at any point in his career. Not by typing it himself, but by writing structured plain text and letting AI do the rest.

What Garry Tan Actually Built

The project is called gstack. At its core, it is a set of Markdown files that turn an AI coding tool called Claude Code into what Tan describes as a virtual engineering team. Each file defines a specific role with detailed instructions, checklists, and workflows, all written entirely in plain text.

The roles cover the entire software development lifecycle. A /plan-ceo-review skill puts AI into founder mode, challenging the scope and strategy of a feature. A /plan-eng-review skill locks down architecture, data flow, and edge cases. A /qa skill opens a real browser, clicks through the application, finds bugs, and writes regression tests. A /ship skill runs the test suite, pushes code, and opens a pull request.

The results, at least by Tan's account, are significant. He claims to have written over 600,000 lines of production code in roughly 60 days, producing between 10,000 and 20,000 usable lines per day, all while continuing his duties as CEO. Whether you take those numbers at face value or with healthy scepticism, the underlying approach is worth paying attention to.

What makes gstack interesting is not the output numbers. It is the input. Tan is not writing Python or JavaScript. He is writing Markdown and letting AI translate that structured text into working software.

So What Is Markdown, Exactly?

If you have spent any time in tech-adjacent spaces, you have probably encountered Markdown without realising it. It was created in 2004 by John Gruber as a way to write formatted text using simple, readable symbols. The idea was elegant: instead of clicking toolbar buttons or learning HTML tags, you could type a few characters and get headings, bold text, lists, links, and more.

Here is what it looks like in practice. To create a heading, you type a hash symbol followed by your text: # This is a heading. For bold text, you wrap the words in double asterisks: **this becomes bold**. A bullet list is just a dash before each item. A numbered list is just numbers followed by periods. A link is the text in square brackets followed by the URL in parentheses. (If you want a quick reference, the cheatsheet at the top of this article covers everything you need to get started.)

That is genuinely all there is to the basics. No software to install, no formatting menus to navigate, no learning curve worth worrying about. If you can type, you can write Markdown.

And here is the thing: you probably already do. When you bold text in Slack using asterisks, that is Markdown. When you format a page in Notion, Markdown is running under the hood. When you have ever looked at a README file on GitHub, you have been reading rendered Markdown. It has quietly become the standard formatting language of the modern internet, used by developers, writers, documentation teams, and now, increasingly, by anyone who works with AI.

Why Markdown Suddenly Matters More Than Ever

Markdown has been a useful skill for two decades. But the rise of AI tools, particularly large language models like Claude, ChatGPT, and Copilot, has elevated it from "nice to know" to something approaching essential.

The reason is straightforward. These AI models process and generate text. When you interact with them, you are having a text-based conversation. And just like communicating with a human colleague, how you structure your instructions dramatically affects the quality of what you get back.

Consider the difference between telling an AI "help me plan a product launch" versus giving it a prompt structured with clear headings for each phase, numbered steps for the sequence of actions, bullet points for the key deliverables, and a checklist for the review criteria. The second version is not just more organised for your own thinking. It is more parseable for the AI. It can see the hierarchy, understand the sequence, and respect the boundaries you have established.

This is where Markdown becomes powerful. Its syntax is essentially a way of encoding structure into plain text. Headings convey importance and hierarchy. Numbered lists convey sequence. Bullet points convey parallel items. Code blocks convey technical specificity. Blockquotes convey referenced information. All of this is information the AI can read and respond to with far more precision than an unstructured wall of text.

How gstack Puts This Into Practice

Garry Tan's gstack is this principle taken to its logical extreme. Each of his Markdown files is a structured prompt that defines a role, a mission, a set of steps, rules to follow, and quality checks to apply.

Here is a simplified example of what one of those files looks like. The QA Lead role opens with a heading declaring the role. Below that, a mission section describes what the AI should focus on. A steps section provides a numbered sequence of actions: open the staging URL, click through every user flow, log each bug, write a regression test for each fix. A rules section provides constraints: never skip a failing test, screenshot every bug, one commit per fix.

It is not magic. It is well-structured communication, formatted in Markdown so the AI can parse it cleanly. And the fact that it works as well as it apparently does should make everyone, not just developers, sit up and pay attention.

As Nick Hodges wrote in his InfoWorld analysis of the gstack phenomenon, all the code developers carefully craft by hand is also "just a bunch of text files posted on GitHub." Docker files, JSON, YAML. All text. The medium was never the point. What changed is who gets to participate and at what level of abstraction.

The Abstraction Ladder

Hodges made an observation in his article that is worth expanding on. The entire history of computing has been a series of abstraction layers designed to let humans express intent with less and less technical effort.

We started by flipping mechanical switches. That was tedious. So we developed binary code. That was tedious. So we created assembly language. That was tedious. So we built high-level languages like Java and Python. But even those were tedious for the vast majority of people who are not professional software engineers.

Markdown, in this framing, is the next rung on that ladder. You do not need to understand memory allocation or syntax rules or compilation processes. You write structured text that describes what you want, and AI translates it into working output, whether that is code, a document, an analysis, or a workflow.

This does not mean traditional programming is dead. Far from it. Someone still needs to understand the underlying systems, debug edge cases, and architect complex infrastructure. But for a growing number of tasks, especially the kind of tasks that knowledge workers, founders, managers, and creators deal with daily, Markdown plus AI is becoming a surprisingly capable combination.

The Debate: Revolutionary or Overhyped?

Not everyone is convinced. The reaction to gstack has been polarised, and the debate is worth acknowledging because it reveals something important about where we are in the AI adoption curve.

YouTuber Mo Bitar released a video calling Tan "delusional" and suggesting he had succumbed to the sycophancy of AI. On Product Hunt, Sherveen Mashayekhi of Free Agency argued that gstack's prominence was a function of Tan's position at YC rather than the tool's inherent merit.

The sceptics raise fair points. Plenty of developers have built similar prompt collections privately. Lines of code is a famously unreliable productivity metric. And there is a legitimate question about whether structured prompts are "coding" in any meaningful sense.

But dismissing gstack as "just prompts" misses the forest for the trees. The same criticism could be levelled at any abstraction layer when it first appeared. Assembly was "just mnemonics for machine code." Python was "just a scripting language." The value was never in the medium itself. It was in what the medium enabled.

When TechCrunch asked Claude, ChatGPT, and Gemini to evaluate gstack, all three gave positive assessments. ChatGPT noted that the real insight was that AI coding works best when you simulate an engineering organisational structure, rather than simply asking it to build a feature. That observation, regardless of what you think about Garry Tan personally, is worth taking seriously.

Beyond Coding: Markdown as a Thinking Tool

Here is where the conversation gets most interesting for people who do not consider themselves technical. Markdown is not just a way to talk to AI more effectively. It is a way to structure your own thinking.

When you write a Markdown document, even a simple one, you are forced to make decisions about hierarchy. What is the main heading? What are the subsections? What is a sequential process versus a collection of parallel ideas? What deserves emphasis and what is supporting detail?

These are the same decisions that make any communication clearer, whether it is a strategy document, an email to your team, a project brief, or a set of operating procedures. Markdown does not add complexity to this process. If anything, it strips away the distractions of formatting toolbars and design choices and forces you to focus on the structure of your ideas.

I have started encouraging the teams I work with to write their SOPs, their internal documentation, and even their AI prompts in Markdown. Not because the syntax matters in itself, but because the discipline of structured writing translates directly into clearer communication, with both humans and machines.

Getting Started

If you have read this far and you are curious, the good news is that the barrier to entry is essentially zero.

The core syntax fits on an index card. Headings are hash symbols. Bold is double asterisks. Italic is single asterisks. Lists are dashes or numbers. Links are square brackets and parentheses. That covers about 80% of what you will ever need.

To practice, you can open any plain text editor and start writing. If you want to see your Markdown rendered in real time, tools like Dillinger offer a side-by-side view where you type on the left and see the formatted output on the right. The Markdown Guide website provides a complete reference if you want to go deeper. And if you want to see what Markdown looks like in the context of AI-powered workflows, Garry Tan's gstack repository on GitHub is a fascinating case study.

You do not need to be a developer to read the files. They are plain text, after all. That is the whole point.

The Question Is Not Whether

The debate around gstack, whether it is "just text files" or something genuinely transformative, misses the more important point. The medium has always been text. Every programming language, every configuration file, every infrastructure-as-code template is "just text." What changes is the level of abstraction and who gets to participate.

Markdown is lowering the barrier to participation in a meaningful way. It is not going to replace software engineering. But it is giving a much broader group of people, founders, operators, content creators, managers, anyone who works with AI tools, a practical way to communicate more effectively with the systems that are increasingly shaping how work gets done.

The question is not whether you will learn Markdown. It is whether you will learn it now, while the advantage is still fresh, or later, when everyone else already has.

It takes about ten minutes. And it might be the most valuable ten minutes you spend this month.