Markdown is drawing attention as a document format optimised for the AI era. With the spread of AI agents, adoption is expanding from private companies to government agencies.

Markdown is a lightweight document markup language created in 2004 by U.S. developers John Gruber and Aaron Swartz. It uses simple symbols such as "#" for headings, "**" for bold and "-" for lists to express document structure. It is saved as a plain text file (.md) and can be opened in any text editor.

It is also widely used in development and AI, including GitHub README files and training data for large language models (LLMs).

Industry participants say the reason markdown is getting more attention recently is its AI-friendliness. For machines, including AI, to interpret natural language, they need a parser suited to the file format. Markdown is a pure text format and does not require a separate parser. It can save resources not only in AI training and processing but also in document exchanges between agents.

In South Korea, markdown is drawing attention for using public data in AI. Most global AI services, including ChatGPT, do not have built-in parsers for HWP and HWPX, which has hindered the use of public data for AI development and services.

The National Artificial Intelligence Strategy Committee said it will write and manage the results of subcommittee meetings and discussions in markdown and disclose them through the committee website, following the outcomes of meetings and debates on March 5. Lim Moon-young (임문영), the committee's standing vice chair, said, "In the AI era, it is important to innovate not only policy content but also the way policies are accumulated and managed." He added, "The transition of the document system will be a starting point for changing the way the government uses AI and its working culture."

Demand is also rising in the private sector alongside the spread of AI agents. In natural-language coding, context engineering — pre-entering background information, rules and context into AI coding agents — determines the quality of output. This process is done by creating markdown files. Major AI coding tools such as Claude Code and Cursor also read markdown files directly and reflect them in code generation.

The markdown-based knowledge management app Obsidian is also gaining ground. Notes written in Obsidian are all saved as markdown files in a local environment. Connecting them to AI coding agents allows immediate use, and user numbers are rising mainly among developers and researchers.

The cloud collaboration tool Notion supports markdown input and output. It does not store files internally as markdown, but it can import or export in markdown format. Notion topped 100 million users worldwide earlier this year. In South Korea, it has secured clients mainly among large companies and IT platforms such as GS, Hyosung, LG AI Research, Today's House and Kakao Style.

A Notion official said, "An era has arrived in which services with a markdown structure that makes it easier for AI to understand context are given more value." The official added, "Even when companies build retrieval-augmented generation (RAG), the reason quality does not come out is ultimately a data structure problem, and organising data based on markdown is the answer."