MicrosoftIt seems highly contradictory. The current consensus among AI thought leaders and KOLs is to educate non-developers on the importance of using native Markdown and HTML for input and output, ensuring everything is inherently "agent-readable."
However, Microsoft’s workflow seems to take the opposite approach: users are expected to package information into PDF or Word formats for Copilot to read. The system then parses these documents into Markdown, processes them, writes the output back to Markdown, and finally uses internal scripts to convert them back into Word or PDF files. This cyclical conversion process results in a significant waste of time and tokens.
It is difficult to understand why Copilot Notebooks do not natively accept Markdown (.md) and plain text (.txt) files as direct input references.
