Back to the newsroom

Op-ed

AI must understand context, not just read documents

Korea Deep LearningPublished

AI must understand context, not just read documents
In an October 2025 contribution to AI Times, Korea Deep Learning pointed out that document automation must go beyond recognizing text to understanding the context of documents and each company's own document grammar.

Kim Ji-hyun, CEO of Korea Deep Learning, contributed "AI must understand context, not just read documents" to AI Times. Kim pointed out that many businesses still think of document automation as pulling out text with OCR or extracting keywords, but that the real value of a document comes from its context.

"What artificial intelligence (AI) needs now is not 'the ability to recognize text' but the ability to interpret context and intent."

— Kim Ji-hyun, CEO, Korea Deep Learning

Same word, different weight

The piece gives examples of meaning that changes with position and use. "Approved" stamped next to a signature line carries legal force, but the same word in the middle of the body text is only a mention. A decision to "cut" a budget also carries completely different implications depending on whether it means reducing marketing costs or adjusting labor costs. Kim's concern is that even if the text is read accurately, extracted data cannot be used in the work if these differences are not told apart.

Documents sit in a flow

Work does not end with a single document. A contract sits in a flow that runs through meeting minutes, review comments, approval history and email, and processing one document alone cannot capture the full context of a decision. Meeting minutes, too, should not stop at summarizing what was said, the piece says, but should make it possible to reconstruct why the meeting was held, what evidence was used, who proposed what, and what follow-up actions were decided.

"In the end, what matters is not the ability to recognize each document on its own but the ability to grasp the context in which that document sits."

Every company has its own document grammar

The piece also points out that every organization uses different forms, terms and practices. One company writes the "total amount" in large type at the top of a contract, another puts it in an attachment or calls it the "sum," and some institutions put key terms in a table footnote or in parentheses. That is why general-purpose SaaS models struggle to keep up with this "document grammar" that differs from company to company. Kim argued that true automation needs company-specific models trained on each organization's real data and workflows.

Documents as units of execution

The piece sees reports, tax invoices, policy proposals and business plans as stores of structured context. When AI truly understands documents, it can take on preparing reports, summarizing key metrics and even managing approval procedures, and people can focus on strategic judgment. The conclusion is that the key is to teach AI the grammar of document forms so that it truly understands documents.

The full piece was published in AI Times.

About Korea Deep Learning

Korea Deep Learning (KDL), founded in 2019, is an enterprise AI agent company. Built on KDL Frontier, its own vision-language model (VLM) specialized for documents, KDL makes AI agents that take on and complete the document work of businesses and public institutions. Its main products are DEEP OCR, DEEP Parser, DEEP Agent and DEEP Ops.

Source