What is JSON?
JSON, short for JavaScript Object Notation, is a plain-text format for representing data as objects made of named fields, where each value can be text, a number, true or false, a list, or another object. A retail order might appear as an object with an order number, a customer object holding name and email, and a list of line items with product code, quantity, and price.
JSON has largely replaced Extensible Markup Language (XML) as the default format for web APIs because it is compact and maps directly onto the structures programmers already use. It is the standard example of semi-structured data: records carry their own field names, but nothing enforces a table schema unless one is added through JSON Schema or a data contract.
Why JSON matters
In AI systems JSON is everywhere. Models call tools by emitting a JSON object of arguments, tool results come back as JSON, and structured outputs force a model to answer in a declared shape so downstream code can rely on it. When an agent extracts the vendor, total, and due date from an invoice, the product is a small JSON record that the accounts payable system can load directly.
Two cautions apply. Models sometimes produce JSON that is almost valid, with a trailing comma or an unquoted value, so production code validates every response against a schema. And JSON has no native date or decimal type, which causes rounding and time-zone defects in financial data unless formats are agreed up front.