AI Chat Helpers Can Send Cards as They Work

Vercel has added a way for AI chat helpers to send cards. A card is a message with parts like buttons or tables. The tool is called json-render and works with Chat SDK agents. These helpers can use one shared list of card parts. They do not need a different tool for every message shape.

The helper describes the card and the system builds its message. It then sends the card in the chat service's own format. The same approach works with Slack, Teams and other supported services. A card can include tables, charts, buttons and other parts. The shared list tells the helper which parts it can use.

A card can appear before the helper has finished its work. The system updates that message as more parts become ready. By default, updates happen no more than once every 500 ms. When the work ends, readers see the final checked card. This lets the same message change while the helper builds it.

The system also checks whether each card part follows its rules. If something is wrong, it gives a readable list of problems. Developers can send that list back to the AI for changes. Chat services also limit what one message can hold. In Slack, a second table becomes a block of code. Developers can choose to reject cards that go beyond these limits.