The grid canvas
for AI agents.

For thirty years, structured data had two destinations: a database a human queried, or a spreadsheet a human edited. In 2026 a third actor showed up and started producing more rows per hour than either of them — agents.

Agents scrape, enrich, classify, summarise, label, score, and rank. The output is structured. The destination, historically, has been a markdown table glued into the bottom of a chat message. That doesn't scale, doesn't persist, doesn't reuse, and doesn't compose. Building yet another database for it doesn't either — agents need a target that's as fast to write to as console.log and as durable as a permalink.

Instadash is that target. One push, one URL, schema inferred, versioned forever, queryable by the agent that made it and by every agent that comes after.

A markdown table with eighty rows is a bad experience for everyone in the room. The human can't filter, sort, or copy it cleanly. The agent burned half its context window painting cells. The data is dead the moment the conversation ends.

An agent that needs to show structured data should publish it and return a link. The reader gets a real grid — search, filter, inspect, edit. The agent gets its context back. The artifact survives the session.

Every row of data an agent carries in its context is a row it pays for on every turn. Five hundred rows of competitor research is fifteen thousand tokens that get re-billed at every reasoning step until the conversation ends. The same rows, sitting on a grid URL, are zero tokens — fetched on demand, paginated, schema-summarised.

instadash_push takes structured output out of the prompt. instadash_read puts only the rows the agent needs back in. That's an order of magnitude of token spend on any non-trivial agent workflow.

Most useful agent workflows have a human in the loop. The agent finds two hundred prospects; a human picks the forty real ones. The agent extracts claims from a hundred PDFs; a human verifies the dozen that matter. That review needs a real surface — checkboxes, edits, comments — not a re-prompt-the-agent loop.

A grid is the bridge. The agent writes, the human reads and edits, the agent reads back what changed via instadash_edits — and proceeds with confidence about what was approved. The loop closes in two MCP calls instead of two thousand tokens of clarifying chat.