A few years ago, watching an AI write three tidy paragraphs felt like the future. Today it hands you three pages on demand, and what you feel is closer to dread.
Count the AI summaries someone sent you this week. Now count the ones you actually read. That gap is the whole story. When generating text cost human effort, text carried weight. Now anyone can produce twenty airtight pages before their coffee cools, which means twenty airtight pages nobody asked for are worth roughly what they cost to make. The scarcity flipped sides: writing became free, reading stayed expensive, and every wall of AI prose sends the bill to the reader.
And the harsh version: your beautifully prompted summary? Nobody read that either. They scrolled it hoping a number would jump out, gave up around paragraph three, and asked you in chat. The wall of text is not a deliverable. It is a queue.
Put the two eras side by side:
| The essay era | What comes next |
|---|---|
| Three pages of prose | One picture |
| The number hides in paragraph nine | The number is the first thing you see |
| You discover the AI misunderstood you after reading all of it | A wrong picture looks wrong in one glance |
| Every follow-up question: another essay | Every follow-up question: a fresh picture |
This is not a prompting problem, so no prompt will fix it. It is a format problem. Text buries answers by design: one word after another, the reader doing all the excavation. Formats that show instead of tell put the answer on the surface. People have known this forever, which is why most serious decisions in your company already happen in front of a chart, a diagram, or a dashboard.
The chat window just caught up. In January 2026 the MCP standard added the missing piece: apps that render inside the conversation itself. I built mcp-dashboards on it, a free MCP server that turns AI chats from a place that describes your numbers into a place that shows them. Charts, dashboards, flow diagrams, timelines, drawn in the reply, interactive, and redrawn every time you ask a follow-up.
That is a real chat. I did not name a chart type or open a BI tool, and nobody will read three pages to find the churn number, because it is sitting on the surface where an answer belongs. Ask "what if traffic triples" and the picture redraws in seconds, in front of everyone. I wrote about what that does to a meeting in the previous post: the short version is that arguments stop surviving on ambiguity when the answer is on screen.
Text is not dying everywhere, and I should say that plainly. A contract needs prose. Code needs prose. A hard tradeoff deserves paragraphs, and this post is one of them: an argument rather than a status report. What is over is text as the default answer format for numbers, status, and comparisons, which happens to be most of what working people actually ask an AI. Those questions were never essay questions. We just did not have a screen in the chat until now.
The essay had a good run. Give your AI a screen.
Set it up in one line
One entry under mcpServers in your MCP config file (claude_desktop_config.json):
"mcp-dashboards": { "command": "npx", "args": ["-y", "mcp-dashboards", "--stdio"] }
Restart Claude and type:
Take these numbers [paste yours] and show me the quarter at a glance: revenue by month, top products, conversion, churn. Pick the charts yourself.
The dashboard renders in the reply, and every follow-up gets a fresh one. Runs on your machine, and no third-party chart service ever sees your data. Full setup in the README.
GitHub: github.com/KyuRish/mcp-dashboards · npm: npmjs.com/package/mcp-dashboards
Connect with me on LinkedIn.
Top comments (0)