Most AI products treat a book as material to compress. Upload a file, ask a question, receive a summary. The answer may be useful, but the encounter with the book is gone: the cadence, examples, hesitation, and exact argument have been replaced by the model’s version of them.
Coffeetable takes the opposite position. When a question touches several books, it finds the pages that matter and arranges those pages into one short reading. The authors keep their words. The model chooses; it does not impersonate.
Summary is substitution. Coffeetable is curation.
What compression removes
A summary preserves conclusions and discards the route taken to reach them. That is sometimes exactly what you want. If you need to decide whether a report matters, compression is efficient.
Books are different. Their value is often in the route: the story that makes an abstraction concrete, the sentence that changes the emotional temperature, the accumulation of examples that earns a conclusion. Reduce that to five bullets and you retain information while losing the experience that made the information persuasive.
The internet already has enough paraphrases of books. Coffeetable should not add another layer between reader and author.
The product is a reading
A person rarely needs an entire shelf at once. They have a question: how to find the first believers, why a strategy feels incoherent, what discipline looks like when motivation disappears.
Coffeetable turns that question into a small sequence of passages. One book names the problem. Another supplies a counterexample. A third offers language that makes the whole thing click. The result is closer to a mixtape than a report.
- Passages remain verbatim.
- Book and author attribution stay visible.
- The order creates a line of thought across sources.
- The reading is short enough to finish inside the conversation.
Selection, not imitation
The model still does important work. It interprets the question, searches the available library, chooses relevant passages, and sequences them. But it should be judged as an editor, not as a ghostwriter.
That changes the quality bar. A fluent paragraph is easy. A strong selection requires understanding why this passage belongs, what the next passage adds, and whether the whole reading has movement rather than repetition.
It also makes errors more legible. A bad selection can be traced to a source and replaced. A synthetic answer may sound confident while hiding where its claims came from.
The limits matter
Coffeetable is intended for public-domain and otherwise permitted material. Not every requested title will be available. A passage can be relevant without being sufficient. A five-minute reading is an invitation into books, not a replacement for them.
Those constraints are not defects to disguise. They keep the product honest about what it is: a way to find and read the right pages at the moment a question is alive.
ChatGPT is good at answering. Coffeetable is for the moments when you would rather read.