Thinkfork
Why we built Thinkfork
Picture a task you do all the time: asking an AI to help you understand a book. You ask for the key ideas of Thinking, Fast and Slow, and the reply gives you five of them:
- Two systems of thinking
- Heuristics and biases
- Prospect theory
- Overconfidence
- The experiencing self and the remembering self
You are curious about the first one, so you ask about it. The answer has its own sub-points, you pick one, and a few rounds later you are deep in a detail. Then you remember: what was that third idea, prospect theory? It is somewhere far up the chat. You scroll, you skim, you lose your place, and often you just give up. The other four ideas are gone, and you never see the whole book in one view.
The same thing happens when an agent does real work for you. Ask it to plan a software release and the answer has several decisions at once: the version scope, the rollout plan, the pricing change, the announcement. You dive into the rollout, then into one risk inside it, and by the time you are done the other decisions have scrolled out of sight.
Almost every serious conversation looks like this: the answer is a set of points, not a single line, and we follow only one or two of them at a time. A chat window is a single line, so everything we did not follow is buried.
Thinkfork fixes exactly this. Every reply is a card on a canvas, and every follow-up grows a branch from the point you chose. The five ideas stay side by side as five branches. When you want prospect theory, you click that spot and ask there, exactly where you left it, and the earlier answers stay as they were. When you finish, the tree is a map of everything you explored, instead of a long scroll to re-read.
