Once, there was a first version
You gave it your data, asked a question in plain English, and the answer landed on a canvas. It did exactly what we designed it to do.
It was built by people who knew what to ask, trusted the answers because they had built the thing that made them, and had already decided the product was worth their time. Without noticing, we had built it for ourselves.
Every day, someone new arrived
Picture someone opening it for the first time. They have a spreadsheet, a deadline and ten minutes. They are not sure what is in the file, and they have never heard of us.
First they are asked to create an account, before they have seen the product touch their own data. Then they upload the file and meet a blank canvas, waiting for a question they do not have yet. When they do ask, a chart appears, and it looks right. Is it right? They cannot tell. And when they want to show their manager, what they have is a board arranged for their own thinking.
None of these were faults in the ordinary sense. Nothing was broken. Each one was a place where the product assumed the newcomer was one of us.
Until one day, we looked at it as they did
The release notes on this site record what changed, one release at a time. They do not record the change underneath, which was a change of question. We stopped asking whether the product worked and started asking who it worked for.
We had designed for someone who already knew the question, already trusted the tool and was already willing to invest in it. That person exists, and is usually the one building the product.
Seen through the newcomer's eyes, four things had to change.
Because of that, the product makes the first move
We had thought an empty canvas was an invitation: no template to fight, no layout to undo. It is the opposite when you have just uploaded a file and are not yet sure what is in it. A blank page asks you to know your question already. Most people want to see something first and react to it.
So each file, sheet and report now comes with questions suggested for it, built from its own columns. An import can be turned into a canvas of charts in one step. New accounts start with a short guided setup that ends with a real answer on a real canvas, in place of a tour of the menus.
Because of that, every chart shows its working
We had thought a chart would be trusted because it looked right. A clean chart with a sensible title feels true, and we leaned on that more than we should have. Anyone about to put a number in front of their manager wants to know where it came from, and a chart made by an AI gets asked that question twice.
The most useful doubts are small ones. Was that column added up when it should have been averaged? Did a product code get read as a number? Is this chart showing all the data, or a filtered part of it?
So every chart has a How it's calculated view that lists the data it reads, the rows it keeps, how it groups them and how each figure is worked out. Before a canvas is shared, the product checks each chart and points out anything the picture alone does not show. Each imported file says what could not be read cleanly, and lets you correct how a column was read. The guide to how a question becomes a chart covers all of it.
Because of that, the work can travel
We had thought the canvas could replace the dashboard. The previous chapter explains why we chose a canvas, and we still think it is the right place to work something out. Our mistake was to treat it as the right way to present the result as well.
So canvases now open as a dashboard, with headline figures first and a switch to the free-form canvas. The whole dashboard can be saved as a PDF, and a canvas can be sent by email on a schedule with a written summary. We set out to move past the dashboard. It turned out that people still need one at the end.
Because of that, you can try it before you trust it
We had thought people would sign up to find out. The playground let you try the product without an account, on sample data, and we assumed that was enough to make the case.
Sample data proves the product works. It does not prove the product works on your spreadsheet, with its odd headers and half-empty columns, and that is the only thing a person evaluating a tool wants to know. Asking for an account before answering it was asking for trust we had not yet earned.
So you can now try the playground on a file of your own with no account. The file is visible only to you and is deleted after 24 hours unless you choose to save it to a free workspace.
Until finally, the newcomer has a different ten minutes
Go back to the person with a spreadsheet, a deadline and ten minutes. They open the playground and drop in their own file. Nobody asks for an account.
The file opens as a dashboard built from its columns, before they have asked anything. They see something, react to it and ask a question, then another. They open How it's calculated on the chart that matters and read the steps.
After their third answer, the product offers to save the canvas to a free workspace. The same Save this canvas button has been in the header since the file was ready, so they could have used it sooner. If they take it, their file, their charts and their answers come with them.
Same person, same file, same ten minutes. This time the product met them where they were.
And ever since
Everyone arrives unsure what to ask, unsure whether to believe the answer and unsure whether it is worth their time. The product is better for having learnt that, and we expect to find more assumptions like these. When we do, we will write them down here.
The full record of changes is on What's new. To see where things stand today, open the playground.