You never pick a mode
For most of last year, the first thing Mastheads asked you was what kind of thing you wanted.
That is a normal way to build software. You have several jobs the product can do, so you put them behind a chooser and the operator picks one. Nobody complains about it, which is exactly the problem - a menu is the kind of friction that never produces a support ticket. It just quietly costs everyone a decision they did not need to make, several times a day, forever.
What changed my mind was watching people use it. They knew precisely what they wanted written. What they did not know was which of our words described it. Is a list of twenty-five topics a bulk run or a campaign? Is one headline an article or a brief? Those are our categories, not theirs, and asking somebody to translate their intention into our vocabulary before we will listen is a strange way to greet them.
So we took the menu out.
What replaced it
Mastheads opens on one field.
Type a line and it is one article. Press Enter, type a second line, and it becomes a list - one article per row. Delete back down to a single line and it folds again. There are seven ways in now: a topic, a list, a structure you reuse crossed with your own data, a URL you already published, two products to compare, a subject and a count, a term to explain. Every one of them is the same field, reading what you typed.
You can drop a file in instead. It reads the lines out of a .txt, .csv, .xlsx, .docx,
.pdf or .json.
Underneath, nothing forked. Whichever door you came through, one article is one full run of the same nine desks.
That was the real reason to do it, and it is worth saying plainly: a mode menu is not only a question for the operator, it is a promise to maintain several products. Each mode grows its own edge cases, its own quality bar, its own bugs and its own paragraph on the pricing page. One field means there is one thing to get right.
The part I was nervous about
Software that infers what you meant can infer wrong, and being wrong about somebody's intention is worse than asking them. I have used enough tools that guessed helpfully and handed me something I never ordered.
So the field does not act on a guess quietly. It shows you what it decided before anything is
spent. One line reads 1 article. Three lines reads 3 articles, and the run that appears
underneath says IN THIS RUN 3. If that number is not the number you expected, you have not paid
for anything yet.
That is the same rule the rest of the product runs on - report what happened, never infer it and present it as fact - moved one step earlier than usual. An inference you can see is not a guess. It is a proposal.
What it is not
It is not natural language. The field is not parsing your sentence for hidden instructions or forming an opinion about your tone. It is reading the shape of what you typed: how many lines there are, whether they are addresses, whether you dropped a file. That is a small, dull mechanism and I would like it to stay that way, because a small dull mechanism is one you can predict. Predictable is most of what an operator actually wants from a tool they use every day.
It also did not touch the newsroom. Nine desks still go and find real sources, and the editor still fixes what it finds instead of flagging it and walking away. The field is a better way to hand them work. They are still what the work is for.
New accounts get ten articles over seven days, free, with no card. Type one line and see what comes back.