How to build new formats without knowing how to code

Every digital newsroom keeps a list of ideas that never get made. Hardly any of them are bad; what they have in common is that somebody has to code them, and coding costs money, costs tender documents and costs an amount of time that almost never lines up with the date the idea made sense for.

That calculation has changed, and I found it out building a live tracker of the Three Kings making their way across Extremadura, with a week of conversations with ChatGPT ahead of me, as a journalist who had never written a line of code in his life. More than 16,000 people followed it.

What opens up

The process now looks a great deal more like a conversation than like a development project, and a newsroom can hold its own in a conversation: you ask for something, a version that almost works comes back, you point at what breaks, you ask again. You iterate until the tool does what you had in your head, and you do it with the people already inside, who happen to be the ones who know what the audience cares about on that particular night.

That changes which ideas are worth attempting. A format for a single night, a special that only makes sense for one week, a map for an election, a small tool for a local festival: things that would never justify a public tender and that now take a few days.

What it costs

It is not magic and it does not come free either. The first versions were a mess, there were whole afternoons going round and round the same error and moments of thinking the thing would never make it to 5 January. What you save in budget you pay in hours of conversation and in learning to ask for things precisely, which is a new skill and one nobody has on day one.

What stays the same

You still need to know what you want, who you are doing it for and why it is worth the trouble, which is exactly the job it always was. The model hands you options, not judgement. And the decision that really made the tracker work, understanding that the only thing that mattered that night was answering where they are right now, came out of no conversation with a machine.

That part, deciding what gets built and what gets dropped, nobody has automated yet.

What comes next

Understanding how to iterate with a model is going to be the difference between the newsrooms that catch this train and the ones that watch it go by. And that strikes me as excellent news, because it hands the content people back the chance to invent fresh formats without waiting for the calendar and the budget to agree with each other.

We still have plenty of ideas in the drawer. I am going to start pulling them out, obviously still without knowing how to code.