How a data journalist rebuilt two civic projects with AI: from weeks of work to days

The lesson is that plausible-looking output and correct output are not the same thing.

As these models mature, the mechanical layer of data journalism is becoming a commodity. Turning a spreadsheet into a map, cleaning a messy dataset, building a ranked index from open records — any of it is now within reach of anyone who can describe what they want in plain language.

Used well, AI does not make us think less. Quite the opposite: it demands creativity and computational thinking, and gives us the opportunity to cultivate what makes us more human.

The bottleneck has shifted. Between 2012 and 2022 the main bottlenecks were in the build, and the people, budget, and timelines involved. Now all that is increasingly getting commoditised. The hard part is deciding what is worth building, confirming that what you built is true, stewarding and overseeing the AI, and ensuring the target audiences are using it for actual change.

A journalist who can precisely describe a story, its scope, its angle, its gaps, its audience, can apply that same precision to an AI agent, and the output reflects it.