The other day, I explored this dataset from the Swedish National Board of Health and Welfare. It's a dataset on external causes of injury and poisoning in children. I used matplotlib to plot the three most common causes in 2001. A tricky aspect was that the dataset contained both specific and aggregated causes. To get a clear picture of the most common causes, I had to remove the aggregated data entries. All in all, it took me about an hour, an hour and a half, to get a nice plot.

Then I asked Claude Code to do the same. And wow. In less than a minute, it spun up a program that listed the most common causes in two decent looking graphs. But the most intereseting part was what happened next. Claude Code not only identified that the dataset contained aggregated causes - it also suggested that it was probably desirable to remove the aggregated entries and look at specific causes.

I must say, I'm impressed.