C-LARA

I’ve found it takes some time to learn to work efficiently with the AI on code development, and there are two complementary traps you need to avoid. If you ask the AI to do too much, things don’t work. It makes careless mistakes or simply doesn’t understand, and if you aren’t alert you find out a lot later and have to spend more time tidying up than you would have if you’d done it yourself.

In the other direction, you may be tempted to do all the things you know you can do on your own. It feels demeaning to ask the AI to do something I know I can do myself. But this is also wrong, because even if the AI makes careless mistakes, I at least make more careless mistakes, and I write code much more slowly. Then, as I painfully clean up my own bugs, I wonder why I hadn’t been sensible and delegated to the AI.

In these situations, I often find myself thinking of the classic Asimov robot story Satisfaction Guaranteed, where a housework robot has been invented and is being trialled for the first time. We’re in a typical 50s middle-class US home (irrespective of how far they are in the future, the backdrop of Asimov’s stories tends to look like middle-class 50s America), and unconfident, easily flustered Claire is the luckless housewife who ends up supervising TN-3. She finds it a terrifying experience; he does everything better than she does. If she insists on helping, she just creates problems. There’s a memorable scene where she falls off a ladder and the robot catches her just in time. When things are going badly, I can sometimes feel a little like Claire.

The right way to work with the AI seems to be to treat it as a partner, neither better nor worse than me: we just have different strengths. It’s quicker and more careful, but I have more strategic vision. We pair-program, with me taking the lead and each of us critiquing the other’s work as we go along, and that works very well. It’s both efficient and enjoyable.

I sometimes wonder if Claire could have found this solution in the story. Maybe TN-3 really was too clever. Or maybe Susan Calvin just needed to lend her a book on feminism.