Generative AI is a chaotic black box.
But then again, so are people.
There's a lot of hype and a lot of cynicism around generative AI, but the only thing anyone knows for sure is that it's found niches where it will stick for the foreseeable future.
We'd like to only write code that interacts with clean, predictable, testable, transparent models. The real world tends to disabuse us of the notion that we can.
We're all going to have to work with AI and interact with it on some level either as producers or consumers.
So how do we ship products when we have these chaotic black boxes that we're depending on?
The same way that we've always shipped products that must interact with the real world:
- We approximate and iterate on our approaches
- We add constraints and provide actionable feedback when they're broken
- We manage expectations on all sides
- We test test test test test test test test with real users, who always surprise us with their ingenuity to get into trouble
Whether you think AI will take all our jobs or not, we're going to be interacting with these messy, unpredictable, useful entities in our day-to-day for a long time.
Just another day in the office.