Garbage In, Garbage Out

I try to put as much code as I can on Github, as an exercise in learning in public. Over the holiday I published something new: a Lambda function deployed with CDK that is called when events fire from Buy Me A Coffee. It’s notable for a couple reasons:

• First, if you haven’t noticed, there’s now a little orange icon down in the lower right of this blog. If you’ve benefited from my writings here, and are feeling generous, you can buy me a coffee using that icon. No pressure, though!
• Second, I wanted a permanent reference repo for doing a basic Lambda deployment with CDK, and now I have one.
• Finally, and most interestingly, my first draft of this code (both the CDK and the function itself) was written by ChatGPT.

As I’ve been reflecting on what artificial intelligence means for the future software engineering, I doubt it’s going to obviate the need for understanding code. What I do think it will be able to do is produce workable drafts, getting maybe half-way to a workable solution, with developers making revisions to get it across the finish line.

If I’m right about that, software professionals are going to need to tweak their skill sets to be less like writers and more like editors. Few developers I know are particularly adept at reading code, though it’s a shortcut for leveling up coding skill. Software is an oddball in that respect: in most other creative domains, one must study the work of the geniuses who came before, but we don’t typically train coders that way. Now it might not even matter, because the machines can do the studying for us.

What will be important, then? Formulating problems, and then tweaking resultant autogenerated code. So code fluency will still be needed. Just not for the boilerplate; for that, we’ll have tools like ChatGPT and AlphaCode.