A Centaur with an Identity Crisis
I finished reading (or, more accurately, listening to) Cory Doctorow's latest book, The Reverse Centaur's Guide to Life After AI.
If you're at all familiar with his blog, or have seen him talk on the subject, nothing in the book will come as a huge surprise. I think what's remarkable about it, tho, is the clarity of the construction. The anecdotes and examples in the book are stories he is often repeating, but they do come together as a coherent, convincing argument about how to confront the AI bubble.
I'd been kind of mulling how to talk about LLMs and related technologies in public for a while now. It's not easy. I'm professionally required to cultivate a pretty sizable amount of enthusiasm for the technology. On the other hand, quite a few friends and online communities talk about it like it's the Great Satan.
I feel much more aligned with Cory's take on things. There is actually something useful here. I hate the billionaire-fueled con job inflating the bubble to utterly preposterous proportions. A data centre boom in the middle of a decidedly laissez-faire regulatory environment means these dumbasses are building out gas turbines instead of being forced to power these things with renewables, which might actually end up helping with the climate emergency. Everything about the bubble economics of this thing is stupid and terrible and is making me incredibly anxious for a variety of reasons.
But there is a there there. This technology is going to have a pretty profound effect on my profession, if nothing else. I'm not going to pretend I can tell you what it will be. This isn't LinkedIn. But it's also interesting.
I've been somewhat fortunate that I've been able to take the centaur role with LLMs. I can use them or (mostly) not use them as I see fit. (My actual work revolves around LLMs, so avoiding them would be a bit silly, but my actual process is mine to develop). I mostly write the code myself, chatting with the coding agent if I get stuck (rubber duck style). Then I get the agent to do a code review and did-I-miss-anything verification with Jira before sending it off to the team to review. And I feel way more productive for it.
I think this is the way to do it. I need to understand the code that has my name on it, and the best way to do that is to write it. Writing code isn't that hard when you've been doing it for a while. I find myself beating my head on the desk a lot less, because the agents are good enough now that they can spot errors I'd have spent hours debugging in the past. So all good, right?
I am, of course, more than a little anxious about where all this is going. As Cory cites Stein's Law, "If something cannot go on forever, it will stop." And this bubble can't go on forever.