A self-induced lobotomy
A few weeks ago, I started reading about Steve Yegge's Gas Town, a fully automated agentic platform so chock full of AI that he could go almost completely hands-free with his software development, while 30+ terminals on his screen coordinated in what sounded to me like a rather slapstick manner, yet supposedly this was the hot new shit that everyone AI-side was going to foam at the mouth over; myself included.
I'm a big old nerd. I absolutely love computing and spend a fair amount of time on my computer, tinkering, building, coding, less gaming than I should; all around having a good time. Like a lot of nerds in the last few years, I've been bitten by the AI bug and have really gone off the deep-end in playing with it. It's fun, it works, it helps my ideas go from just that to a reality really quickly.
Steve's (very funny) Gas Town introduction article triggered something in my brain. With AI up until reading that article, I was keeping myself in the loop on purpose, checking code, really breaking down implementations so that I got what was going on in whatever it was that I was working on at the time. But I read his article and it "inspired" me to just let go and fully let the AI take over: and take over it did.
"My" output went through the roof. Huge features in my coffee cupping app ticked off in minutes. Agents delegating tasks to sub-agents to do a thing with another whatsit, while me, three levels above in delegation-nirvana waited for the hierarchy below to complete whatever slog I'd set out for them. Chop-chop, little robots, I haven't got all day! (I actually do, I don't work on Fridays.)
This was and still is really fun. I took this ethos to work where we are big on AI tools to get shit done. Jira going mad with agents helping groom the backlog, Slack messages getting scanned for emergencies from people's half-heartedly filled pings obviously written before the first coffee was drunk, internal tools getting one-shot by Claude to deliver to our teams. It's incredible.
It's more incredible how hard it is for me to engage my brain. I can feel my bigger ideas starting to get harder to grasp. Ideas I need to get my work done.
There's a lot of chatter online about this delegation of duty and the loss of our need to think, but I'm starting to think it's happening way faster than I'd expected from myself. Cory Doctorow of "Enshittification" fame, wrote a fascinating article on this topic, talking about the concept of "centaurs" and "reverse centaurs". The centaur is an actual person getting assisted by machines (in my context it's all the AI delegation mentioned above), while the reverse centaur is something darker. The reverse centaur is exactly what it sounds like: it's us doing the grunt work.
I was explaining to a colleague how we can make our AI "assistant" in Jira (called Rovo, which when you say with a really Aussie accent makes it clear that this is a terrible name) more effective when people ask it questions that need to have the answer come from internal documentation.
What happens is:
- Human asks Rovo a question in the chat interface
- Rovo interprets the user's request and queries Confluence (internal knowledge base) using natural language
- Hopefully an article we've written matches that search string
- It surfaces that to the human in the chat
So what if there's no matches? Rovo shits the bed, human leaves empty handed, we get a ticket to solve things. Not ideal.
What I was explaining to my colleague was that it's up to us to build that bridge between Rovo and our knowledge. Whatever Rovo (I keep saying it in my head with a really brash west Sydney accent and it's horrible) searched for is what our article should be called. We should adapt our knowledge base in Confluence to conform to it's understanding and interpretation of the initial request.
We should become the reverse centaur.
It's a hard line to tread, because I'm a believer that at work, it's worthwhile doing whatever you need to to get the job done in as efficient a manner as possible, because it's not "your" time, it's your employers. They're paying you to not fuck around. But then where does our professional growth come from when we are the extension of the robot, not the other way around?
I asked Rovo to analyse tickets in the backlog and evaluate them on some criteria that I designed and it did it and the outputs were sound, so should I just go ahead and implicitly trust it's recommendation to de-prioritise that two-month old ticket? After all, I told it how to evaluate it, so didn't I actually do the analysis? Or am I losing nuance because every ticket is slightly different in their context, and if I don't evaluate it myself and just assume 100% correct evaluation every time from the bot, then do we end up missing something critical? Does it even matter because more and more, the requests we get are obviously written by AI?
It's reverse centaurs all the way down on a timescale long enough.
I started this blog because I, me with my actual hands, wanted to craft something that my nephew could read in the future and learn a bit more about who I was and am. I love history and it feels like my little slice of it, something to look back on, as I am rather sentimental. This might end up being one of the last refuges I have left, especially if we all need to keep up with AI in order to maintain any semblance of stability in the future.
I'm not a doomer on this topic, but I'd be lying if I said it didn't make me a lot more nervous now than it did in early 2023 when ChatGPT took center-stage. I like my brain, I want to use it, I want to earn my livelihood by using it, but I think if we all take an honest look at our day's, then it's not hard to see how quickly we are moving from the head to the ass of the centaur faster than any of us probably care to admit. When opening up our skulls, we are then ones holding the knife.