AI & Overstimulation
My poor, poor monkey brain.
If you work in software, you’re aware coding agents have come a seriously long way in the last 6 months.
This time last year I was very much of the opinion that IDE auto-complete was the sensible endpoint for LLMs in development. I thought that Vibe Coding was a dead-end that would probably give me and my colleagues years of job security fixing messes for clients!
Since around January it became abundantly clear to me that I might have failed to foresee the scaling… and I don’t think I was the only one. Timescales for work have collapsed significantly, at the same time as the models become more intelligent and their harnesses refined.
Now don’t get me wrong, the velocity is dizzying. I can take ideas and experiments and have something real in minutes. I can tackle annoyances in projects that I would otherwise never be afforded the time to touch.
But my god. The context switching is truly brutal. I think for a lot of us the fact that we can do so much, inevitably leads us to doing so much. I’m waiting on Claude to finish X? Awesome, no problem. I’ll open another tab and have Claude window number 2 do Y… and Z, and so on and so on.
I finish the end of the day with three times as much done as I could ever have done prior. But my poor mammalian brain has had zero time to catch up alongside it. Manual coding had a natural speed limit built in. You can move at best, very slightly slower than the speed of thought. This created a cadence of thinking, building, digesting. A nice loop that, whilst still tiring when pushed, allowed you the breathing room to really internalise what you had made. Now all buffer is removed. The consequence is this pervasive, potent mixture of wired, exhausted, overstimulated and still thirsty to create more.
The comparison that comes to mind for me is social media and the internet, and their endless firehose of content. But that’s consumption, and I think this taps into something even more primal because it provides us with a near-endless supply of power to create. This is neither 100% bad, nor 100% good.
I think we’ll all have to grapple with how we keep this sustainable, especially as the models are only going to get faster.