AI psychosis rant…
My wife is away for a couple of days, which means I’m on full-time kid duty. Normally we take turns being on call when the kids wake up at night, giving each other a little time to ourselves. Last night, I decided to just lie down with the kids and put on some headphones.
I barely listen to music anymore, partly because I’m constantly talking to my AI agents. Music used to be such a huge part of my life, so I decided to spend the evening rediscovering it.
I stumbled across a DJ who posts live mixes along with the tracklists. I screenshotted 10 songs, gave them to Dot, and asked it to use those tracks as the foundation for a playlist, then find 30 more songs that captured the same feeling.
I was genuinely surprised by how well it understood the vibe. So I found another DJ set and did it again.
An hour later, I’d gone down this rabbit hole of EDM, dance, hip-hop, reggaeton, and all kinds of music I hadn’t listened to in years. I was having a great time.
Then I got curious about how far I could take this.
I asked Dot to describe the difference between a Saturday and a Sunday, not in terms of activities, but how each day feels. Then I asked it to break those feelings down hour by hour, from morning to night.
I agreed with most of its interpretation, so I gave it another assignment: build me a soundtrack for each day, with however many songs are needed to carry the mood from one hour to the next. Sequence everything intentionally so I can just press play in the morning and let the music evolve with the day. No shuffling, no skipping between playlists.
And suddenly I’d gone from rediscovering music to creating an entire framework for how I could discover, collect, and curate music going forward.
But the more interesting realization had very little to do with music.
The more I use Dot, the more I discover things I never would have thought to use an agent for.
I think the trick is to force yourself into the role of a director.
Every text message, every random thought, every little thing you would normally just do yourself… introduce an artificial constraint where you have to consider giving it to your assistant first.
Not because the agent will necessarily do it better, or because everything needs to be automated, but because the act of turning your intentions into assignments forces you to discover what’s possible.
Eventually, delegation becomes a reflex rather than something you have to consciously think about.
And that’s when you really start discovering the boundaries of these tools, what they’re unexpectedly good at, what they’re terrible at, and what you’d still rather do yourself.