You have a side project? How cute.

I feel like these days, with AI assistance, having one side project is considered rookie numbers.

New workflows involve waiting. You give an LLM a task, and off it goes. Whether we’re talking about an LLM as an assistant, an agent loop, or a software factory, every once in a while you free your attention.

What to do with this free time? Maybe I’ll finally review that vibe-coded feature. Or think about what to build next. Or do the laundry.

Or… why not start another LLM session?

And when two LLMs are busy thinking, why not a third?

With two basic subscriptions, three concurrent AI-assisted projects feel to me like a reasonable number. My attention ends up distributed roughly like this:

  • 60% – The One Thing. It’s probably still a good idea to have top priority
  • 30% – The Something Else. I mean, the other subscription have to play for itself
  • 10% – The Maybe That Thing Too. Because sometimes the first two are both thinking

The upside? It’s genuinely cool. You can make meaningful progress on three projects in a single evening. Nowadays, two basic subscriptions can get you surprisingly far, without running into usage limits every 30 minutes. Good luck trying to develop three software projects at once five years ago.

The downsides? Constant context switching – I don’t believe three projects can get enough love and attention this way. And mind you, attention to detail is becoming more and more valuable now that everyone is basically a product engineer.

Plus, there’s another slight inconvenience: the human brain turns into a salad.

Our attention has been under attack for quite some time anyway (by doomscrolling, notifications, etc). I’ve spent pretty much my entire career trying to protect my attention, ideally by staying in deep work for hours.

Now, ironically, the very tools that make me so much more productive seem to be pulling me in the opposite direction.

And the worst part? It’s working. I have 14 side projects, literally. Some even close to finish (finally! After all these years!). I’m getting more done than ever, so I’m not exactly complaining.

But somewhere between the second and third LLM session, I can hear my poor mind crying for help.

What helps me:

  • Mediate (or swim – it’s the same thing)
  • Read books – brains seem to like that
  • Do something truly monumental without LLMs – like writing a blog post (not even the brain sandwich helps, I realized)

Those are nice ways to recharge, sure, but I’d rather stop frying my brain in the first place. Perhaps the tools themselves should be designed differently.

My bet is that a properly designed software factory could actually be the savior here. Not a fully autonomous one – I don’t believe in those yet. But one designed around managing human attention, rather than maximizing the number of agents or their output.

(And yes, that’s one of my side projects too).