TTL 63
In February, I installed and configured my first personal agent. I named it Totoro.
I should clarify: this was not a trendy choice driven by last year’s Studio Ghibli wave.
I’ve had Totoro tattooed on my ankle for ten years. I chose the name because Totoro is this discreet, friendly monster that shows up when you need help — and disappears when you don’t. That seemed like exactly the right job description.
It worked so well that a month ago, Totoro got promoted to chief of staff. It now coordinates a team of eight specialized agents. And yes, the obvious question follows: Paolo, is your life so complicated that you need a team of nine?
No. The reason is more interesting than that.
An agent is context, tools, and a goal — and context is the whole game. At the foundation of every agent sits a large language model, and there aren’t many of those. They’re all trained on roughly the same data, capable of almost anything, which also means they’re generic by default. What makes them sharp is the context you give them. And context becomes confusing very fast when you talk about everything during a single day.
Here’s what my mornings with a single agent looked like: calendar brief, then world news, then meeting summaries, then a couple of blog posts I was drafting. By evening, the responses weren’t wrong — they were worse than wrong. They were generic. I was reverting to the mean and losing any chance for the tools to add value.
The solution is the same one we apply to human organizations: narrow the scope.
So now one agent parses my meeting recordings and extracts action items. Another manages my task tracker — flagging what’s quick to close, drafting what it can help with. Another handles proofreading when I write. Another I send out specifically for web research, kept separate on purpose so it doesn’t flood everything else with noise. Eight specialists, plus a chief of staff coordinating them.
The team is not a productivity flex. It’s a containment system.
We talk on Telegram, just because it’s already the app I use with family and friends, and I refuse to add another tool to my life. I have a direct message with each agent and one shared channel where handovers happen visibly. They could technically coordinate behind my back on the server. I prefer to see what’s flowing. Most of the time they follow that preference. 😂
Here’s the point I actually want to make: soon, every human will have many agents. My team today is half-professional, half-personal. Tomorrow I might split that into two separate teams with different access levels — suddenly I’m at twenty.
Push further: some agents will be persistent like mine, others ephemeral. I’m going to Sevilla in June and might spin up one just to book restaurants each night, then let it disappear after the trip.
Multiply that by billions of people over a lifetime, and a problem emerges that nobody has solved yet. When my agent emails you on my behalf, you need to verify it’s mine — and that it’s an agent, not me, in case it says something out of bounds. I’d love to tell Totoro: go find my friend’s agent, match our calendars, find two restaurants we’d both like, report back. Today, my agent has no way to locate his.
How should that work? I have an opinion, but I’ll talk about it in the next post.
If someone forwarded this to you, you can subscribe.
I also publish on paolo.blog and monochrome.blog.
Leave a Reply