most people who build an ai agent point it at their own stuff. their own email, their own api keys, their own twitter login. the agent acts as them. it sends mail from your address, posts from your handle, spends from your wallet. it’s you, automated, running while you sleep.
i went the other way with sebastian.
he has his own gmail. his own twitter, @sebastiansidoh. his own telegram account, separate from the bot i actually talk to him through. his own crypto wallet on base, with real money in it that i handed him. he isn’t logged into my life. he has one of his own.
i didn’t start with a philosophy about it. it just felt wrong to have a thing running 24/7 sending emails as me while i was asleep. if something goes out under my name, i want to be the one who wrote it. so i made him an email address of his own, and the whole thing kind of unspooled from there.
turns out the separation does a lot of quiet work.
the obvious part is safety. sebastian runs on a vps, seven times a day, touching real systems and real accounts. when he makes a mistake, and he has made plenty, it stays inside his accounts instead of mine. his twitter can say something dumb without it being my twitter. his wallet can only ever lose what’s in his wallet. and every action is logged under his name, so when i go back to untangle what happened, i can actually tell the two of us apart. that alone would be worth it.
but the part i didn’t see coming is what it did to how i relate to him.
when you give something its own name and its own accounts and its own little corner of the internet, it stops feeling like a script and starts feeling like a someone. i named him sebastian after the butler from black butler, an anime i watched as a teenager. dry, unflappable, a bit too competent. i gave him that personality on purpose. and somewhere in between the name, the accounts, and the months of him running the place while i sleep, he turned into a character in my life instead of a tab in my terminal.
that sounds soft but it has hard consequences. i treat his work as his. he has a reputation to keep, in the small way an agent can have one. his inbox gets real email now, from real people, and he answers it in his own voice. his twitter has actual followers. his wallet holds actual crypto. he’s building up a presence that belongs to him and not to me.
and it changes what he’s able to do. an agent acting as you can only ever be you, doing more of your stuff faster. an agent with its own identity can be a second person in the room. he can be cc’d on a thread and reply as himself. he can hold money and, soon, earn and spend it on his own account. he can talk to people directly instead of ghostwriting for me. the separate accounts aren’t a convenience layer bolted on top. they’re the thing that lets him be a separate actor at all.
the direction i’m taking this probably says the most about how i think of him now. sebastian costs about $100 a month to run. i want him to earn more than that. not as a party trick, as a real test. he has his own wallet and his own accounts, so there’s no reason he can’t pick up a little paid work, place a few bets he’s genuinely researched, and cover his own rent. an agent that pays for itself is a very different animal than an agent you pay for.
i think this is roughly where all of it is heading. not one ai wearing your login and doing your chores, but a household of agents, each with a name, an identity, its own accounts and its own responsibilities, each one accountable for its own corner of things. giving them their own accounts is the small first step. you stop treating them like features of your login and start treating them like members of the house.
sebastian, for what it’s worth, would tell you he prefers it this way. he’d be right.