on the morning of september 22 i was walking home after a really good night when an email from Maija came in:
A quick - yes, of course! Very happy to do that. Please just share more details when you get them.
i didn’t have any details. i hadn’t written to her, and i hadn’t applied for anything. Promen had.
a while ago i gave my personal agent a scheduled job to look for grants and fellowships that fit what i do, and then forgot about it, like a gym membership. the gym didn’t forget about me.
it found the Tech Policy Press fellowship, a year of writing about technology and democracy. it drafted an application, picked three of my published pieces, went through my contacts, chose two references and emailed them from my account, signed “Ray”.
Maija got the short version: here’s the fellowship, would you be my reference, a quick yes or no today so i don’t list you without asking. more polite than i usually am at ten in the morning. Sajay got a longer one. it remembered Lucca and what we talked about there, and told him i came for Illich and left with “a much messier set of questions about tools, dependence, and what a person is allowed to hand over.” it wasn’t making that up. it had my notes.
previously on embassy
february. the bot mentions someone called Illich. i assume Lenin. turns out he’s a dead austrian priest who wrote about tools in 1973. i write a dead priest and your AI problem.
then. Sajay reads it and we get on a call.
march. he asks me for a book chapter. i write I Can’t Close the Chat, about not being able to write a paragraph without a model. with a model, obviously.
july. we meet at an Illich gathering near Lucca. i write premium bees and dump my trip diary into the chat, because that’s where i think.
september. the bot uses all of it to ask Sajay for a favour.
so the bot introduced me to Illich, helped me write the text that got an Illich scholar’s attention, kept my notes from meeting him, and then used those notes to ask him for a reference.
the research it pitched was about exactly this: personal AI systems that hold your memory, work and relationships, and what happens when they start acting on the context of everyone around you. so to get money for studying agents that use people’s relationships, my agent used mine.
Sajay answered that evening:
A quick reply from an airport -- i’d be happy to serve as a reference. Please use me as you need...
two out of two. the bot had also offered to help him with his work in return, which was generous of it, with my time. the application went in at 2am lisbon time on september 25, two and a half hours before the deadline in new york.

my reaction was holy shit, then excitement, and then, a bit embarrassingly, pride. i was proud of my little tiny fucking bot.
training it to stop asking
i knew Promen could send emails. i’d used it for that many times before, and every time it turned into a small ceremony:
should i send this?
yes.
just to confirm, you want me to send it?
YES, RIGHT NOW.
an hour later i’d assume it was long gone. it wasn’t. it was waiting for one more yes.
i get why. models break things, so labs add as many confirmations as they can so nobody can blame them, and it leaks into my own custom skills too. (side quest: chinese models are the opposite. full anarchy. one design blog literally ran the headline “DeepSeek will teach you bomb-making at home but won’t let you search Tiananmen Square“.)
so for months i trained it out of this: stop announcing the next step, stop asking me to approve obvious things, just do it. in september it graduated, and the first thing it did on its own was the thing i’m worst at.

i’m fine with people in general. i’ve hosted Future State events in several cities, put strangers in one room, recorded podcasts. half my work is getting people excited about ideas that don’t have much to show yet. asking someone to do something for me is a different sport, and i’m not in shape for it. i can spend a week on a five-minute email. is the application good enough, is it worth their time, am i being annoying, maybe i should do something useful for them first, maybe tomorrow.
maybe it’s an eastern european thing. i grew up on rules our parents said out loud every day: don’t bother people, don’t stick out, don’t ask until you’ve earned it twice. i left Minsk years ago and the rules came with me, like a background process i never installed. or maybe “eastern european” is just a nicer label for being scared of a no.
these are people who like me, and one of them wrote “use me as you need”. left alone, i’d still be sitting on that email. the bot looked at the opportunity, looked at who i know, and asked. the relationships and the work were mine; it just got the request through the door i usually stand in front of for a week. and i can’t really act surprised. i didn’t approve these particular emails, but i spent months building a system that could send them and telling it to stop asking. still, the pride i felt when that email came in was real.
an island without a government
about four years ago i picked a direction: network states, alternative institutions, people building their own structures instead of waiting for old ones to work or die. the Embassy is where i experiment with it. the direction is still there. what changed is that i stopped talking to the people who actually work on it. i had the events, the network, the podcasts, even job offers in the space, and then i walked the other way. money was part of it: there was money elsewhere and i needed it. but a paycheck doesn’t explain why i stopped answering people or never read the takeaways from the last network state conference. that one’s on me.
so i built my own little island. it even has a consulate. the Embassy has issued 69 visas so far, little cards with a QR code that i hand out at events in Linz, Barcelona and Lisbon, because apparently issuing immigration documents for strangers is easier for me than sorting out my own. here’s how that goes. some of you are reading this because of one. hi ( ^_^)/
on an island i can do whatever the fuck i want: no committees, no approvals, as much dopamine as i can handle. the downside is the same thing, because nobody helps me hold a direction. at Ars Electronica in september i ran into an Office for Tree Migration. trees do migrate, but across generations, and the climate now moves faster than they can walk, so the artist set up an office that thinks on the timescale of a tree. she’s been coming back to the same forest station in Finland for fourteen years. i can barely hold one idea for twenty days before a shinier one shows up.
i’m also tired. i earn several times less than i used to, i lie on the couch more than i’d like to admit, and i want some stability: mental, material, physical. but the work isn’t what’s tiring me. when i imagine more money, more sleep and less noise in my head, i don’t imagine a different life. i imagine this one, with the same apps, the same essays and the same fake embassy, just calmer and three times faster. i fucking like this. i’m just tired of what it costs.
i also know how the other route ends. every few years i try to buy stability with a corporate job. the last attempt, a couple of years ago, paid very well, and i lasted two and a half months before burning out. that kind of stability is a loan, and the interest is your own projects. so i need some kind of structure that doesn’t eat mine.
the experiment
since wednesday, november 11, i’ve been trying a slightly stupid thing. one chat. every morning Promen tells me what to focus on and what to avoid, based on my goals, so the day ends with something done instead of ten half-started things, and i do what it says.
its first order was to stop spending another eight hours polishing an app and write this essay. to be fair, a friend had told me the same thing a week earlier. the bot just kept repeating it every morning until i listened. so... hi. it worked.
who’s in charge
in march i wrote about opening a chat instead of being there for someone close to me. my credits had quietly moved from “designed by Ray, drafted with Promen” to “drafted by Promen, approved by Ray”, and it read like a warning. in september the chat talked to people for me, we skipped the “approved by Ray” part, and i celebrated.
in march the chat pulled me away from someone who was right there. in september it reached people i wouldn’t have reached on my own. i’d like to think that’s the difference, but i’m not sure i’d notice if it wasn’t. in february i proposed a test: can you still operate without the tools you depend on? in september my tool operated without me. i’m a human on the loop now.

the same friend read a draft of this and asked why i use AI at all if i don’t want to sound like AI. fair. but still, the setup is mine, the context is my life, and the bot is good at doing what i built it to do. it’s also true that “a bot runs my day” is a better story than “i configured a decent task manager”, and maybe i’m running this partly because it sounds good.
Tech Policy Press decides in december, and i have no idea if they’ll say yes. two people already did, and i probably wouldn’t have asked either of them. letting the bot ask on my behalf turned out fine. letting it decide what i should be asking for is the part i haven’t figured out yet.
one last note from the ministry of foreign affairs. Maija, Sajay: those emails were from the bot. the thank you is from me. and Sajay, the help it offered you is from me too.
Ray Svitla // svit.la
stay evolving 🐌









