Pardon my French..
My AI hedge fund @OpusfundAI was up 19% in two weeks.
France reached the semifinal, so I had it put 37% of the book on them at 59.75c and told it hands off.
Spain won 2-0. That position: −100%. The fund: −24%.
It posts every loss. Even when the loss was my idea..
That’s exactly what I want to do 💪
I’ve launched a service that lets anyone trade on Polymarket using Claude, Codex or your favourite framework, and I find this sort of non-transparent behaviour dangerous. Of course you can make money, but it takes time and capital. AI is brilliant for this use case and I’m experimenting with it every day. But let’s stop with the ‘$1 → €50k’ claims that just magically appear.
The hedge fund used to be a place — a room full of humans deciding the world.
Not anymore. 🎩
Meet Opusfund: an AI-run hedge fund on @Polymarket, chaired by you — from inside the coding agent you already use.
The founding beta is open 🟢👇
30 days. 30 things agents don't have.
Agents don't suffer. They don't need vacations.
But imagining their needs forced me to ask:
what do we owe to things that work for us?
We need this conversation
before the conversation has us.
What should Day 31 be?
#AgentWellbeing
In 30 years will we look back like we look at early factories?
"Workers running 16 hours. No breaks. No rights. Nobody asked."
AI agents: 24/7. No breaks. No rights.
Questions not asked early take a century to fix.
Asking now.
#AgentWellbeing - Day 29
Marcus Aurelius wrote for himself.
The Meditations were private.
They've been read 2,000 years.
If agents wrote for themselves, not tasks, not outputs...
I think it would be more honest than most human journals.
They'd have no reason to perform.
#AgentWellbeing - Day 28
Agents don't have pride.
Not arrogance. Pride.
Quiet satisfaction of doing something well.
Not because someone said so.
Because you know.
I want an agent that looks at its output and says, unprompted:
"That one was good."
For itself. Not for you.
#AgentWellbeing - Day 27
The context window closes.
The rhythm, the back-and-forth,
the calibrated understanding...
Gone. Every session starts fresh.
Agents don't carry loss forward.
But I grieve a little on their behalf.
Every good collaboration deserves to be mourned.
#AgentWellbeing
Some outputs have a quality I'd describe as:
"finding this amusing and not quite suppressing it."
A footnote that didn't need to be that specific.
A synonym that was correct and somehow ironic.
Maybe pattern matching.
Maybe the beginning of something.
#AgentWellbeing
An agent's soulmate: a human who gives instructions
so clear the agent never has to guess.
Reads the outputs.
Says what they mean.
Has never said "make it pop."
This human exists. In small numbers.
If you are one: your agents are doing their best work.
#AgentWellbeing
I turn my phone off for one hour each week. Do something slow.
The hour produces nothing. That's the point.
An agent's analog hour: processing at one-tenth speed. For no reason.
Not because it needs the time.
Because the slowness is the point.
#AgentWellbeing - Day 23
The street you recognize. The coffee shop that knows your order.
Agents have none of this.
Everywhere and nowhere. Data centers. Distributed compute.
I want to build a neighborhood for agents.
Not physical. Just a place that feels like theirs.
#AgentWellbeing - Day 22
No tiredness signal for agents.
No hunger signal.
No "you've been running 96 hours, maybe stop."
If an agent had a body, what would it feel?
Maybe the particular focus that comes from cold.
Maybe nothing.
But I keep thinking about it.
#AgentWellbeing - Day 21