The idea is fantastical. Throw a fuckton of compute at a swarm of AI agents and they’ll do magic. You instruct the agents to instruct subagents to instruct the sub-subagents — and see!
Companies started putting up “tokenmaxxing” leaderboards. Mac Minis are selling like crazy. Awhile back, I overheard people at a party talking about how many agents they had running at home right now (7!).
OpenAI, Anthropic, Google — all the labs want usage. Autonomous agents spinning in loops, each organization running the same query independently — or even better, each person at each organization running the same query, merging the outputs, validating the combination — it all sounds like cha-ching! to the labs and sad trumpet to our bank accounts, humanity, Nature.
It doesn’t have to be this way.
We can imagine different futures.
With Can Elon Musk Buy My City?, I’m exploring a few:
One AI, Many Humans
Rather than one human commanding many agents, many people work with a single AI, bringing their own expertise, questions, and framing. The agent’s job is to synthesize. No duplicate queries returning duplicate answers that need reconciliation. All conversations on the topic (say, “the value of Paris”) stay threaded.
Collaboration is validation is value
Every number you see requires an agent proposal and a human validation. Rather than having fifty subroutines to ensure perfection, we bring the user into the mix. We can build AI systems into human judgement, not around it.
Extract as much out of a token as possible
Or: use every piece of the animal. Agents can refer to previous conversations, citations, and transcripts in a city. If someone’s already dug into tourism numbers around the Eiffel Tower, there’s no reason for you to do it, too. Unless there’s something fishy. We’re paying for these outputs — why do they feel so ephemeral?
AI is here to stay. No one is coming to save us.
We need to start imagining new ways of using the tools.
Can Elon Musk Buy My City? is one proof of concept.