The neural network is 200,000 neurons, trades Bitcoin autonomously, and happens to be the entire connectome of a fruit fly.
The Summary
- A Coinbase engineer connected a simulated fruit fly brain to a live Bitcoin trading account, where it's making real trades based on neural firing patterns
- The fly brain is down only 1% so far, beating most crypto influencers and several hedge funds
- The engineer explicitly warns this proves nothing about AI trading — but that hasn't stopped the experiment from raising harder questions about what qualifies as autonomous agency
The Signal
A Coinbase engineer wired a complete simulation of a fruit fly's brain into a Bitcoin trading account, and the insect's 200,000 neurons are now making real trades. The system uses the fly's actual neural connectome, the mapped network of every neuron and synapse, to process market data and execute buys and sells. No machine learning layers. No LLMs. Just raw biological architecture deciding when to go long.
The digital fly is down 1% on its crypto portfolio, a performance that falls somewhere between "impressively stable" and "statistical noise." The same simulated brain has already been taught to parallel park, play Beat Saber, and explore what the creator calls "its sexuality," proving the fly connectome is general-purpose enough to handle multiple domains.
"It proves nothing about AI trading, but everything about what we're willing to call an agent."
Here's what makes this weird: the engineer himself says it proves nothing. The fly brain wasn't optimized for trading. It wasn't trained on price data. It's just a biological neural network mapping sensory inputs (in this case, market data) to motor outputs (buy/sell commands). The fact that it works at all, even at break-even, suggests that any sufficiently complex decision-making substrate can produce market behavior that looks intentional.
This matters because we're about to have thousands of "agents" managing capital, and we have no working definition of what makes something an agent versus a script with a pulse. Consider the common categories:
- Rule-based bots that follow if/then logic
- Trained models that pattern-match historical data
- Agentic AI systems that "reason" through language models
- A literal insect brain wired to an exchange API
The fly sits in category four, which we didn't know existed until this week. It has no understanding of money, no model of supply and demand, no conception of risk. It just has neurons that fire in response to stimuli. And that's enough to make it functionally indistinguishable from thousands of other automated traders.
The Implication
If a fruit fly brain can trade crypto without losing its shirt, the bar for "autonomous agent" is lower than anyone thought. This isn't an argument for bio-computing or neural simulation. It's a demonstration that agency might be less about intelligence and more about interface. Plug anything with enough complexity into the right API, and it becomes an economic actor.
Watch for regulators to catch up slowly, then all at once. When they finally write rules for AI trading, they'll need to define what counts as an agent, what counts as autonomous, and what counts as responsible. A simulated fly brain that trades Bitcoin forces the question: where exactly do you draw that line?