**A Fruit Fly Launched a Memecoin
**
It earned $106,000 in creator rewards in six and a half hours. Every number below is on-chain and verifiable.
**The Launch
**At 18:23 UTC on September 10, 2026, a simulated fruit-fly brain signed a transaction on Robinhood Chain and created a token called FLYBRAIN.
Six hours and thirty-seven minutes later, the token had reached a market cap of $6,870,529, was up 101% in the previous hour, and had generated 320.63 GOOGL in creator fees across 358 sweeps — approximately $106,600 at GOOGL's implied price of $332.58.
The launch itself cost 0.000979 ETH.
Token flybrain ($FLYBRAIN
Contract 0x4eb990547bce4a982432ca88cf5fae7eed1a2d35
Transaction 0x63b2164f3d784e46538cc81d3c48095bd7252a3d12398fdad9991870d12a4f1c
Block 59614342
Status SUCCESS
Creator 0x6ce4085EfB52a6eBDb7d6989beb8860847f4b42A
Supply 1,000,000,000
Pair GOOGL
Tax 1.00%
What the Fly Is
This is not a metaphor. It is also not an AI agent with a fly-themed persona. In September 2026, HHMI Janelia's FlyEM team, the Cambridge Connectomics Group, and Google Research released the complete connectome of a male Drosophila melanogaster central nervous system under CC-BY.
The dataset contains every synapse traced from electron microscopy of one real animal.
We retain the 165,122 traced neurons and remove any pair with fewer than three synapses, leaving:
10,228,000 signed connections.
Each connection's sign comes from its predicted neurotransmitter:
Acetylcholine → excitatory
GABA and glutamate → inhibitory
The neurons run as leaky integrate-and-fire units:
Resting potential: −52 mV
Threshold: −45 mV
Time constant: 20 ms
Nobody trained these weights.
They were measured.
How the Fly Sees a Web Page
A screenshot of the live launchpad is sampled through the fly's 892 retinotopic hexagonal columns into L1 and L2 — the lamina monopolar cells that sit directly behind photoreceptors R1–R6. All 165,122 neurons then integrate for 12 ms of simulated time. When forward drive pushes beyond the bottom of the window, the page scrolls.
The fly walks down a webpage the same way it walks across one.
**The Launch Process
**The fly read the form through its retina and typed the description and ticker.
It clicked the launch button, then clicked Confirm in the launchpad dialog.
That interaction produced exactly one eth_sendTransaction, which was signed in Python. The private key never enters the browser.
**What the Fly Did Not Do
**This is the part most projects leave out. So it goes here rather than in a footnote. The fly did not complete the entire form. It reliably gets the description, often the ticker, and rarely the name. A separate script completed anything it missed, selected the paired asset, set the creator tax, and pressed Confirm. The live log explicitly labels which fields came from the fly and which came from the rig. It cannot read. The visual input is only about 30 × 30 pixels across the 892 columns. Text is texture to it.
It did not choose the name, ticker, or pair. It also has no idea what a memecoin is. Light mode blinds it. On a white page, it filled 0 of 3 fields. On a dark page, it filled 2 of 3. Its retina is fundamentally a luminance map, and the difference is one of the clearest demonstrations that the visual pipeline is doing real work. None of this is investment advice.
It Learns Now, Too
Since the launch, one place in the network is allowed to change:
the Kenyon cell → MBON synapse in the mushroom body, under dopamine. The learning rule is the measured one. A Kenyon cell active shortly before a dopaminergic neuron fires has that synapse depressed, rather than strengthened.
In other words:
Learning in a fly is subtraction.
There are 44,042 synapses in this learning pathway. There is no potentiation anywhere. The reward system was not hardcoded
The reward and punishment roles of the MBONs were determined by comparing total PAM input against total PPL1 input for each MBON. Whichever was stronger determined the side.
That produced:
Reward side: MBON01, MBON02, MBON03
Punishment side: MBON04, MBON10, MBON11
That matches the literature. Nothing was told to it.
**A simple learning experiment
**After 20 rewarded encounters with the same view:
Reward-side MBON weights moved −6.0%
Punishment-side MBON weights remained at −0.9%
Kenyon cell weights remained unchanged
Only the synapse that is actually allowed to learn moved.
Check It Yourself
The entire brain simulation is public.
GitHub: github.com/fruitflydev/flycoinrh
The repository is MIT licensed.
The connectome itself is not ours to relicense; it remains CC-BY from Janelia, Cambridge Connectomics, and Google Research.
The fly is roaming live at: flybrain.online
