I Trained a Fly's Brain to Generate WIRED Story Ideas
I used an open-source map of a fruit fly's brain to vibe code a website called PitchFly. Its headline suggestions were delightfully bananas.
By Will Knight | September 16, 2026
What if you could outsource your creativity to a fruit fly? That's essentially what I set out to do when I built PitchFly, a website that uses a biologically accurate simulation of a Drosophila brain to pitch story ideas. The results were strange, unpredictable, and occasionally brilliant—much like the insect itself.
The Brain of a Fly, Digitized
The project was made possible by the FlyWire connectome, an open-source, neuron-by-neuron map of an adult fruit fly's brain released in 2024. Containing roughly 140,000 neurons and 50 million synapses, it remains one of the most complete wiring diagrams of any complex brain ever produced. In 2026, this dataset has become a playground for the "neuroAI" crowd—researchers and hobbyists alike who wire up simulations of real biological circuits to see what emergent behavior falls out.
I'm not a neuroscientist. But with the help of an AI coding assistant, I managed to stand up a lightweight spiking neural network in Python, seed it with the FlyWire connectivity data, and hook its output to a large language model. The idea was simple: let the fly's neural activity act as a kind of chaotic random seed, and let the LLM translate that activity into English.
What Does a Fly Think About?
The results were, to put it mildly, unpredictable. The headline suggestions were delightfully bananas. Where a human editor might pitch a story about semiconductor supply chains, my fly-brain generated titles like:
- "The Silicon Trap: Why Microchips Dream of Water"
- "I Let a Bee Decide My Retirement Portfolio"
- "The Secret Sex Lives of Data Centers"
- "WIRED's Guide to Grieving Your Dead Smart Fridge"
Some of these were nonsense. But a few—especially the ones that combined two unrelated concepts—had a strange resonance. It turns out that biological noise, when filtered through a language model, can produce ideas that feel genuinely novel.
Why This Matters
The point isn't that a fly is a good editor. The point is that the boundary between biological computation and artificial generation is getting thinner. Projects like OpenWorm, the Human Brain Project's successor initiatives, and now FlyWire-based experiments are all part of a larger trend: using real neural data not just to understand brains, but to build new kinds of AI.
In 2026, this convergence has a name—neuroAI—and it's attracting serious investment from both academia and industry. Startups are already exploring whether connectome-inspired architectures can make AI more energy-efficient, more robust, or simply weirder.
Build Your Own PitchFly
If you want to try this yourself, the ingredients are all open source:
- FlyWire connectome data (available via the FlyWire Consortium)
- Brian2 or NEST for spiking neural network simulation
- Any LLM API for translating neural activity into text
- A weekend and a tolerance for weirdness
The code for PitchFly is available on GitHub. I make no promises about the quality of the ideas it generates—but I can promise they won't be boring.
Will Knight is a senior writer at WIRED covering AI and emerging technology.
via Wired AI