Mike Gowen

Rodeo was Foundation's more casual, social take on collecting art. I co-led design from early concept through launch and continued product development.

Much of the work was hiding the complexity of web3 from people who just wanted to share and collect, and turning the ways people already showed off their collections into features.

With no sign of the NFT market recovering, we made the hard call to shift our focus to fungible tokens. The plan was to apply the same agent-first approach to a new market: Pico would let people spin up agents for tasks like managing a liquidity pool, dollar-cost averaging into a favorite coin, or automatically rebalancing a portfolio.

We started with a much narrower wedge: a dead-simple way for people new to crypto to manage a Uniswap LP. The user funds the account, picks a few options, and Pico handles the rest.

After a lifetime of apartments, my wife and I built our first house, and immediately realized we had no idea how to take care of one. I built Gable to help us keep up. It builds a picture of your specific home, factoring in its systems, age, and local climate, then uses AI to generate a personalized upkeep plan, explaining each task in plain language with the timing, effort, and cost surfaced up front.

Every illustration and animation was generated with image and video models, kept to a consistent visual style throughout. Gable is still in development.

Pico began as Deca.ai, a personal assistant focused on NFTs. But building another chatbot felt like a missed opportunity, so I threw out the original direction and shifted the product from answering questions to taking action: buying on your behalf, monitoring prices, and alerting you the moment an artist you follow released something new.

This was before agents became the dominant frame for AI products. The instinct was right, but with NFT market conditions continuing to decline, we ultimately had to pivot.

Twinkle is an AI-powered bedtime story app I built with my two young sons in mind. It asks for your child's name and age, an object to build the story around, and an optional lesson you'd like them to take away. From there it writes a custom story, hands it to a text-to-voice model to read aloud, and generates a cover illustration to match.

I eventually set it aside for two reasons: it relied on three separate models, so cost was a real concern, and the stories had to be genuinely good rather than the generic AI slop that's so easy to produce. Both were hard problems at the time, though I suspect they'd be far more tractable today.