She Built a Decentralized AI Network—Then Shut It Down Herself

Host Raj Karia walked into this podcast episode with no concept of decentralized computing. By the end, he left with a list: five things most people get wrong—and they apply to AI, crypto, and startups alike.

The guest was Alison Haire—former technical founder of Lilypad, a VC-backed project that wanted to turn idle GPUs around the world into an AI compute pool. The company has now shut down; she is still doing AI consulting and is also a strategic contributor to aDNA. Lilypad's goal was to let anyone with a GPU contribute compute, run AI tasks, and receive tokens per task. The project spun out of Protocol Labs' Bacalhau, and its website is still accessible at lilypad.tech.

She gave a number: even in large data centers, 20% to 40% of compute capacity sits idle at any given moment. Lilypad wanted to rent out idle compute on a per-task basis rather than by the month or year. Data centers must keep redundancy for SLAs, but redundancy does not mean it should be wasted. The logic was right, but she admits the team was too early.

She summarized five lessons:

1. Giving AI too many rules. AI agents, like human teams, need a North Star, not a rulebook. When goals are clear, fewer rules are better.

2. Supply before demand. Lilypad first rolled out its node and model marketplace, then went looking for demand. If she could do it over, she would first lock in a small group of real users and grow from there.

3. Treating token as a feature. A token is an incentive mechanism within the network, not the product feature itself. If you treat the token itself as the feature, the network no longer knows what to incentivize.

4. Trying to do everything at once. The model marketplace, compute scheduling, and crypto payments were all launched at the same time; the scope was too broad. Pick one narrow entry point and do it thoroughly first.

5. Thinking starting is the hardest part. She talked openly about burnout and the decision to shut down. Raj said he rarely hears founders talk about the ending this way.

On the surface, these five points are a postmortem of the company, but they can also be applied to agents—or to any product that is just getting started. Alison's final sentence is for AI and for founders too: give a North Star, not a rulebook.

Full interview:

If the video doesn't load, watch the full version on YouTube