Alessio Fanelli

Alessio Fanelli is the founder of Kernel Labs. His public AI work includes coding-agent workflows and commentary on how vendors build advantages around models. Earlier, he cohosted Latent Space with swyx and published smol-podcaster, an open-source tool used for the show's transcription work.[3][8][9][2][6][7]

Fanelli introduced himself as Kernel Labs' founder in the February 12, 2026 interview with Jeff Dean. He asked how Google balances maximum frontier performance with the lower-cost end of the model performance frontier, connecting capability to the practical requirement of deployment at Google's scale.[3]

In his April 14 essay “Startups are context arbitrages”, Fanelli argued that running more workloads lets vendors learn and codify knowledge within their platforms, widening their context advantage.[9]

On May 1, he published an account of his OpenAI Symphony setup. Its task-starting workflow uses Todo as the status Symphony polls to begin a new task.[8]

His earlier Latent Space work provides concrete examples of both technical interviewing and applied AI tooling: he cohosted its first numbered episode in February 2023 and announced smol-podcaster that August.[2][6]

In the first Latent Space episode, published February 23, 2023, Fanelli introduced himself as a partner and CTO in residence at Decibel Partners alongside cohost swyx. He used the same role description in the May 3 episode with Replit's Reza Shabani.[2][4]

In a May 2023 discussion with Simon Willison, Fanelli distinguished obtaining the best model from building the best product around it. He also predicted that first-party data loops would become especially important if high-quality data reduced the amount of data needed.[5]

Fanelli announced the open sourcing of smol-podcaster on August 9, 2023. The MIT-licensed repository documents a publishing workflow that generates diarized transcripts with speaker labels and timestamps, timestamped chapters, and title suggestions informed by previous episode titles. Its default title examples come from Latent Space, which uses the tool for most of its transcription work.[6][7]

For parallel or remote processing, smol-podcaster documents a web interface with Celery background runs and a required message broker, naming RabbitMQ as the author's choice. Fanelli described its personal benefit in his release announcement: “It’s been saving me a lot of time, and hopefully it helps other creators out there!”[7][6]

Beyond the podcast, Fanelli hosted the Architect 1 track at the 2025 AI Engineer World's Fair, as credited in the June 13 conference recap.[1]