swyx's 'Agent Lab' Thesis: AI Startups Should Own the Problem, Not the Model, to Survive Frontier-Lab Upgrades
Latent Space·medium signal
On Latent Space, swyx laid out the agent lab framing — startups that own a specific customer problem end-to-end survive frontier model releases, while startups that own a model capability get absorbed by the next upgrade. It's the sharpest available answer to the recurring "did OpenAI just kill your startup" cycle, and it converges with Guillermo Rauch's argument that the real fight is splitting models from agents. Both are betting that durable value sits in the agent/product layer, not the model layer.