Flying without knowing how the plane works

The author argues that software engineers still spend too much effort managing the machines and tools needed to build software, even as consumer apps have long become independent of the devices people use. Drawing on a career spanning web, mobile, infrastructure, and AI, he recalls repeatedly imagining ambitious projects only to hit limits in computing power, memory, security, or setup. AI has renewed the excitement of making things, helping developers tackle more work and solve problems they could not manage alone, but it has also added a moving set of models, tools, skills, and workflows to maintain. Development with AI has shifted through copy-and-paste, autocomplete, terminal tools, web interfaces, and now orchestration of agents; keeping up with each transition can become a job in itself.

His work at Airbnb focused on letting engineers use new AI capabilities without having to master or continually rebuild the underlying systems. A key advance was moving agents from company laptops to server infrastructure, where many can work together and tackle larger tasks. He now presents Hive, a product from his new company, Friendly AI Labs, as a way to make that model available more broadly. Hive aims to run agents in secure, preconfigured remote environments, let users coordinate and review their work from different devices, and keep tasks running when a laptop is closed. Building this experience requires substantial infrastructure for compute, credentials, communication, and interfaces; Hive packages those pieces so people can direct agents without managing the machinery. It is opening a small early-access beta.