AI's 'Boil the Ocean' Mandate: A Call to Reinvent Core Developer Infrastructure

A prominent voice in the developer community recently articulated a profound shift in software development, asserting that AI’s capabilities now demand a fundamental rethinking of core infrastructure. This new paradigm, dubbed the ‘boil the ocean mindset,’ encourages developers to move beyond building highly specialized, vertical solutions towards creating broader, even ‘shitty horizontal plays’ – all-encompassing, functional-but-imperfect tools that were previously infeasible due to prohibitive costs and complexity. This mirrors the revolutionary impact of cloud computing, which democratized access to scalable infrastructure; now AI is democratizing the scope of what individual developers or small teams can build, as exemplified by projects like ‘Lakebed,’ a personal ‘shitty cloud for shitty apps’ built with agentic workflows.

This visionary perspective specifically targets foundational developer tools for overhaul. Package managers like npm and npx are criticized for systemic security vulnerabilities, opaque publishing processes, and a dire lack of actionable metadata, with calls for features like auditable releases, name squatting prevention, and granular permission displays for agents. Version control systems like Git and platforms like GitHub are deemed outdated, lacking granular permissions for private files or branches, and suffering from file system inefficiencies (e.g., APFS issues), advocating for a ‘Dropbox for devs’ model and in-memory source control. Communication platforms like Slack are highlighted for poor UX in message prioritization and thread management, suggesting a ‘posts’ primitive akin to Facebook Workplaces for better context management and agent integration. Furthermore, the state of mobile development (iOS restrictions, Android’s stagnation) prompts a call for a new, open mobile OS that supports Android apps while fostering developer customization and experimentation. The speaker underscores that AI agents, operating in dynamic, self-prompting ‘loops,’ are key to executing these ambitious projects, enabling complex multi-stage work and dramatically increasing developer output, thereby making such comprehensive re-imaginings both viable and necessary in the current tech landscape. The community is also urged to develop more diverse and ‘weird’ benchmarks to accurately measure and drive model capabilities.