AI coding agents and a rerun of the operating-system wars (gaseri.org)

The Agentic AI for Science workshop highlighted the evolving landscape of AI coding agents, drawing parallels with historical operating-system wars. It underscored the importance of the platforms on which scientific workflows are built, emphasizing the distinction between proprietary AI models like Claude and GPT and the open-source tooling surrounding them. While proprietary solutions can provide a rich ecosystem for applications, the challenge arises in how users and developers can independently modify and migrate their work across platforms. This raises pertinent questions about control, flexibility, and collaboration, reminiscent of past debates in software development and deployment. The article discussed how harnesses—frameworks that interact with AI models—can function similarly to operating systems, determining the capabilities and integrations available to developers. Examples like Claude Code, Codex CLI, and OpenCode illustrate the trade-offs between proprietary solutions and those offering more freedom and flexibility. The comparison reveals that choosing a model does not dictate the entire development environment, as harness choice significantly influences workflows and tool compatibility. Ultimately, this evolution in AI harnesses signals a significant moment for the AI/ML community, as it reshapes how developers think about integration, collaboration, and the future of scientific computing with AI agents.