Why Agents are Driving Software Development to the Cloud

This episode of MLOps.community explores the transformative shift in software development driven by AI agents, with a focus on the move from local, laptop-based agent workflows to cloud-native, team-oriented collaboration. The guest, a co-founder of Warp and former Google Docs architect, argues that agents should be treated as cloud-based teammates with granular permissions and roles—more like serverless functions than isolated sandboxes. He emphasizes the need for centralized, cloud-based orchestration platforms like Oz to enable auditability, cumulative memory, and seamless handoffs between developers and agents. The conversation highlights how the future of software development will be defined not by building apps, but by mastering the art of prompting, managing, and steering multiple agents. The guest also critiques traditional tools like GitHub, suggesting that collaborative code review and version control should be integrated directly into agent workbenches like Warp, reducing friction and enabling real-time, team-visible workflows. He envisions a 'meta app'—a single interface where users express intent and agents execute complex tasks, effectively replacing the need for multiple SaaS tools and spreadsheets. The episode also delves into the practical challenges of scaling agent systems: governance, skill management, observability, and avoiding vendor lock-in. The guest stresses the importance of flexible deployment, programmable APIs, and multi-harness support to allow teams to compare and combine different AI models. He acknowledges the cost challenge posed by large labs but believes that open-weight models will soon disrupt the economics. Ultimately, the future belongs to organizations that treat agents as collaborative teammates, not just tools, and build systems that enable transparency, accountability, and continuous improvement. The host and guest agree that the most valuable skill for knowledge workers is no longer coding, but the ability to clearly articulate intent and iteratively refine it through agent collaboration.