OpenAI executive Thibault Sottiaux, General Manager of Products and Platform overseeing ChatGPT and Codex, predicted that in another 2-3 months, Codex will be an outdated tool. He stated that AI agents are evolving from single-machine tools to cloud-native infrastructure, emphasizing that "the laptop is no longer enough" for running next-generation models with long-context reasoning and high autonomy. The Harness framework, a suite of features including context management, tool invocation, state persistence, environment isolation, and error recovery, has gained widespread adoption among developers and office workers. However, running agents on laptops faces three major physical bottlenecks: computing power and memory limitations when executing 20 subtasks concurrently, long-running tasks requiring hours or days that force laptops to stay powered on, and parallel explosion of context and toolchains that local lightweight harnesses cannot handle. In response, Agent tools are transitioning toward heavy infrastructure, with OpenAI's Codex already supporting cloud-based asynchronous execution, sending tasks to isolated cloud containers to run autonomously.