According to OpenAIDevs on X, subagents are now supported in Codex, enabling developers to spin up specialized agents to keep the main context window clean, tackle parts of a task in parallel, and steer individual agents as work unfolds (source: OpenAIDevs). As reported by Greg Brockman on X, the feature is positioned to help teams complete large amounts of work quickly via parallelization and scoped contexts (source: Greg Brockman). According to the OpenAIDevs announcement video, business impact includes faster iteration cycles, reduced context-switching overhead, and clearer orchestration of complex, multi-step pipelines—key for use cases like multi-repo code refactors, data pipeline validation, and evaluation harnesses for model experiments (source: OpenAIDevs). For engineering leaders, the opportunity is to design agent architectures that allocate subagents to discrete responsibilities—planning, retrieval, code generation, testing—and consolidate results into a primary agent, improving throughput while preserving auditability and cost control (source: OpenAIDevs and Greg Brockman).

According to DeepLearning.AI, OpenAI introduced Frontier as an enterprise platform to build, coordinate, and evaluate organizational AI agents, enabling unified control over agent identities, permissions, shared context, and performance from a single interface (as reported by The Batch via DeepLearning.AI). According to DeepLearning.AI, the goal is to help companies manage growing teams of AI agents working alongside employees, centralizing governance and monitoring for compliance and reliability. According to DeepLearning.AI, this positions Frontier as an orchestration and evaluation layer on top of OpenAI models, supporting scale-out agent workflows, auditability, and role-based access that can reduce operational risk and accelerate deployment across functions like support, sales ops, and IT automation.