2. OpenAI’s Codex Gets Subagent Support, Shifting Agentic Coding From Sequential to Parallel
OpenAI researcher Greg Brockman (@gdb) announced that subagent support is now live in Codex, OpenAI’s cloud-based AI coding agent. The feature allows Codex to spawn and coordinate multiple subordinate agents working in parallel, enabling what Brockman describes as the ability to “get large amounts of work done quickly.” This is a functional architecture shift, not a marginal capability update: Codex can now decompose complex engineering tasks, distribute them across concurrent subagents, and aggregate results, rather than processing instructions in a single sequential thread.
The competitive implications are immediate and pointed. Anthropic’s Claude, through its “computer use” and multi-step tool-calling capabilities, and Google’s Gemini-based coding integrations have been pushing toward similar agentic workflows, but parallel subagent orchestration is a meaningful differentiator in throughput. For enterprise engineering teams and developer-focused SaaS platforms building on top of OpenAI APIs, this substantially raises the ceiling on what a single Codex session can accomplish per unit of time. Startups in the AI coding assistant space, including Cursor and Cognition (Devin), now face a more capable native OpenAI offering that undercuts their core value proposition around autonomous, multi-step code execution.
The broader signal here is that the agentic coding race is consolidating around orchestration depth, not just code generation quality. The ability to parallelize work through subagents mirrors patterns in distributed software systems, and the labs that nail reliable multi-agent coordination, including task delegation, error recovery, and result synthesis, will define what “AI software engineer” actually means in practice. Brockman’s framing of subagents as “very fun” is casual, but the architectural move is not: this is OpenAI staking a position that Codex should be evaluated not as a coding assistant but as a coding workforce.