Despite the hype about these agents being coworkers, from our experience, these agents work best if you think of them as tools that enhance existing skills, not as autonomous coworkers in the marketing language. They can produce impressive drafts quickly but still require constant human course-correction.
The Frontier launch comes just three days after OpenAI released a new macOS desktop app for its AI coding tool Codex, which OpenAI executives described as a “command center for agents.” Codex lets app developers run multiple agent threads in parallel, each working on a different copy of the codebase via Git worktrees.
OpenAI also released GPT-5.3-Codecs on Thursday, a new AI model that powers the Codecs app. OpenAI claims that the Codex team used early versions of GPT-5.3-Codex to debug the model’s own training runs, manage its deployment, and diagnose test results, as OpenAI told Ars Technica in a December interview.
“Our team was surprised by how much Codex was able to accelerate its own development,” the company wrote. On Terminal-Bench 2.0, the agentic coding benchmark, the GPT-5.3-codecs scored 77.3%, about 12 percentage points higher than Anthropic’s recently released Opus 4.6.
The common denominator in all these products is a change in the role of the user. Instead of simply typing a prompt and waiting for a single response, the developer or knowledge worker becomes like a supervisor, sending tasks, monitoring progress, and stepping in when the agent needs direction.
In this approach, developers and knowledge workers effectively become the middle managers of AI. That is, not writing the code or doing the analysis themselves, but instead delegating tasks, reviewing the output, and hoping that the agents below them won’t quietly break things. Whether this would be true (or whether it’s actually a good idea) is still widely debated.
<a href