Designers Vibe-Code Prototypes and Submit PRs

Alexander Imbirikos leads Codex at OpenAI with a vision of AI as a software engineering teammate rather than just a coding tool. He believes AI should proactively participate across the entire development lifecycle, not just write code when prompted.

At the core of Imbirikos' philosophy is the idea that AI should function as a true teammate that helps with planning, coding, testing, and maintenance. He describes current AI tools as "really smart interns that refuse to read Slack and don't check Datadog unless you ask them to." This perspective shapes how he approaches product development—focusing on building systems that can operate autonomously while still keeping humans in control of the process.

Imbirikos believes the current limiting factor in AI advancement isn't model capability but human interaction bottlenecks. "The current underappreciated limiting factor is literally human typing speed or human multitasking speed," he explains. This insight drives his team to focus on reducing the need for humans to prompt and manually validate AI work, creating systems where AI can be "helpful by default."

For engineering leaders, this means rethinking how teams integrate AI. Rather than treating AI as a tool that requires constant direction, Imbirikos suggests configuring it as a teammate that can operate independently within defined boundaries. Teams should start by having AI handle specific, well-defined tasks like previewing builds or running tests, then gradually expand its responsibilities as trust develops.

For individual contributors, this perspective suggests focusing less on coding mechanics and more on systems thinking and collaboration. As Imbirikos notes, "It's still deeply worth understanding what makes a good overall software system... skills like really strong systems engineering or effective communication and collaboration with your team will continue to matter for quite some time."

The practical implications are significant. At OpenAI, teams have dramatically accelerated development—building the Sora Android app in just 28 days with a small team, which then became the #1 app in the App Store. This acceleration comes from treating AI as a teammate that can handle increasingly complex tasks while humans focus on higher-level direction and validation.

Imbirikos' approach to product development is empirical and bottoms-up. Rather than planning everything in advance, his team ships quickly, learns from user feedback, and iterates. This requires humility and a willingness to learn from real-world usage rather than theoretical planning. For teams adopting AI tools, this suggests starting with practical, real-world problems rather than hypothetical use cases.