OpenAI’s leadership is pushing forward at breakneck speed in the race toward artificial general intelligence (AGI), and one key bottleneck they’ve identified is something surprisingly mundane: human typing speed. Alexander Embiricos, who leads product development for Codex—the AI coding assistant at OpenAI—recently highlighted this limitation during an appearance on “Lenny’s Podcast.” According to Embiricos, the current underappreciated constraint on advancing AGI isn’t computational power or model architecture, but rather the pace at which humans can write prompts and validate AI-generated outputs. “Human typing speed,” he said, “or human multi-tasking speed on writing prompts,” is a critical bottleneck. AGI refers to a theoretical form of artificial intelligence capable of understanding, learning, and applying knowledge across a broad range of domains—matching or surpassing human cognitive abilities. All major AI companies are racing to achieve it, and Embiricos believes the path forward requires minimizing human involvement in the AI workflow. “If you can have an agent watch everything you’re doing, but you’re still stuck reviewing every piece of code it produces, then you’re still bottlenecked by how fast you can read and respond,” he explained. The solution, he argues, lies in designing systems where AI agents are not only capable but also trusted and self-validating—able to check their own work without constant human oversight. “If we can rebuild systems so that the agent is default useful, we’ll start unlocking hockey stick growth,” Embiricos said. In tech parlance, “hockey stick growth” describes a sudden, exponential increase in productivity or performance after a period of slow progress. He acknowledged that achieving fully automated workflows isn’t straightforward—each application will require tailored solutions—but he’s confident progress is imminent. “Starting next year, we’re going to see early adopters begin to hockey stick their productivity,” he predicted. “Over the coming years, larger organizations will follow suit.” Ultimately, Embiricos sees this surge in automated productivity as the feedback loop that will propel AI development toward AGI. “That hockey-sticking will flow back into the AI labs,” he said. “And that’s when we’ll basically be at AGI.”