You're typing too slowly. At OpenAI, researchers say that humans are the main bottleneck on the path to artificial general intelligence

OpenAI’s Codex product lead, Alexander Embiricos, recently stated on Lenny’s Podcast that one of the key barriers is human «typing speed.» In other words, we type too slowly and switch between tasks inefficiently.

According to him, AI agents can already do much of the work independently, but the real «bottleneck» emerges when humans need to write prompts and verify or approve results—a process that takes time and attention. «If an agent can’t validate its own work, you’re still limited by whether you can review all that code,» Embiricos explained.

He sees the solution as removing humans from this cycle as much as possible—making AI systems «useful by default,» able to self-verify and work autonomously. When this happens, productivity will skyrocket—growth will look like a «hockey stick» curve.

Embiricos predicts that starting next year, individual users will begin to notice this effect, followed by companies. Somewhere between these initial productivity spikes and the mass adoption of agent systems, he believes that artificial general intelligence will emerge.