AGI Development Hits Unexpected Roadblock: Human Typing
OpenAI’s push toward artificial general intelligence (AGI) is racing ahead, but one surprising obstacle is slowing progress: human typing speed. Alexander Embiricos, head of Codex at OpenAI, revealed on Lenny’s Podcast that humans themselves may be the bottleneck in achieving AGI. As AI systems become increasingly capable, the pace of human input—writing prompts, validating AI outputs—cannot keep up, potentially limiting breakthroughs in AI reasoning.
AGI represents the next frontier in AI technology, a system that can understand, learn, and reason at or beyond human-level capabilities. Major AI companies are competing to achieve it first, but Embiricos believes that humans are currently the limiting factor, not the machines.
Why Typing Speed Matters for AI Progress
At the heart of the problem is the way humans interact with AI. Most advanced AI systems, including OpenAI’s Codex, rely on users to provide prompts, review results, and validate outputs. This workflow, Embiricos says, is slow and prone to bottlenecks.
“You can have an agent watch all the work you’re doing, but if you don’t have the agent also validating its work, then you’re still bottlenecked,” Embiricos explained. The result is that even the most capable AI cannot reach its full potential if humans cannot feed it instructions quickly enough.
From Prompting to Automation: The Next Step
The solution, according to Embiricos, lies in reducing human dependency in AI workflows. Instead of humans constantly guiding and checking AI outputs, future systems could enable AI agents to validate their own work. This shift could unlock what he calls “hockey stick growth”—rapid, exponential acceleration in AI performance.
In practical terms, this means moving toward fully automated systems where AI is not just executing tasks but also monitoring and improving itself. The promise is enormous: faster development cycles, more accurate outputs, and AI that can function with minimal human intervention.
The Concept of “Hockey Stick” Growth in AI
“Hockey stick growth” is a term often used in startups and tech to describe a growth curve that starts slowly and then spikes dramatically. Embiricos believes AI development could follow a similar trajectory once human bottlenecks are removed.
By letting AI agents operate independently, developers anticipate that progress in AGI could accelerate far beyond current expectations. The combination of autonomous AI agents and human oversight only when necessary could revolutionize the speed at which AI systems improve.
Challenges to Fully Automated AI Workflows
Despite the potential, Embiricos cautions that creating fully automated AI systems is not straightforward. Each application—coding, content creation, or decision-making—requires a tailored approach. There’s no universal formula for letting AI self-validate without oversight.
However, the push for smarter agents is already reshaping OpenAI’s priorities. By focusing on removing the dependency on human prompts and reviews, researchers hope to create a more scalable AI infrastructure that accelerates progress toward AGI.
Implications for Codex and AI Developers
For Codex, OpenAI’s coding agent, this shift could be transformative. Developers could see AI taking on more complex coding tasks independently, reducing time spent on repetitive reviews and manual debugging. Embiricos’ vision points to a future where AI can handle multi-step workflows with minimal human input.
This evolution not only improves efficiency but also opens new possibilities in software development, data analysis, and beyond. By freeing humans from the slow task of reviewing outputs, AI can operate closer to its full potential.
Human Limitations in the Age of AI
Embircicos’ comments highlight a broader tension in AI development: humans are no longer the fastest link in the chain. As machines surpass human capabilities in computation and reasoning, human speed—whether typing, reviewing, or multitasking—becomes a limiting factor.
This has sparked discussions in the AI community about how to design systems that complement human abilities rather than depend on them. The goal is not to replace humans entirely but to offload the slowest and most repetitive tasks to AI agents.
Smarter AI Agents
The future, Embiricos suggests, involves AI agents capable of assessing their own work and collaborating with humans only when needed. This approach could dramatically accelerate progress toward AGI, making it less dependent on the current human workflow.
Experts believe this could lead to a new era in AI development, where human guidance becomes more strategic and less hands-on. Smarter agents could reduce bottlenecks across industries, from software to research, enabling faster and more reliable AI-driven solutions.
The Race to AGI Intensifies
OpenAI is not alone in this race. Companies across the tech world are striving to achieve AGI, and efficiency bottlenecks like human typing speed are becoming a key competitive consideration.
Those who figure out how to let AI agents self-validate could gain a significant edge. In a field where milliseconds matter and scaling solutions is crucial, removing human bottlenecks could define the next decade of AI progress.
Human-AI Collaboration Reimagined
Embiricos’ insight underscores a fundamental shift: the path to AGI isn’t just about smarter AI, it’s about rethinking human-AI collaboration. Faster typing alone won’t solve the problem—but designing systems where AI can operate independently might.
As research accelerates, one thing is clear: the next breakthroughs in AI may come less from humans typing faster and more from AI learning to manage itself. The human role in AGI will evolve, focusing on strategic oversight rather than manual prompting.
