Anthropic’s Andrej Karpathy says “the external world has seen the preview” of what AI can actually do
What happened
Andrej Karpathy, now at Anthropic and previously a co-founder of OpenAI, pointed out a stark divide in real-world AI exposure. While about 75 percent of the global population has barely interacted with large language models, a very narrow group—around 5,000 people—have access to advanced internal AI systems. These systems use agent swarms to solve problems that would take top human experts years, giving those with access a massive productivity advantage. Meanwhile, the public mostly sees chatbots that showcase only a fragment of AI’s actual capabilities.
Why it matters
This uneven access creates a sharp knowledge and capability gap between insiders and the rest of the world. Most businesses and users outside this elite group won’t experience AI’s full potential anytime soon. That means the hype around AI chatbots overpromises what the general public can currently do with AI. Builders and investors must recognize the pressure on AI developers to move beyond demo chatbots and deliver tools that handle complex tasks across industries. The risk for outside operators is being caught in an outdated narrative of AI as a novelty conversation partner, while actual AI-driven workflow automation advances rapidly in closed environments.
What to watch next
Tracking when and how broader access to these internal frontier AI systems expands will reveal shifts in market power. Watch Anthropic and other cutting-edge AI labs for moves to democratize more advanced agent technologies. Also monitor whether new user-facing tools start reflecting the efficiency gains agent swarms provide internally. If this gap closes, expect increased automation on expert-level workflows and pressure on companies to adopt AI faster to keep up. If not, expect persistent skepticism and slow adoption outside the AI elite.
AI Quick Briefs Editorial Desk