What the Top AI Users Are Doing Differently

The disparity between advanced AI users and average users has widened dramatically, with advanced users leveraging AI eight times more effectively than their average counterparts. This gap has increased from 2.6 times to 8.3 times between January and June, indicating that average users have not significantly improved the value derived from their AI usage.

Meta is preparing to launch a consumer agent named Hatch, designed to offer a more streamlined experience. The company is also enhancing WhatsApp to better integrate third-party agents and plans to release a larger model called Watermelon in October, which has reportedly matched GPT-55 on internal benchmarks, although it may struggle to keep up with future advancements like GPT-56.

Grockbott's initial pricing strategy was confusing and high, leading to a limited launch. It is now part of the $60 a month Cursor Pro subscription and the $100 a month Super Grock subscription. OpenAI is reducing prices for GPT-56 Sol, making API access more affordable. Meanwhile, Huggingface is exploring acquisition opportunities at a $13 billion valuation, despite 97% of its users accessing the platform for free.

NVIDIA has invested $42.3 billion in private companies but faces supply chain constraints that limit growth in chip revenue. The company is also involved in multiple acquisitions and investments within the AI ecosystem. In a related incident, Taiwanese prosecutors charged nine individuals, including an NVIDIA manager, for chip smuggling linked to SuperMicro servers.

At Microsoft, there is significant variance in AI usage across departments, with some employees reporting monthly spends as high as $28,000. A study identified top performers, termed AI amplifiers, who effectively guide and refine AI outputs to enhance their value. The transition to AI-driven work paradigms is expected to cause personal disruptions that society needs to prepare for.

Sam Altman acknowledged underestimating the speed of AI disruption, suggesting that the transition may be smoother and slower than initially expected due to institutional inertia. Despite access to advanced AI tools, even advanced users experience inertia in their computer usage.

Non-technical roles have seen substantial growth in Codex usage, with finance and accounting increasing by 20 times, marketing and communications by 26 times, and legal by 108 times since February. The gap between leading AI users and general enterprise users continues to widen, with frontier firms now using 8.3 times more output tokens than typical firms, and 17 times more than a year and a half ago.

The fastest growth in agentic use cases is occurring among knowledge workers outside of software and engineering functions. OpenAI defines frontier firms as those in the top 10% of usage based on output tokens per active user. This increase in agentic usage reflects a growing understanding of patterns that allow agents to thrive in various contexts, with frontier firms showing higher engagement with plugins and skills compared to typical firms.

A significant shift is occurring in the type of work being done as users transition from chat-based AI to agentic AI, which enables systems-level work that impacts broader workflows. For example, writing constitutes 57% of legal work in chat but drops to 16.2% in agentic AI, while coding rises to 32.9% in agentic contexts, even among non-software engineers. By 2026, agents are expected to become more prevalent, with frontier firms leading the way in deploying agentic uses, although many agents still operate within individual silos.