The recent launch of the Chinese AI model DeepSeek-V3 caused alarm in tech circles, suggesting China was months closer to achieving artificial general intelligence than had been previously thought. Tech stocks tumbled and President Trump described it as a “wake-up call” for America’s AI industry.
More intriguingly, however, is this month’s comparatively quiet debut of Manus, a Chinese general-purpose AI agent. The demo for this fully autonomous agent shows it carrying out a range of tasks, including buying property, booking holidays and developing video games. Analysts have dismissed it as “merely a wrapper” because it builds off models by Western AI companies like Anthropic. Yet, as the AI researcher Dean Ball has observed, Manus represents the state-of-the-art in agentic product development, whereas Deepseek remains inferior to cutting-edge Western models across most key performance metrics.
This highlights a blind spot in much of the policy and media discourse on AI. Our conversations overwhelmingly focus on innovation and capability breakthroughs, while neglecting the equally vital challenge of actually delivering AI solutions to businesses and consumers. We risk winning the research race but losing the implementation marathon....