Being responsible with Chinese AI hype
Lessons from Manus and the dangers of uncritical tech narratives
Over the weekend, a previously unknown Chinese AI company called Manus captured the attention of tech Twitter with bold claims of having developed “the first general AI agent.” Already some users took to Twitter to claim Manus would “replace 50% of all white collar jobs” within six months and was proof AGI would arrive this year. Even serious people like Tyler Cowen were taken in, calling Manus “for real, and ahead of its American counterparts.”
But then people’s hearts were broken. They learned that Manus was heavily using Claude models from Anthropic (a US company). Now Manus is “a huge marketing scam” and “fraud”.
This 48 hour hype cycle is a perfect case study in how AI narratives spiral out of control. And this whiplash-inducing narrative shift isn't just confusing - it actively undermines serious analysis of international AI competition.
Let's break down what's actually happening with Manus, how it fits into the established pattern of Chinese AI hype cycles, and what it tells us about the future of AI competition.
What Manus actually is
After examining available evidence from technical users who have investigated the system, Manus appears to be a product that extends Claude 3.6 models to include planning, web browsing, computer use, rudimentary memory, and code use to combine to accomplish tasks within a well-designed user interface.
Manus is neither pure hype nor a scam. Despite not being revolutionary, Manus does represent legitimate innovation in user interface design (appearing intuitive and well-executed), web agent capabilities (handling some real-world tasks that even OpenAI and Anthropic do struggle with), and the ability to coordinate multiple tools with relative ease. Nathan Labenz noted Manus offers “legit value on travel search and planning” with “smarter planning” than some competitors. Ethan Mollick described it as “a Claude wrapper, but a very clever one” capable of “some good stuff” despite typical agent limitations.
Manus marketing mania
What differentiates Manus isn't just its technology but its marketing approach. The rollout followed a familiar playbook:
Viral demo: A highly produced demo video showing impressive but potentially misleading capabilities
Limited access: Creating artificial scarcity with invitation codes
Influencer targeting: Providing early access to tech influencers and getting them to uncritically promote the product.
Benchmark focus: Promoting performance on a single benchmark (GAIA) with favorable comparisons to OpenAI and Anthropic, without revealing that the benchmark was essentially “gamed”.
This strategy created an invitation code market where codes reportedly sold for over $13,000 - remarkable for a company valued at only around $100 million.
As Twitter user Johns pointed out:
Actually, this product began a large-scale promotional campaign in China two days ago. Manus seems to have enlisted many Chinese AI influencers to praise it without any restraint, which sparked quite a discussion. However, most ordinary users still do not have access to it.
[…] Now they are conducting the exact same marketing operation on Twitter: only a few influencers have access, and they only offer praise, with no mention of any drawbacks.
Frankly speaking, after the release of deepseek, the whole world is prone to believe that there will be another outstanding Chinese AI product, and manus is exploiting this mindset.
A pattern of Chinese AI announcements
If this feels familiar, it should. We've been here before with DeepSeek in January - initial claims about a ‘$6M model beating billion dollar American AI’ generated frenzied headlines about the end of US AI dominance but didn’t hold up to sober scrutiny. This pattern of extreme hype has become a recurring feature of Chinese AI announcements.
In fact, the Manus episode fits into a recognizable pattern we've observed with Chinese AI announcements over the past several years: bold initial claims that a new Chinese AI system has dramatically leapfrogged Western capabilities are made with minimal information and technical transparency. There is heavy emphasis on a few specific benchmarks where the system performs well and minimal scrutiny around the context and meaning behind those benchmarks. This then comes with patriotic framing, positioning advancements as evidence of China's technological ascendance.
Manus appears to be deliberately exploiting this receptivity to Chinese AI hype. If an American startup with no resources did this, it would be a curiosity. When a Chinese startup does it, it becomes an international power story. As Zvi Mowshowitz aptly put it:
There's a lot of momentum in that [China beats US] narrative, and a lot of people want to push it. It's in the air. This is also partly due to other Chinese successes like TikTok and Temu, in general so many want to say China is winning.
The Real Implications
The most important aspect of the Manus story isn't about Manus itself, but what it signals about the evolving AI landscape:
Marketing will continue to outpace reality: AI startups know how to play the media like a fiddle. This is becoming increasingly sophisticated, with benchmark optimization and controlled rollouts designed to maximize publicity. The public is not prepared and the media is not helping.
Be particularly skeptical of initial claims of Chinese AI: The Chinese AI discourse tends to be unusually hype-driven.
There is an evaluation gap: These days, we have minimal way to distinguish genuine technical advances from clever marketing.
Demos don’t work to evaluate AI: If you could judge by demos alone, we would've had driverless cars a decade ago.
Wrappers that extend AI capabilities can still matter: AI agents that combine existing models with web access and tools around browser/computer use can deliver significant value even without breakthrough capabilities.
Breaking the Hype Cycle
What's particularly concerning is how predictable this cycle has become. From DeepSeek to Manus and likely future announcements, we see the same pattern: extraordinary claims generate headlines and investment interest, followed by technical reality checks that receive far less attention. This dynamic distorts our understanding of the actual state of AI competition and makes thoughtful policy responses more difficult.
For policymakers and analysts evaluating Chinese AI progress, several principles can help cut through the noise:
Wait for independent technical evaluation: Allow time for researchers to analyze systems before drawing conclusions. With Manus, like other AI systems, tuning out the entire first 24 hours and waiting another 24-48 hours was the right call. This will be true for other developments.
Focus on actual capabilities: Evaluate what systems can consistently do rather than cherry-picked demonstrations. Don’t trust demos.
Track deployment at scale: Assess whether achievements represent incremental improvements or genuine breakthroughs. Note that the ability to operationalize AI at scale remains a critical differentiator.
As we evaluate future Chinese AI announcements — and there will certainly be more — we should approach them with neither dismissive skepticism nor uncritical acceptance, but sober technical analysis.
And if you want more of that, that’s what I do. I’ll leave you my subscribe button here:
Appreciate the well balanced thoughts!
Chinese hype is actively hindering the AI safety cause. It's very hard to convince the general public that AI is a threat and should be handled with great care, when they see news of debunking yet another huge claim about a leap in AI. These crashes in excitement only prime people's tolerance and can blind them to actual threats. It will be the story of a boy who cried "paradigm shift".