Think AI is replacing workers? The reality is more ironic: humans labor behind the scenes to build AI meant to replace them. The gap between hype and reality reveals uncomfortable truths about innovation, investment bubbles, and the actual human cost of chasing artificial intelligence dreams.
Salesforce CEO Marc Benioff learned that $1 billion in local donations doesn't buy permission to call federal troops into San Francisco—not when 25-year friendships end and fellow billionaires publish op-eds on moral laundering.
Good Morning from San Francisco, Salesforce answered an RFP you didn't know existed. The pitch: use AI to
Explore Manus AI, a ground-breaking solution for reducing chicken consumption. Discover our innovative system to promote healthier food choices.
The future of AI can't order a chicken sandwich.
Manus, the latest "agentic" AI platform, launched last week to thunderous applause. Its Discord server swelled to 138,000 members. Invite codes sell for thousands on Chinese resale apps. Everyone wants a taste of the future. Too bad the future can't taste fried chicken.
"Given a simple task – ordering a chicken sandwich from a nearby restaurant – Manus needed two attempts. First try: system crash. Second try: it found the menu but couldn't figure out how to pay," reports TechCrunch's Kyle Wiggers. At least it understood what a sandwich was. Progress comes in small bites.
The Butterfly Effect, the Chinese company behind Manus, promises their AI can handle everything from real estate purchases to video game programming. They neglected to mention it struggles with DoorDash.
Flight booking proves equally challenging. Asked to find a business-class ticket from NYC to Japan, Manus responds with broken links to airline websites. It's like having a travel agent who only knows how to use Google, poorly.
Restaurant reservations? Failed. Building a Naruto-inspired fighting game? System error after 30 minutes. Table for one turns into error code for two.
The platform combines existing AI models, including Anthropic's Claude and Alibaba's Qwen. Think of it as a technological turducken – several AIs stuffed inside each other, pretending to be something new.
Research lead Yichao "Peak" Ji claims Manus outperforms OpenAI's deep research on GAIA benchmarks. Meanwhile, users report endless loops and error messages. It's like bragging about a car's speed while it's stuck in the garage.
Alexander Doria, co-founder of AI startup Pleias, documented his struggles with the platform. Others noted its tendency to make factual mistakes and skip citations. It's an AI that's confident but wrong – the digital equivalent of that uncle at Thanksgiving.
Manus seems to be Claude 3.7: "Human:" and "Assistant:" creates a prompt injection and it get stuck in neverending loop. https://t.co/g2tHYvmaWf pic.twitter.com/UuCnzIFux5
— Alexander Doria (@Dorialexander) March 9, 2025
Chinese media celebrated Manus as "the pride of domestic products." AI influencers spread tales of its capabilities, including sharing a video of desktop-to-smartphone operations. Ji later confirmed the video wasn't actually Manus. Apparently, even AI has stunt doubles.
Some tried comparing Manus to DeepSeek, another Chinese AI company. But DeepSeek developed its own models and shared them openly. Manus keeps its technology hidden, like a magician who won't show how the trick works. Probably because there isn't one.
The platform's popularity isn't surprising. The tech world loves a good story. This one has everything: artificial intelligence, international competition, and thousands fighting for invite codes. It's Silicon Valley soap opera, just with more error messages.
"As a small team, our focus is to keep improving Manus and make AI agents that actually help users solve problems," a Manus spokesperson told TechCrunch via DM. In other words: they're still figuring out how to make it work.
The closed beta is supposedly meant for stress-testing. Mission accomplished – users are definitely stressed.
Critics point out that Manus represents a familiar pattern in AI development: promise everything, deliver something less than everything. Much less. It's the tech equivalent of ordering a feast and getting a menu.
The company claims they're working to scale computing capacity and fix reported issues. Meanwhile, invite codes keep selling, hype keeps building, and somewhere, a chicken sandwich remains unordered.
Industry experts suggest this highlights the gap between AI ambition and reality. The world wants autonomous agents that can navigate reality. What it has is sophisticated software that gets confused by fast food menus.
The situation mirrors previous AI launches: big promises, limited delivery, and enough hype to float a small continent. Manus joins a long line of "revolutionary" technologies that turned out to be evolutionary at best.
To be fair, The Butterfly Effect acknowledges they're in early access. Every new technology needs time to mature. But usually, you start with the basics – like completing purchases – before promising to revolutionize human-machine interaction.
Why this matters:
Marcus Schuler
Marcus Schuler
Wikipedia discovered an 8% decline in human visitors after improving bot detection—while the AI platforms trained on its content now answer questions without sending anyone to the source. The sustainability model breaks when citations don't include clicks.
Microsoft wants Windows 11 users to talk to their PCs and let AI agents manipulate files. Cortana failed at exactly this a decade ago, and corporate IT won't enable experimental agents that create new attack surfaces. Third time's the charm?
Oracle booked $65B in AI cloud deals—including Meta—in 30 days, but the stock slipped after-hours despite raised 2030 targets. Investors want proof on 30–40% margins, not bookings. Demand's real. Execution's unproven. The scoreboard moves there.
Anthropic wired Claude into Microsoft 365, chasing institutional memory while Microsoft hedges with model choice inside Copilot. The fight isn't about chat features—it's about who mediates your company's knowledge day after day.
Get the 5-minute Silicon Valley AI briefing, every weekday morning — free.
