Margaret Atwood — the most famous living author — tried Claude once and concluded: “garbage in, garbage out.” Meanwhile, 49% of Americans use AI daily. Solopreneurs earning $10M tripled. OpenAI’s top users generate 60 hours of agent work per day. The gap between what AI’s critics experience and what AI’s users achieve has never been wider.
Margaret Atwood — Babell Literary Festival, Porto
“The thing about AI is that it’s garbage in, garbage out.”
What Atwood Said
At the inaugural Babell Literary and Cultural Festival in Porto, Portugal, Atwood told the audience she had used an AI model exactly once in her life — Anthropic’s Claude. Her conclusion: the technology “simply isn’t reliable enough for humans to depend on” and “even people who use it for business reasons have to check it because it makes mistakes.”
She’s not wrong. AI makes mistakes. But the statement reveals something more interesting than a critique of AI quality — it reveals the experience gap between someone who tried AI once and the people who use it every day.
The Data Says the Opposite
The key insight: Atwood used Claude once. The people generating $10M in solo revenue use it hundreds of times a day. The gap isn’t about AI quality — it’s about the harness. A single prompt to a chatbot produces garbage. A system of agents, loops, memory, and feedback produces 60 hours of parallel work per day. The tool is the same. The orchestration is everything.
The Structural Read
THE HARNESS IS THE DIFFERENCE
Atwood’s experience proves the Harness Theory thesis by inversion. Without a harness — without framing, without loops, without memory — AI is indeed garbage in, garbage out. With a harness, one person produces the output of a department. The model is the same. The orchestration is what separates Atwood’s experience from a $10M solopreneur’s.
SHE CHOSE CLAUDE — AND THAT MATTERS
Of all the models she could have tried, Atwood chose Claude — the model Polymarket gives 94.8% odds of being the best. Even the best model, used once without a harness, produced a “garbage” experience for the world’s most famous author. The tool doesn’t matter without the system around it.
THE 67% AGREE WITH HER — AND USE AI ANYWAY
67% of Americans say AI advances too fast. They’re skeptical, like Atwood. But 49% use it daily anyway. The paradox: people distrust AI’s societal impact while trusting it for their own productivity. Atwood represents the 51% who don’t use AI. The data represents the 49% who do — and who are compounding an advantage the 51% can’t see.
The Bottom Line
Margaret Atwood is one of the greatest writers alive. She tried AI once and called it garbage. She’s describing the experience of using a tool without a system — and she’s right that a single prompt to a chatbot is unreliable. But the people building $10M businesses, generating 60 hours of parallel agent work, and seeing 86% productivity gains aren’t using AI the way Atwood did. They’re not typing one prompt. They’re building harnesses — orchestration systems that compound. “Garbage in, garbage out” is the correct description of AI without a harness. The Harness Theory is the answer to why it doesn’t have to be.