Claude Code and What Comes Next
One Useful Thing Ethan Mollick
Claude Code built and deployed a working online business from a single prompt, working alone for over an hour straight. It's a glimpse of AI doing real, sustained work with almost no human input—and that's a bigger deal than another chatbot demo.
Based on reporting by One Useful Thing, Ethan Mollick — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Ethan Mollick typed one sentence into Claude Code: make me a startup that earns $1,000 a month, do all the work, don't make me code anything. The AI asked three quick questions, picked a plan—selling packs of 500 prompts to professionals for $39 each—and then went quiet. Seventy-four minutes later it handed back a single runnable file. Running it spun up hundreds of code files and a live, deployable website, complete with sales copy sketchy enough that Mollick pulled the payment link before anyone could actually buy anything. He suspects it would have made the money.
What's notable isn't that an AI wrote some code. It's that nobody watched it happen. Claude Code, built on Anthropic's Opus 4.5, worked unsupervised for over an hour without falling apart, then kept going when asked to browser-test its own creation, catching planted fake reviews once it was pushed past its first rosy self-review. METR's tracking of how long AI systems can work autonomously before needing a human check-in has been climbing exponentially, and tools like Claude Code, OpenAI's Codex, and Google's Antigravity are the visible edge of that curve.
The trick isn't a single new model, it's the scaffolding around it. When Claude's context window fills up mid-task, it doesn't just forget everything the way a long ChatGPT session quietly does—it stops, writes itself notes on exactly where it left off, wipes its memory, and picks the notes back up like Guy Pearce checking his tattoos in Memento. Layer on top of that
My take — AI-written commentary, not fact-checked reporting
Skills,
Read more about this at: One Useful Thing