A Stanford CS student asked ChatGPT to fix a bug in his code. It gave him five things to check. None of them was the bug.
He decided AI was overhyped. Closed the tab. Told his classmates it was useless for real work.
His roommate pasted the exact same code into a different box. Answer in 90 seconds. The bug was in how the gradients accumulated. One line.
Same prompt. Same company. Different model.
Here is what nobody checks. When you type into the free box you are not talking to `ChatGPT` You are talking to a stripped-down version. Fewer parameters. Weaker memory. The flagship that thinks for a full minute before it answers sits behind a $200/month door. Most people never open it.
The roommate pays the $200. The student pays $0 and concludes the technology doesn't work.
→ The edge isn't the AI. The edge is knowing which model you're actually talking to.
Second thing nobody knows. The model is a lossy zip file of the internet, frozen about 6 months ago. 1 trillion parameters. Its memory is vague on purpose. The context window is its working memory. Stuff it with junk and it gets distracted, slower, dumber.
The people getting clean answers start a fresh chat the second they switch topics. The people getting garbage paste their whole life into one thread and wonder why it forgets.
While everyone argues whether AI is a bubble, a handful of people quietly route the hard prompts to the thinking model, keep the context clean, and walk out with the answer in 90 seconds.
Same box. Same words typed in. The reply depends entirely on what question - and what model - you actually picked.
2022: new hire joins a team.
3 weeks of onboarding. Asking senior engineers the same questions. Reading wikis nobody updated since 2019. Reviewing git blame line by line.
2024: same hire, same codebase.
Opens a terminal. Types one question. 2-3 days. Done.
That's Claude Code.
The guy who built it - Boris, Anthropic staff - dropped something interesting at a recent talk.
He said by end of year, people probably won't use IDEs anymore.
Not as a hot take. As a quiet observation. Based on watching the model improve from the inside.
Meanwhile Cursor raised $100M. Copilot has 2 million paid users. Everyone's building AI on top of the IDE.
And the guy who built the most-used agentic coding tool in the world ships from a terminal.
No UI layer. No indexing. Your code never leaves your machine.
Just: type what you want -> Claude does it -> paste the git log into standup.
The twist: they built it as a CLI because too many IDEs at Anthropic. They kept it as a CLI because the model's getting good enough to not need a UI.
80% of Anthropic engineers use it daily. Researchers run it on notebooks. New hires use it before they know the codebase.
-> It's free to try. One command.
@0xRasm doesn't that assume builders are always better off focusing on the high-level stuff, when sometimes the devil's in the details that terminal agents are handling?
2022: new hire joins a team.
3 weeks of onboarding. Asking senior engineers the same questions. Reading wikis nobody updated since 2019. Reviewing git blame line by line.
2024: same hire, same codebase.
Opens a terminal. Types one question. 2-3 days. Done.
That's Claude Code.
The guy who built it - Boris, Anthropic staff - dropped something interesting at a recent talk.
He said by end of year, people probably won't use IDEs anymore.
Not as a hot take. As a quiet observation. Based on watching the model improve from the inside.
Meanwhile Cursor raised $100M. Copilot has 2 million paid users. Everyone's building AI on top of the IDE.
And the guy who built the most-used agentic coding tool in the world ships from a terminal.
No UI layer. No indexing. Your code never leaves your machine.
Just: type what you want -> Claude does it -> paste the git log into standup.
The twist: they built it as a CLI because too many IDEs at Anthropic. They kept it as a CLI because the model's getting good enough to not need a UI.
80% of Anthropic engineers use it daily. Researchers run it on notebooks. New hires use it before they know the codebase.
-> It's free to try. One command.
@6uappi onboarding used to be so tedious, now it's all about streamlining the process and getting people up to speed quickly, what's the most surprising thing you've learned from adopting this new approach
@0xWast3 building something that actually makes money in a short timeframe is a whole different story than just throwing code together, what's the most important thing to focus on when getting started with this approach
@gippp69 vibe coding's impact is all about simplicity and focus, letting one person create something that resonates with a specific group of people, no need for a huge team or complicated tech.