Adamu Shehu

This post has millions of views and I understand why. People are curious and a little scared about AI. But the way this article frames what's happening is deeply misleading, and people deserve better than that. AI is going to change every industry. It's another revolution, bigger and faster than the advent of computers and the internet You should absolutely prepare for what's coming. But the current AI stack, at least the models from the top labs that we have access to, is still far from self-operating or AGI level. Matt's article presents these tools as nearly flawless. They are not. Matt says he walks away from his computer for four hours and comes back to perfect, finished work with no corrections needed. I build with these tools every day. It does not work like that. I was coding with Opus 4.6 recently, which is in my opinion and many others' the best coding model out there right now. It could not fix a spacing issue in my Tailwind code. After several prompts I had to go in and change the code myself. Before you try to tell me I'm bad at prompting, just ask and I'll show you what I've built with these tools. These models hallucinate. They will confidently give you wrong answers with a straight face. They will delete your work and not think twice about it. They suffer from sycophancy, which means they will agree with you and tell you what you want to hear even when they're wrong. Anthropic themselves recently released a video discussing this exact problem in their own models. These are fundamental issues still being solved. Matt talks about AI having "judgment" and "taste" now. Those are big words with zero evidence behind them. What I see daily is a tool that is incredibly powerful but still makes decisions any junior developer would catch. He selectively cites METR's data to show AI completing longer tasks but conveniently leaves out that the accuracy bar for those benchmarks is 50%, not 100%. And it's specific to coding tasks, not general work. He picked the numbers that support his story and left out the ones that don't. He quotes Dario Amodei's predictions about job displacement as near-certainty. But Amodei runs an AI company. He has a financial interest in you believing AI is the most transformative thing ever. He might not be wrong, but you should weigh his predictions accordingly instead of treating them as gospel. Matt frames "AI helped build itself" like some sci-fi intelligence explosion. What actually happened is OpenAI used their own tools during development. That's like saying Microsoft Word helped write itself because someone at Microsoft used Word to write documentation. Technically true, deliberately misleading. And here's the part that says more than anything in the article. At the very end, Matt thanks three humans for reviewing his drafts. Not AI. Humans. If AI can truly do everything he claims, why did he need humans to review a blog post? The AI labs themselves are still figuring things out as they go. That's why a lab like OpenAI is still investing heavily in side quests like Sora and image generation. If AGI was truly around the corner, why would they be doing that? Your job will probably look very different in the near future. But that future is not here yet and you still have time. The people who will thrive are not the ones panicking because of a viral post. They're the ones learning to work with these tools right now while understanding where they fall short. There are great, honest resources out there by Anthropic, Thinking Machines and others that actually explain what AI can and can't do without the fear tactics. Read those instead. Prepare, yes. Panic, no.