@hurls

Dwarkesh just dropped his thoughts on the state of AI for December 2025, and he’s calling the top. He’s moderately bearish. He argues that because these models can’t "learn on the job" like a human, AGI is much further away than we thought. https://www.dwarkesh.com/p/thoughts-on-ai-progress-dec-2025 He’s right but it doesn’t matter. I’m a developer. I love solving problems, but I've always hated typing. I hate syntax and I don't know much about algorithms. Dwarkesh sees the current state of AI as a "modern expert system," implying that it’s brittle but from where I’m sitting, its set me free. If you looked at my workflow in January of this year, I was tab-completing lines of code. It was faster, sure, but I was still in the weeds. Now the game has fundamentally changed. I don’t tab-complete anymore. I don’t even really "write" code. I assign agents. I break a complex task into features. I hand them off. The agents go do the work. I review the code, not to check for syntax errors (Opus is a better programmer than I will ever be) but to check the logic. The agent is the virtuoso musician, but I am the conductor. The model is still missing that key spark of human intelligence. It can build the feature, but it doesn't know why the user needs that feature. That's why I still have a job. This time next year I bet I won't be typing as much and talking more. I want the full Minority Report experience. I want to stand in front of a screen, dictating changes, and watching the software assemble itself in real-time. I’ll be building things in a day that used to take a week. I will be absolutely unstoppable. If I am 10x more productive, so are the researchers at OpenAI and Anthropic. So are the students studying deep learning. So are the engineers building the next generation of hardware. We aren't just using these tools to build CRUD apps; we are using them to build the next version of the tools themselves. LLMs are accelerating the frontier of knowledge because they are removing the friction between "having an idea" and "testing that idea."

1 reply

0 recast

5 reactions