Behdad Esfahbod (@behdadesfahbod)

Since late December I have shifted all my coding work to agentic development with OpenAI Codex and Claude AI coding agents. This has completely transformed how I work, mostly in English now, and I cannot imagine going back to the old days. Is as if I have two very capable employees now. I can pump features out at incredible speed, features that previously we didn't have resources to work on. I was going to write about my experience and learning so far, but Andrej nails it. I have experienced every point he is making. To answer two of his questions: - "What happens to the "10X engineer" - the ratio of productivity between the mean and the max engineer?" The gap is about to widen significantly. I believe AI coding will make the mean engineers disposable, while making max engineers multiple times more productive. - "Armed with LLMs, do generalists increasingly outperform specialists? LLMs are a lot better at fill in the blanks (the micro) than grand strategy (the macro)." I think specialists are going to gain from this revolution even more than generalists. Here's the thing: specialists are scarce, and hold decades of tribal knowledge the generalists *and* the AI lack. As such, specialists can "replicate" themselves by guiding the AI to perform much more than they could do by hand. Generalists are scalable by adding more of them to the task at hand. Not with specialists. I will write later about the problems I tackled with Codex & Claude. I have achieved three tasks, each within a day or two or three, whereas they would have taken weeks if not months before, if I could pull them off at all. The future is here. If you are software engineer and you have not embraced these technologies already, you are falling behind.