Anthropic Developer Andrej Karpathy: "At OpenAI I saw two types of engineers: ones who write every instruction - ones who describe what they want The second type replaced the first in a year " the Karpathy framework - 3 eras of programming: 1.0 → you write the code →... takes weeks 2.0 → you feed data, model writes the algorithm → takes hours 3.0 → you describe what you want in English → takes minutes each era killed the last. most people are still in 1.0 "the best code is the code you never write. you don't program the solution - you program the data -the solution programs itself" he built this at OpenAI - now applies it at Anthropic - the framework never changed watch & save - guide about Software by Andrej ↓show more

i think he's ones who understand AI fully another smartest video by Karpathy, booked!

hehe, he's always give brilliant material and always try to share it

Most people are still in 1.0 is the line that stings the most honestly, easy to say you've moved on and still be writing every instruction manually. Which era are you actually operating in day to day? @0xCodila

4th type. I give a simple reference code that I write and ai uses as a reference to write more correct code. Less token waste.

Pure intent leverage.

ngl the 3.0 devs who cant explain WHAT they want still fumble tho the bottleneck moved from writing code to having a brain

I think soon enough we'll be at a point where our computers will know what we want before we actually do

love the comparison

Programming 3.0 doesn't eliminate engineering—it shifts the bottleneck. Less time writing code means more time validating outputs, defining intent, and building reliable systems. Have you found evaluation to be a bigger challenge than code generation in practice?

era 4.0 is just pointing at your screen and going 'you know' and the computer goes 'yeah' 💀

loop engineering > prompting

The best code is the code you never write is such a good line to actually sit with

This is the real shift. People who still treat AI like search box will lose to people who run it like a worker.

Soon we will have era 4 just explain in words

The framework is right, but "takes minutes" undersells the new bottleneck. When the coding takes minutes, the constraint shifts to knowing what to ask for and verifying what comes back. Era 3.0 didn't remove the skill — it moved it upstream.

Karpathy’s framework is spot on. Moving from strict instruction to high-level intent is how we scale.The real magic happens in the Outer Loop — when you transition from "prompting" to deploying autonomous looping systems that evaluate and self-correct on the fly. 🤖🔄

wild how 3.0 still rewards the people who understand 1.0 tho like you cant describe what you want if you dont know whats actually hard

Future of building entirely

most people are jumping right from 1 to 3 as 2 was not very accessible.

Drug dealer sells the drugs.

The three-eras framing is a useful mental model, though 'takes minutes' still oversells how much verification and iteration real production code needs.

nobody actually stopped writing 1.0 code though, it just moved under the stuff you describe

确实如此，不过能把“想要什么”精准描述出来，门槛其实比手撸代码高多了。 过去大家还能靠熟悉语法栈建立点护城河，现在 AI 把这层壁垒直接击穿了。当写代码的体力活被抹平，拼的完全就是架构感、问题拆解能力和审美。 以后纯粹的码农确实危险，但对那些有极强系统思维的人来说，现在的杠杆率简直是无限放大的。

The shift is real, but I think it's more about how you *describe* intent to AI. The best builders I know do both: they know the code when it matters and describe when speed matters.

the ones still writing every instruction just dont know theyre already replaced

This works if what you are doing is simple and doesn't require foresight. Each approach has merit and each approach deserves its day.

Interested

The transition to 3.0 is mostly about learning how to specify constraints better. The "describe what you want" part is actually the hard part now.

Love this shift. Describing what you want beats micromanaging every line. Here’s to smarter tooling and faster breakthroughs.