# Dario Amodei On How His AI Predictions Have Been Misunderstood *Dwarkesh Clips* ยท https://www.youtube.com/watch?v=d3x-n0C3-a4 ## Summary Dario Amodei explains that his prediction that AI would write 90% of code was widely misunderstood as meaning software engineers would become unnecessary. He lays out a spectrum of milestones from 90% of code to 100% of end-to-end SWE tasks to reduced demand for engineers, arguing progress through them is fast but that engineers will shift to higher-level work. The interviewer pushes back that no software renaissance is visible yet, and Amodei responds that diffusion takes time but the effect will still be very fast, citing Anthropic's roughly 10x annual revenue growth. ## Key points - Dario Amodei says his prediction that AI models would write 90% of code within 3 to 6 months was widely misunderstood as meaning 90% of software engineers would no longer be needed. - He lays out a spectrum of benchmarks: 90% of code written by models, 100% of code, 90% of end-to-end SWE tasks, 100% of today's SWE tasks, and eventually 90% less demand for SWEs. - Even when models write all of today's software engineering tasks, he argues engineers will move to new higher-level work like managing systems. - The interviewer notes that even in greenfield projects using tools like Claude Code, there is no visible renaissance of new software features, which he says should dilute estimates of a 'country of geniuses' effect. - Amodei agrees diffusion takes time but insists the effect will still be very fast. - He contrasts two extreme views: that AI will diffuse too slowly to matter, and that recursive self-improvement will lead instantly to Dyson spheres. - He cites Anthropic's revenue growth as evidence of speed: roughly zero to 100 million in 2023, 100 million to a billion in 2024, and a billion to 9 or 10 billion in 2025. ## Notes ### Misunderstood predictions - Dario Amodei says people repeatedly misunderstood his prediction that AI models would write 90% of code in 3 to 6 months. - He notes this happened at least in some places, including at Anthropic and among downstream users of their models. - He stresses that "90% of code written by the model" is a weak criterion and does not mean 90% of software engineers are no longer needed โ€” those are "worlds apart." ### The spectrum of benchmarks - Amodei lays out a progression: 90% of code written by the model, then 100% of code, which is a big productivity difference. - Further along: 90% of end-to-end SWE tasks (compiling, setting up clusters and environments, testing features, writing memos), then 100% of today's SWE tasks. - Even at 100% of today's tasks, software engineers are not out of a job โ€” they move to new higher-level work like managing. - The end of the spectrum is 90% less demand for SWEs, which he thinks will happen. - He says he wrote about this spectrum with farming in "the adolescence of technology." ### The interviewer's pushback - The interviewer observes that even in greenfield projects started with Claude Code, there is no visible renaissance of new software features that wouldn't otherwise exist. - He suggests the world and jobs are complicated, and closing the loop on self-contained systems may yield limited broader gains โ€” which should dilute estimates of the "country of geniuses." ### Amodei's response: fast but not instant - He simultaneously agrees diffusion takes time and insists the effect will be very fast. - He caricatures two extremes: "economic diffusion" as a buzzword for why AI progress won't matter, versus drawing an exponential line to Dyson spheres after recursive self-improvement. - As evidence of speed, he cites Anthropic's "bizarre 10x per year" revenue growth: zero to 100 million in 2023, 100 million to a billion in 2024, and a billion to 9 or 10 billion in 2025. ## Transcript [00:00] So I I've made this series of predictions on um code and software engineering and and and I think people have repeatedly kind of misunderstood them. So, so let me let me let me let me let me lay out the spectrum, right? like I think it was like you know like you know eight eight or nine months ago or something I said you know the AI model will be writing 90 90% of the lines of code in like you know 3 to 6 months which which happened at least at some places right happened happened at Antropic happened with many people downstream using our models but but that's actually a very weak criterion right people thought I was saying like we won't need 90% of the software engineers those things are worlds apart [00:40] right like I would put the spectrum as 90% of code is written by the model, 100% of code is written by the model and that's a big difference in productivity. Um 90% of the end toend SWE tasks right including things like compiling including things like setting up clusters and environments, testing features, writing memos. 90% of the tasks are written by the models. 100% of today's SUI tasks are are are written by the models. and and even when when when that happens doesn't mean software engineers are out of a job like there's like new higher level things they can do where they can they can manage and then there's a further down the spectrum like you know there's 90% less demand for SWES which I think will happen but like [01:24] this is this this is a spectrum and you know I wrote about it in in the adolescence of technology where I went through this kind of spectrum with farming um uh uh and so I I actually totally agree with you on that it's just these are very different benchmarks from each other but we're proceeding through them super fast. It seems like in part of your vision it's like going from 90 to 100. Um first it's going to happen fast and two that somehow that leads to huge productivity improvements. Um whereas when I noticed even in green field projects that people start with cloud code or something people report starting a lot of projects and I'm like do we see in the world out there a renaissance of software all these new [02:03] features that wouldn't exist otherwise and at least so far it doesn't seem like we see that and so that does make me wonder even if even if like I never had to intervene on cloud code um there is this thing of like there's just the world is complicated jobs are complicated and closing the loop on self-contained systems whether it's just writing software or something, how much sort of how much broader gains we would see just from that. And so maybe that makes us this should dilute our estimation of the country of geniuses. >> I well well I actually I I like I like simultaneously I simultaneously agree with you agree that it's a reason why these things don't happen instantly, but at the same time I think the the the effect is going [02:45] to be very fast. So like I don't know you could have these two polls, right? one is like um you know AI is like you know it's not going to make progress it's slow like it's going to take you know kind of forever to diffuse within the economy right economic diffusion has become one of these buzzwords that's like a a reason why we're not going to make AI progress or why AI progress doesn't matter and and you know the other axis is like we'll get recursive self-improvement you know the whole thing you know can't you just draw an exponential line on the on the curve you know it's it's we're going to have you know Dyson spheres around the sun and like you you know so many nanconds after uh you know after after we get recursive I mean [03:24] I'm completely caricaturing the view here but like you know there there there are these two extremes but what we've seen from from the beginning you know at least if you look within anthropic there's this bizarre 10x per year growth in revenue that we've seen right so you know in 2023 it was like zero to 100 million 2024 it was 100 million to a billion 2025 it was a billion to like 9 or 10 billion. If you enjoyed this clip, you can watch the full episode here and subscribe for more clips. Thanks.