WEBVTT 00:00:00.000 --> 00:00:04.797 So I I've made this series of predictions on um code and software 00:00:04.880 --> 00:00:08.557 engineering and and and I think people have repeatedly kind of misunderstood 00:00:08.640 --> 00:00:12.477 them. So, so let me let me let me let me let me lay out the spectrum, right? like 00:00:12.560 --> 00:00:16.077 I think it was like you know like you know eight eight or nine months ago or 00:00:16.160 --> 00:00:18.077 something I said you know the AI model 00:00:18.160 --> 00:00:20.237 will be writing 90 90% of the lines of 00:00:20.320 --> 00:00:23.597 code in like you know 3 to 6 months 00:00:23.680 --> 00:00:27.277 which which happened at least at some places right happened happened at 00:00:27.360 --> 00:00:31.517 Antropic happened with many people downstream using our models but but 00:00:31.600 --> 00:00:36.237 that's actually a very weak criterion right people thought I was saying like 00:00:36.320 --> 00:00:40.317 we won't need 90% of the software engineers those things are worlds apart 00:00:40.400 --> 00:00:43.037 right like I would put the spectrum as 00:00:43.120 --> 00:00:47.997 90% of code is written by the model, 100% of code is written by the model and 00:00:48.080 --> 00:00:49.837 that's a big difference in productivity. 00:00:49.920 --> 00:00:53.917 Um 90% of the end toend SWE tasks right 00:00:54.000 --> 00:00:58.237 including things like compiling including things like setting up 00:00:58.320 --> 00:01:00.397 clusters and environments, testing 00:01:00.480 --> 00:01:03.117 features, writing memos. 90% of the 00:01:03.200 --> 00:01:08.157 tasks are written by the models. 100% of today's SUI tasks are are are written by 00:01:08.240 --> 00:01:11.197 the models. and and even when when when that happens doesn't mean software 00:01:11.280 --> 00:01:14.957 engineers are out of a job like there's like new higher level things they can do 00:01:15.040 --> 00:01:19.197 where they can they can manage and then there's a further down the spectrum like 00:01:19.280 --> 00:01:24.157 you know there's 90% less demand for SWES which I think will happen but like 00:01:24.240 --> 00:01:27.597 this is this this is a spectrum and you know I wrote about it in in the 00:01:27.680 --> 00:01:31.037 adolescence of technology where I went through this kind of spectrum with 00:01:31.120 --> 00:01:34.237 farming um uh uh and so I I actually 00:01:34.320 --> 00:01:38.397 totally agree with you on that it's just these are very different benchmarks from 00:01:38.480 --> 00:01:41.837 each other but we're proceeding through them super fast. It seems like in part 00:01:41.920 --> 00:01:43.117 of your vision it's like going from 90 00:01:43.200 --> 00:01:45.917 to 100. Um first it's going to happen 00:01:46.000 --> 00:01:49.917 fast and two that somehow that leads to 00:01:50.000 --> 00:01:54.157 huge productivity improvements. Um whereas when I noticed even in green 00:01:54.240 --> 00:01:57.437 field projects that people start with cloud code or something people report 00:01:57.520 --> 00:02:01.677 starting a lot of projects and I'm like do we see in the world out there a 00:02:01.760 --> 00:02:04.797 renaissance of software all these new features that wouldn't exist otherwise 00:02:04.880 --> 00:02:08.477 and at least so far it doesn't seem like we see that and so that does make me 00:02:08.560 --> 00:02:12.957 wonder even if even if like I never had to intervene on cloud code um there is 00:02:13.040 --> 00:02:16.397 this thing of like there's just the world is complicated jobs are 00:02:16.480 --> 00:02:20.317 complicated and closing the loop on self-contained 00:02:20.400 --> 00:02:23.277 systems whether it's just writing software or something, how much sort of 00:02:23.360 --> 00:02:26.077 how much broader gains we would see just 00:02:26.160 --> 00:02:30.477 from that. And so maybe that makes us this should dilute our estimation of the 00:02:30.560 --> 00:02:33.837 country of geniuses. >> I well well I actually I I like I like 00:02:33.920 --> 00:02:35.677 simultaneously 00:02:35.760 --> 00:02:38.957 I simultaneously agree with you agree 00:02:39.040 --> 00:02:43.037 that it's a reason why these things don't happen instantly, but at the same 00:02:43.120 --> 00:02:48.157 time I think the the the effect is going to be very fast. So like I don't know 00:02:48.240 --> 00:02:49.757 you could have these two polls, right? 00:02:49.840 --> 00:02:52.957 one is like um you know AI is like you 00:02:53.040 --> 00:02:56.797 know it's not going to make progress it's slow like it's going to take you 00:02:56.880 --> 00:03:00.317 know kind of forever to diffuse within the economy right economic diffusion has 00:03:00.400 --> 00:03:03.917 become one of these buzzwords that's like a a reason why we're not going to 00:03:04.000 --> 00:03:07.517 make AI progress or why AI progress doesn't matter and and you know the 00:03:07.600 --> 00:03:10.957 other axis is like we'll get recursive self-improvement you know the whole 00:03:11.040 --> 00:03:14.557 thing you know can't you just draw an exponential line on the on the curve you 00:03:14.640 --> 00:03:17.837 know it's it's we're going to have you know Dyson spheres around the sun and 00:03:17.920 --> 00:03:20.077 like you 00:03:20.160 --> 00:03:24.877 you know so many nanconds after uh you know after after we get recursive I mean 00:03:24.960 --> 00:03:28.477 I'm completely caricaturing the view here but like you know there there there 00:03:28.560 --> 00:03:30.877 are these two extremes but what we've 00:03:30.960 --> 00:03:33.597 seen from from the beginning you know at 00:03:33.680 --> 00:03:35.517 least if you look within anthropic 00:03:35.600 --> 00:03:39.517 there's this bizarre 10x per year growth 00:03:39.600 --> 00:03:41.837 in revenue that we've seen right so you 00:03:41.920 --> 00:03:47.357 know in 2023 it was like zero to 100 million 2024 it was 100 million to a 00:03:47.440 --> 00:03:51.597 billion 2025 it was a billion to like 9 or 10 00:03:51.680 --> 00:03:53.597 billion. If you enjoyed this clip, you 00:03:53.680 --> 00:03:59.600 can watch the full episode here and subscribe for more clips. Thanks.