I'm not so sure about Dario's Davos Predictions

00:00:00: At this year's World Economic Forum in Davos, Dario

00:00:03: Amodei predicted that within the next 12

00:00:07: months or so, AI would be able to write

00:00:11: all code, fully automated, on its own, essentially.

00:00:15: And it's worth paying attention to what this man has to say and I'll

00:00:19: also share my thoughts on it and why I think that you should take a

00:00:23: nuanced look at that. Not just because Dario, of course,

00:00:27: is the CEO of Anthropic which is one of the most important players in

00:00:30: generative AI space, especially when it comes to models related to

00:00:34: coding. But it's also worth paying attention

00:00:38: year, in the beginning of 2025, he predicted

00:00:42: that AI would be able to write 90% of all code

00:00:45: within the next three or six months and depending on how

00:00:49: you look at that, he was not entirely wrong.

00:00:52: Now, sure, this timeline is probably, or was

00:00:56: probably a bit too aggressive. It maybe wasn't three or

00:01:00: six months, more like six to nine months, not sure.

00:01:04: And, of course, it totally depends on your field, so in

00:01:08: which area you're programming, also in the programming language you're

00:01:12: It will depend on your company, the policies you have in your company.

00:01:16: And of course, it also depends on, on you, your

00:01:19: preferences, your experiences with AI.

00:01:22: I can say that for me, I am at a point I would say since

00:01:26: maybe December, November, where for some

00:01:30: projects, most of my projects, AI does write

00:01:34: 80, 90% of all code for me probably.

00:01:38: But this is now where things get interesting.

00:01:42: I'm not talking about wipe coding and I'm not

00:01:45: talking about AI doing it all on its own and I'm also not

00:01:49: talking about AI getting it all right.

00:01:53: Instead, things are way more difficult than that.

00:01:56: So it does absolutely not do that

00:02:00: on its own. It also is

00:02:04: not about wipe coding here. I mean,

00:02:07: you can probably wipe code but I've shared my view on that before

00:02:11: me, by the way, just to make sure we all got the same

00:02:15: really means that you don't look at the code at all, that you don't care about the

00:02:18: code, that you don't necessarily even know how

00:02:21: So yeah, this is also not about wipe

00:02:25: coding here. Instead, I would say we are

00:02:29: at a point where you can hand off many tasks

00:02:33: to AI with a good plan. So

00:02:37: if you have a good plan, AI can

00:02:40: implement it. So I think we're at that

00:02:44: point right now. You, you can definitely do that, at least

00:02:48: for some tasks. But of course, you wanna

00:02:52: review that code and, uh, I know there are

00:02:56: people that tell you you don't need to do that anymore. Might work for them.

00:03:00: Definitely not working for me and if you're shipping anything, if you're

00:03:04: shipping anything to customers, you're responsible for that.

00:03:07: You, as a developer, you can't say, "Oh, the AI got that wrong." No, it's

00:03:12: your responsibility and I won't take responsibility in code

00:03:16: I haven't reviewed, I don't understand.

00:03:20: Also, still, AI makes many

00:03:23: mistakes, so I need to fix mistakes

00:03:27: or steer AI in the right

00:03:31: direction.

00:03:34: And that is also very important. That's, of course,

00:03:38: important to the planning part or related to the

00:03:42: planning part, that you steer AI in the right direction.

00:03:44: But it's also related to what you do with code AI gives you.

00:03:48: So very far away from AI

00:03:52: generating 90% of the code and that means I have

00:03:56: no work to do. Far away from that. It just means that I

00:04:00: got a very fast typer who can

00:04:03: implement my plans, but those plans need to be good

00:04:07: and the output is kind of varying in quality.

00:04:11: Still, for me, it's very likely faster than

00:04:15: it all from scratch. If I instead try to build good plans,

00:04:19: split that up into smaller chunks, let AI write the code for me and

00:04:23: then review and fine-tune that code.

00:04:24: And by the way, with planning, I mean really detailed

00:04:28: plans where I also break down the exact libraries I wanna use, the

00:04:32: patterns I wanna use, the architecture of the software I wanna

00:04:35: step-by-step. So it's not a, a rough plan or a general plan; it's

00:04:39: a very detailed plan. That works for me and therefore, for

00:04:43: me, with, uh, with these restrictions here in mind, I

00:04:47: would say, yeah, it can probably write

00:04:50: 90%-ish of the code. But that does not

00:04:54: mean that it does 90% of the work on its own.

00:04:56: Now, that's important context because that's

00:05:00: to last year's statement by Dario.

00:05:03: So yeah, I would say we are kind of there, but not in the sense of AI doing

00:05:07: it all on its own. Now, what about this year's

00:05:11: statement? Though, in this year, Dario essentially said that AI will be

00:05:15: able to do what software engineers do and write software fully on its own,

00:05:19: end-to-end, within six to 12 months.

00:05:22: That's this year's statement. Now, I fully recommend

00:05:25: watching this entire talk. It's very interesting, but I have some

00:05:29: thoughts about this statement. And obviously, just to make

00:05:33: and obvious, I'm in no way smarter or more capable

00:05:37: in judging the performance of AI models than Dario.

00:05:41: But I'm also not the CEO of a company that needs to

00:05:45: sell these AI models and I can share from my own

00:05:48: experiences. Now, as I said, I would agree with last year's

00:05:52: statement to some degree, but with many caveats, with many

00:05:56: restrictions. The AI is definitely not writing 90% of my code

00:06:00: its own. So naturally, I have a very hard time

00:06:04: imagining... that this will come true, and I have a very hard time

00:06:08: imagining that this will come true, not just within the next 6 or 12

00:06:11: months, but anywhere in the near future.

00:06:15: Now, I totally see that AI is

00:06:19: capable of building software on its

00:06:22: own in a loop, with the Rolf loop that's getting a lot of

00:06:26: hype, uh, related to Claude code. I totally see

00:06:30: that. But the full work of a software

00:06:33: engineer includes the tasks I outlined here

00:06:37: and arguably more than that. It includes building a

00:06:41: good plan, defining which architecture, which

00:06:44: patterns, which technologies to use, reviewing the

00:06:48: code, and also, of course, analyzing the

00:06:52: code, fixing problems with the code, taking responsibility for that

00:06:56: code. That is something I have a very hard time

00:07:00: within the near future because right now, the AI I can

00:07:04: use is too far away from that. It's a

00:07:07: talented, fast writer that makes a lot of mistakes along

00:07:11: the way and that needs very clear guidance.

00:07:14: And that going to models that can do it all on its own, that can

00:07:18: plan out entire architectures cleanly on their

00:07:22: own, that are capable of using the latest

00:07:25: technologies, of writing error-free secure code totally

00:07:29: on their own without review or only with review by other

00:07:33: AIs, that is something I have a very hard time

00:07:37: seeing when I look at today's models and also when I look at the

00:07:41: progress of models over the last years.

00:07:44: Because, yeah, that progress, of course, has been remarkable.

00:07:47: It has been steady and, and good, and especially the tools I shared that

00:07:51: before got way better. So I shared it before

00:07:55: with models. I'm not sure if we're the raw model intelligence,

00:07:59: we're still on a linear, leave alone exponential

00:08:02: trajectory. For the tools, I would say we definitely are on some

00:08:06: linear trajectory. But I have a hard

00:08:10: time seeing that being enough to get us to that full

00:08:14: automation anytime soon. Now,

00:08:18: naturally, these CEOs have other interests

00:08:22: than, than me and you could say I have the

00:08:26: us developers also because I also sell programming

00:08:29: courses. But let's be totally honest, that is not my role

00:08:33: and nothing I will be able to do. I'm just sharing here what

00:08:37: my experience with AI has been, and I'm very open-minded, I would

00:08:41: say, for AI. I'm using it a lot, as I said.

00:08:43: It does write 90% of my code. It's just far away from

00:08:47: that full automation. But yeah, please let me know your

00:08:51: thoughts on this as well, your experiences with AI,

00:08:55: working because you might not be using AI at all or only for

00:08:59: small isolated tasks and you might still write most of the code

00:09:03: or you're using it for everything, you're not even looking on the c- uh,

00:09:07: code, and you're having great results.

00:09:09: I'm interested to learn more about that, so please let me know.

00:09:12: And yeah, have a great time