WEBVTT

NOTE Sentence-level transcript of https://www.youtube.com/watch?v=9fubhllmsBU

NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/9fubhllmsBU.html. A cue ends where the next begins, or 2 s after its last word.

s1
00:00:01.309 --> 00:00:03.309
[music]

s2
00:00:12.400 --> 00:00:19.320
Please welcome to the stage member of technical staff at Anthropic, Tarik Shaupar.

s3
00:00:20.305 --> 00:00:22.305
[music]

s4
00:00:33.080 --> 00:00:34.520
Hey everyone, I'm Tarik.

s5
00:00:34.520 --> 00:00:37.120
I work at Anthropic on Claude Code.

s6
00:00:37.120 --> 00:00:41.880
Before we get started, we have a tradition on Claude Code where we take a selfie before a talk.

s7
00:00:41.880 --> 00:00:48.120
So, if you don't mind, if you strike a pose with me, I'll take a quick selfie at AI engineer.

s8
00:00:50.360 --> 00:00:51.720
Okay, incredible.

s9
00:00:51.720 --> 00:00:57.640
Well, uh yeah, to kick things off, like we said, Fable is back.

s10
00:00:57.640 --> 00:01:03.280
Um we're rolling it out later today.

s11
00:01:03.280 --> 00:01:05.680
Uh keep stay tuned for exact timeline.

s12
00:01:05.680 --> 00:01:10.240
Me and Kat Woo and Simon Wilson will be doing a fireside chat at 12:30.

s13
00:01:10.240 --> 00:01:12.880
We might have some updates for you then.

s14
00:01:12.880 --> 00:01:18.760
Um but Fable is a model I'm just so so excited about.

s15
00:01:18.760 --> 00:01:31.120
It's one of those Anthropic models where you just like you're just going to remember it, like Sonnet 3.5 new, Opus 4, Opus 4.5. It's a model that I just have a lot of like affection and excitement for.

s16
00:01:31.120 --> 00:01:41.840
And the best way to describe Fable to me is like the the map is opening up, you know, like you were playing like an RPG and you've been on the tutorial.

s17
00:01:41.840 --> 00:01:46.520
And now you get to the point where the like, you know, the open world starts, right?

s18
00:01:46.520 --> 00:01:50.200
And there's so much that you can do and explore.

s19
00:01:50.200 --> 00:01:54.640
Uh but there's also it's also a little bit intimidating and confusing, right?

s20
00:01:54.640 --> 00:01:56.600
Cuz there's so much you can do.

s21
00:01:56.600 --> 00:02:03.600
And so, what I wanted to do in this talk is give you guys a field guide to Fable, right?

s22
00:02:03.600 --> 00:02:07.760
How do you work with this new class of models?

s23
00:02:07.760 --> 00:02:10.200
So, I've got four parts to it.

s24
00:02:10.200 --> 00:02:14.520
I've been working on this as a series of articles and blog post.

s25
00:02:14.520 --> 00:02:23.400
Uh but, you know, when we announced Fable is coming out, I was like, "Okay, let me do uh all of this at once at the talk, uh you know,

s26
00:02:23.400 --> 00:02:24.200
uh speed run."

s27
00:02:24.200 --> 00:02:32.720
So, the four parts: unhobbling Claude, finding your unknowns, dealing with the grief, and being unreasonable.

s28
00:02:32.720 --> 00:02:37.240
So, first, unhobbling Claude.

s29
00:02:37.680 --> 00:02:44.560
Um I think something we say really often is that the models are grown, not designed, right?

s30
00:02:44.560 --> 00:02:49.880
We don't wake up and be like, "We need 99% on SWE-bench," right?

s31
00:02:49.880 --> 00:02:52.760
Like, the models are, you know, something we we grow carefully.

s32
00:02:52.760 --> 00:02:55.600
We give it data and feedback and compute.

s33
00:02:55.600 --> 00:03:06.000
Um but, ultimately, it's, you know, something that we it's a little bit organic, and we sort of figure out and learn with the model as we use it.

s34
00:03:06.000 --> 00:03:11.000
And so, um that what that also means is that what contains them is us, right?

s35
00:03:11.000 --> 00:03:19.120
The harness we put them in and the way we prompt them is basically like a function of our understanding of Claude, right?

s36
00:03:19.120 --> 00:03:25.680
And by unhobbling it, I mean, how can we understand Claude better to unleash it?

s37
00:03:25.680 --> 00:03:28.360
And we need to understand Fable more.

s38
00:03:28.360 --> 00:03:39.000
So, I think one of my points is that, you know, uh we're still so early, and I think there's a lot more it understanding in Fable uh to unlock.

s39
00:03:39.000 --> 00:03:46.160
And uh I I think I'll give you a quick example about how models get smarter, cuz it's a little bit unintuitive, right?

s40
00:03:46.160 --> 00:03:54.080
Like, there I saw this viral tweet a couple weeks ago being like, you know, "Why can't LLMs say which Pokémon end in AW?

s41
00:03:54.080 --> 00:03:56.280
There are a thousand Pokémon, right?

s42
00:03:56.280 --> 00:04:01.680
And turns out there are two who whose names end in AW, Croconaw and Drednaw, right?

s43
00:04:01.680 --> 00:04:04.960
And it turns out if you ask like a normal chat model, it can't answer it.

s44
00:04:04.960 --> 00:04:10.240
Which is kind of confusing because like, you know, it definitely knows all the names of the Pokémon, right?

s45
00:04:10.240 --> 00:04:13.520
But if you ask Claude Code, it can, right?

s46
00:04:13.520 --> 00:04:20.840
Cuz what it does is that it fetches every Pokémon and writes a script to filter for AW, right?

s47
00:04:20.840 --> 00:04:26.160
And so, this is what I mean by like unhobbling Claude.

s48
00:04:26.160 --> 00:04:28.680
We call this capability overhang, right?

s49
00:04:28.680 --> 00:04:31.360
Claude gets smarter in spiky ways.

s50
00:04:31.360 --> 00:04:41.360
So, it doesn't just remember every Pokémon and reason through it, but if you give it the code execution tool, it can find the two Pokémon that end with AW, right?

s51
00:04:41.360 --> 00:04:45.919
And so, this is I think part of the challenge with Fable is figuring out this capability overhang.

s52
00:04:45.919 --> 00:04:47.280
What is now possible?

s53
00:04:47.280 --> 00:04:50.640
And I think this is like a discovery that I'm excited to go on with you.

s54
00:04:50.640 --> 00:04:57.520
Uh to make this a little bit clearer, I'm going to talk about a few different examples of how models have progressed in the past.

s55
00:04:57.520 --> 00:05:04.360
Um one of the big examples obviously is like chat, you know, the chat models were had to be given context, right?

s56
00:05:04.360 --> 00:05:16.800
Like maybe you paste in your code base and maybe naively you might have thought like, you know, the way we solve coding is by the context just gets really large and I can just paste in my entire code base, you know, it'll be a 100 million context window.

s57
00:05:16.800 --> 00:05:25.480
But it turns out that instead, if you give it arms, like you give it the bash tool and ways to work with the environment, it can build and search its own context.

s58
00:05:25.480 --> 00:05:29.000
And that's sort of like the insight that led to Claude Code, right?

s59
00:05:29.000 --> 00:05:34.080
And so, again, spiky, like a new like innovation kind of, right?

s60
00:05:34.080 --> 00:05:37.000
In how we think about and work with the model.

s61
00:05:37.000 --> 00:05:39.760
And then recently we we rolled out Claude Tag.

s62
00:05:39.760 --> 00:05:45.520
Uh and what sort of unlocked Claude Tag is its ability to work proactively in multiplayer.

s63
00:05:45.520 --> 00:05:49.880
Uh Claude Code, you know, is something that you have to prompt for it to do work, right?

s64
00:05:49.880 --> 00:05:58.840
And uh, this ability for Claude to wake itself up and do work is something that we think is unlocking the new wave of agents.

s65
00:05:58.840 --> 00:05:59.880
But there's there's more here.

s66
00:05:59.880 --> 00:06:06.120
So, for example, uh, we recently removed 80% of the system prompt for Claude code, right?

s67
00:06:06.120 --> 00:06:12.000
And this is one of the ways in which models, you know, and what they need uh, changes over time.

s68
00:06:12.000 --> 00:06:22.120
So, originally, like, you know, maybe back in 3 sign of 3.5 new, the best practices for a system prompt was a small system prompt, few tools, and lots of examples,

s69
00:06:22.120 --> 00:06:22.760
right?

s70
00:06:22.760 --> 00:06:27.720
And then as the models get smarter, you can give it more information and more instructions, and they start following them.

s71
00:06:27.720 --> 00:06:32.800
And so, it's a larger system prompt with lots of examples and many tools, right?

s72
00:06:32.800 --> 00:06:40.160
But most recently, we found this new class of models want fewer what want a smaller system prompt.

s73
00:06:40.160 --> 00:06:45.640
The examples tend to constrain it cuz it's actually more imaginative than the examples we give it.

s74
00:06:45.640 --> 00:06:50.400
And so, uh, and we tried to give it context and not just constraints.

s75
00:06:50.400 --> 00:06:53.440
We're really trying to avoid being like, "Do not do this."

s76
00:06:53.440 --> 00:06:56.080
Um, which was really necessary for the previous models.

s77
00:06:56.080 --> 00:07:02.720
Um, and so, this is like, uh, a way that the system prompt is changing and and probably will continue to change.

s78
00:07:02.720 --> 00:07:06.640
Uh, another feature I really like is the ask you a question tool.

s79
00:07:06.640 --> 00:07:09.440
This is something I worked on when I first got to Claude code.

s80
00:07:09.440 --> 00:07:17.400
And and it's uh, when Claude, you know, is is planning or wants to ask you a question, it can show you a multiple choice dialogue.

s81
00:07:17.400 --> 00:07:20.080
Uh, for Opus 4, it could barely call it.

s82
00:07:20.080 --> 00:07:26.760
I had to like really tweak the tool to make sure that it was uh, that it would work, right?

s83
00:07:26.760 --> 00:07:33.360
And then, sometime at Opus 4.5, I was like, well, what if I asked it to like, you know, ask me 40 questions about the spec.

s84
00:07:33.360 --> 00:07:34.760
It could start interviewing me, right?

s85
00:07:34.760 --> 00:07:38.600
And so, its ability to ask questions jumped, right?

s86
00:07:38.600 --> 00:07:47.240
And then, most recently with Opus 4.8 and Fable, I can now build a whole HTML report with the questions embedded inside of them.

s87
00:07:47.240 --> 00:07:52.400
And uh it's just like a whole new way of interacting with uh with Claude, right?

s88
00:07:52.400 --> 00:07:58.000
And and so this progression of like how Claude can get information from you has also changed.

s89
00:07:58.000 --> 00:08:03.360
Um speaking of which, uh Markdown and HTML is something I've also talked a lot about.

s90
00:08:03.360 --> 00:08:07.880
Um you know, it turned Initially, Markdown was a a good output for the model.

s91
00:08:07.880 --> 00:08:11.560
Um you know, it could show a little bit of rich information.

s92
00:08:11.560 --> 00:08:14.400
And then, you know, with plan mode, it started to be for you.

s93
00:08:14.400 --> 00:08:17.400
Like you could understand what Claude was about to do.

s94
00:08:17.400 --> 00:08:22.520
Um and now, you know, Claude can build you these in-depth HTML reports, right?

s95
00:08:22.520 --> 00:08:27.880
So again, a way of this the model is getting smarter in a spiky way.

s96
00:08:27.880 --> 00:08:33.320
I really like to emphasize that this is closer to a biology than a physics, right?

s97
00:08:33.320 --> 00:08:36.560
It's still very empirical, very organic.

s98
00:08:36.560 --> 00:08:40.520
Um we don't know all the rules, but there is some sort of science behind it, right?

s99
00:08:40.520 --> 00:08:42.960
Like there is an intuition to build as well.

s100
00:08:42.960 --> 00:08:46.600
And so I really, you know, encourage you to treat Fable like that.

s101
00:08:46.600 --> 00:08:52.560
Uh one of my favorite papers uh that at Anthropic that we've written is on the biology of a large language model.

s102
00:08:52.560 --> 00:08:59.240
Um all of our research papers are meant to be read by, you know, people with various degrees of technical expertise, but this is one of my favorite.

s103
00:08:59.240 --> 00:09:04.440
So uh if you're looking to learn a little bit more, suggest you check it out.

s104
00:09:04.440 --> 00:09:08.839
But so uh yeah, we talked about unhobbling Claude.

s105
00:09:08.839 --> 00:09:14.160
But it turns out when you're working with Fable, you also need to unhobble yourself, right?

s106
00:09:14.160 --> 00:09:19.720
And so one of the things that I think a lot about is that the map is not the territory, right?

s107
00:09:19.720 --> 00:09:27.040
When I'm working on a coding problem, the plan and prompt and spec that I have in my mind is the map, right?

s108
00:09:27.040 --> 00:09:34.160
But the territory is the actual code base, the real world, the constraints that Claude needs to navigate, right?

s109
00:09:34.160 --> 00:09:39.400
And whenever Claude runs into something in the territory that's not in the map, I call that an unknown, right?

s110
00:09:39.400 --> 00:09:41.680
Claude has to figure out what to do about it.

s111
00:09:41.680 --> 00:09:44.920
It's a decision point that I haven't specified.

s112
00:09:44.920 --> 00:09:57.400
And Fable is one of the first models where I felt that like I really have to figure out my unknowns because if not, it's going to traverse such a large area that like it's going to run into a lot of them.

s113
00:09:57.400 --> 00:10:00.680
So, how do you figure out your unknowns?

s114
00:10:00.680 --> 00:10:09.520
Um Yeah, it I Fable's bottleneck my ability by my ability to match the map and the territory to find my unknowns.

s115
00:10:09.520 --> 00:10:13.840
So, a few um few ways to think about this.

s116
00:10:13.840 --> 00:10:15.840
I like to think of it in a a matrix.

s117
00:10:15.840 --> 00:10:19.640
So, like for any problem, I have a bunch of known knowns.

s118
00:10:19.640 --> 00:10:21.760
This is usually like what I write in my prompt.

s119
00:10:21.760 --> 00:10:23.000
What do I want, right?

s120
00:10:23.000 --> 00:10:24.600
Then I have known unknowns.

s121
00:10:24.600 --> 00:10:29.000
Things that like I know I haven't don't really know yet, but I just haven't figured out yet.

s122
00:10:29.000 --> 00:10:31.560
I can um Uh yeah.

s123
00:10:31.560 --> 00:10:32.600
Then I've got unknown knowns.

s124
00:10:32.600 --> 00:10:38.080
Like what's so obvious that I just wouldn't write it down, you know, but I I'd know it when I see it, right?

s125
00:10:38.080 --> 00:10:40.600
And then finally unknowns unknowns.

s126
00:10:40.600 --> 00:10:41.920
What haven't I considered at all?

s127
00:10:41.920 --> 00:10:43.440
What do I not know, right?

s128
00:10:43.440 --> 00:10:48.520
Like what is something that if I knew could change how I prompt Claude?

s129
00:10:48.520 --> 00:10:53.440
And and luckily, you can use Claude, you can use Fable to find your unknowns.

s130
00:10:53.440 --> 00:10:57.440
So, I'm going to go over a few examples of how I do that with Fable.

s131
00:10:57.440 --> 00:11:02.400
Um The first is I like to do what I call a blind spot pass.

s132
00:11:02.400 --> 00:11:09.240
So, I like to say something like, "Hey, I'm working on a new auth provider that I know nothing about uh like in this code base.

s133
00:11:09.240 --> 00:11:14.240
Can you do a blind spot pass to help me figure out my relevant unknown unknowns and help me prompt better, right?"

s134
00:11:14.240 --> 00:11:23.680
And so, this like might have Claude go through the the auth module and figure out like, "Oh, you know, this is kind of like a hairy little uh dead end that comes up a lot."

s135
00:11:23.680 --> 00:11:25.560
Maybe search in my Git diff or Slack.

s136
00:11:25.560 --> 00:11:28.200
I might tell it where there's context, right?

s137
00:11:28.200 --> 00:11:31.680
So, that I can learn about, you know, all the gotchas.

s138
00:11:31.680 --> 00:11:33.480
And and you can use this very broadly, right?

s139
00:11:33.480 --> 00:11:35.800
You can use it to teach you about new fields.

s140
00:11:35.800 --> 00:11:39.200
I I recently did this for color grading when doing video editing.

s141
00:11:39.200 --> 00:11:42.640
Um I think this is really powerful and and Sable is incredible at it.

s142
00:11:42.640 --> 00:11:49.040
Um in many ways, the model knows more about you know, almost everything than I do.

s143
00:11:49.040 --> 00:11:50.880
I just need to get it out of it.

s144
00:11:50.880 --> 00:11:55.880
Um then I like to do some brainstorms and prototypes.

s145
00:11:55.880 --> 00:11:58.880
Uh this helps me figure out my unknown knowns, right?

s146
00:11:58.880 --> 00:12:02.760
Things like especially for design, it for me it's like know it when you see it, right?

s147
00:12:02.760 --> 00:12:05.920
So, I might ask it to uh create a dashboard.

s148
00:12:05.920 --> 00:12:08.960
Um and I tell it I have no visual taste.

s149
00:12:08.960 --> 00:12:13.840
Uh make me an HTML page with four widely different design decisions so I can react to them, right?

s150
00:12:13.840 --> 00:12:21.720
And then you know, you tweak this as you want, but like the idea is to sort of get an idea of like what are the things that you uh

s151
00:12:21.720 --> 00:12:23.960
you know, you can't describe in words, right?

s152
00:12:23.960 --> 00:12:28.120
And uh like work with the model to help figure that out.

s153
00:12:28.120 --> 00:12:30.480
Uh then then interviews.

s154
00:12:30.480 --> 00:12:39.120
So, once I have an idea of like, you know, this is what I want to do, uh there's probably still a lot of like uh unknowns here, right?

s155
00:12:39.120 --> 00:12:43.360
Where I might not have considered something, I might not have specified it.

s156
00:12:43.360 --> 00:12:45.800
And so, I'll ask Claude to interview me, right?

s157
00:12:45.800 --> 00:12:47.880
And I'll give it a little bit more context.

s158
00:12:47.880 --> 00:12:55.880
In any of these questions, like giving it a little bit more context about you and the work and the stage you're at, like, "Hey, yeah, prioritize questions that would change the architecture."

s159
00:12:55.880 --> 00:12:58.360
is extremely helpful.

s160
00:12:58.720 --> 00:13:00.440
Uh then references.

s161
00:13:00.440 --> 00:13:05.360
One of the best ways to give Claude a map is to give it another map, right?

s162
00:13:05.360 --> 00:13:13.720
So, instead of me writing out the spec, uh I can just say, "Hey, here's some code that represents what I want to be done, right?

s163
00:13:13.720 --> 00:13:17.320
It could be in a different uh system or language.

s164
00:13:17.320 --> 00:13:22.600
Uh but just read this code, understand it, and then use that to start your work, right?

s165
00:13:22.600 --> 00:13:25.280
And uh again, this can be in a lot of different ways.

s166
00:13:25.280 --> 00:13:29.680
If I'm making a a React component, I might have an HTML mock-up that is my map, right?

s167
00:13:29.680 --> 00:13:31.240
That I pass in as a reference.

s168
00:13:31.240 --> 00:13:32.720
I think this is a really, really powerful.

s169
00:13:32.720 --> 00:13:35.800
Fable is really incredible at it.

s170
00:13:35.800 --> 00:13:39.360
Uh something else I've like really appreciated is implementation notes.

s171
00:13:39.360 --> 00:13:46.640
So, if uh while you're running Fable uh and it runs into an unknown, ask it to log it, right?

s172
00:13:46.640 --> 00:13:54.560
So that um you uh you can see where the deviations happened and then you can sort of figure out why as well, you know?

s173
00:13:54.560 --> 00:13:57.920
It will usually give you some context about what happened.

s174
00:13:57.920 --> 00:14:08.600
And then finally, I'd like to get a Fable to quiz me about what happened uh just to make sure I understand what I'm doing and I can represent this work, you know, when I'm creating a PR or merging it.

s175
00:14:08.600 --> 00:14:14.520
Um this is a really great way of like making sure that you're like really in the loop with Fable.

s176
00:14:14.520 --> 00:14:23.560
And I think that's like one of the most important parts of Fable is like staying in the loop and making sure that you uh you get what you want.

s177
00:14:23.560 --> 00:14:28.120
So, um those are some of my tips for working with Fable.

s178
00:14:28.120 --> 00:14:40.360
Uh I also want to say that the first time I used a Mithril class model, uh used Fable, I felt both a huge sense of like gain, but also a sense of loss.

s179
00:14:40.360 --> 00:14:44.200
And I I wanted to talk a little bit about that, you know?

s180
00:14:44.200 --> 00:14:49.640
Um when I think about coding before LLMs, it feels like a foreign country, you know?

s181
00:14:49.640 --> 00:14:57.920
Like I used to run a YC startup about 30 people and we were just constantly forced into trade-offs because of how hard code was, right?

s182
00:14:57.920 --> 00:15:04.080
Like we could make the the app fast or we could try prototyping new feature and and this might take a month or this would take 2 months.

s183
00:15:04.080 --> 00:15:05.280
And so we had to choose.

s184
00:15:05.280 --> 00:15:07.480
It was just really, really hard.

s185
00:15:07.480 --> 00:15:15.240
Um and now I went back to that code base a couple weeks ago and I thought about some of the things that I wanted to do.

s186
00:15:15.240 --> 00:15:17.760
And uh it was just way easier.

s187
00:15:17.760 --> 00:15:23.040
It was like the things that would have taken me weeks, I could do in hours, you know?

s188
00:15:23.040 --> 00:15:25.920
And uh at some point it's like, yeah, like how can you not laugh?

s189
00:15:25.920 --> 00:15:28.560
Also, how can you not cry, all honestly?

s190
00:15:28.560 --> 00:15:34.400
Like, it's like one of these things where I really, really loved programming and writing code by hand.

s191
00:15:34.400 --> 00:15:39.720
I love the feeling of like seeing the code base in my mind and like rotating it.

s192
00:15:39.720 --> 00:15:48.080
But I also remember just, you know, like staying up late nights trying to debug, working on things for weeks without working, right?

s193
00:15:48.080 --> 00:15:50.880
I just remember swimming in failure.

s194
00:15:50.880 --> 00:15:54.240
I just remember that like the most of the projects I've ever worked on have failed.

s195
00:15:54.240 --> 00:15:56.800
Most startups go bankrupt, you know?

s196
00:15:56.800 --> 00:16:00.560
I think just overall programming and coding is extremely hard.

s197
00:16:00.560 --> 00:16:07.240
And like as much as I enjoyed those highs, I I can cannot go back, right?

s198
00:16:07.240 --> 00:16:11.880
And the way my reflection here is like the only way out is through, right?

s199
00:16:11.880 --> 00:16:14.240
There's still a lot to learn with the gentle coding.

s200
00:16:14.240 --> 00:16:16.360
There's a lot to learn with Fable.

s201
00:16:16.360 --> 00:16:22.560
But I think if we try really hard and if we like stay in the loop, we on hobble it, we can get there, you know?

s202
00:16:22.560 --> 00:16:28.360
And we can come out on the other side with just so much more.

s203
00:16:28.360 --> 00:16:33.480
And so, the last bit I wanted to talk about is is the so much more part, right?

s204
00:16:33.480 --> 00:16:36.760
I call this being unreasonable.

s205
00:16:36.760 --> 00:16:44.320
One of my favorite parts of Anthropic is that we believe that tradeoffs are not real.

s206
00:16:44.320 --> 00:16:50.640
Um, like I think that very often, I like in my previous company, I was very used to being reasonable.

s207
00:16:50.640 --> 00:16:56.839
So I'd like write down this list of priorities and I'd be like, well, I guess we can prioritize this against this, right?

s208
00:16:56.839 --> 00:16:58.960
And like, you know, that makes sense.

s209
00:16:58.960 --> 00:17:01.960
So we'll this will be our priority this quarter.

s210
00:17:01.960 --> 00:17:04.600
But what if you just did all of it, you know?

s211
00:17:04.600 --> 00:17:09.120
What if you forced reality to show you the tradeoff, right?

s212
00:17:09.120 --> 00:17:11.640
Um, this is one thing I really valued out of our culture in Anthropic.

s213
00:17:11.640 --> 00:17:16.319
And that my reflection going forward is that I'm going to be a lot less reasonable.

s214
00:17:16.319 --> 00:17:23.640
Um, I think one of this like the math of Claude and Fable really changes how you think about tradeoffs.

s215
00:17:23.640 --> 00:17:26.760
And there are so many tradeoffs that you make implicitly in your head, right?

s216
00:17:26.760 --> 00:17:28.400
Like good, fast, cheap.

s217
00:17:28.400 --> 00:17:30.560
Now it's pick three, right?

s218
00:17:30.560 --> 00:17:40.960
Um I think that like the best way to like do more ambitious work is to uh like reframe and make big make ourselves more ambitious.

s219
00:17:40.960 --> 00:17:48.960
Because I think the only way to prove that agents work is to do the best work of our lives faster than ever before.

s220
00:17:48.960 --> 00:17:54.880
Um you know, for example, I made this deck last night in about 4 hours with Fable.

s221
00:17:54.880 --> 00:18:01.520
I feel like it's a it's a deck I really like and I I really enjoyed it, but I also um you know, I did it really fast.

s222
00:18:01.520 --> 00:18:08.560
Uh and I think that if you're here, you know, at AI engineer, the world is kind of looking at you to prove that AI works, right?

s223
00:18:08.560 --> 00:18:14.120
That it's not just like a fad or something, but that it can make us more productive and also save us time.

s224
00:18:14.120 --> 00:18:22.600
And and that's my resolution for this year is to work be more productive, but work less and spend more time with people I really care about.

s225
00:18:22.600 --> 00:18:29.880
Uh I think it's also worth calling out that building is easier, but generating value is still hard.

s226
00:18:29.880 --> 00:18:38.560
And I think this is something that we run into, you know, as AI engineers sometimes where we think so much about the process of building and our our setups.

s227
00:18:38.560 --> 00:18:42.040
Um but the the point is to generate value, right?

s228
00:18:42.040 --> 00:18:49.640
And uh there it takes a lot of swings, it takes a lot of tries to find the valuable stuff.

s229
00:18:49.640 --> 00:18:58.680
Uh but that really is the goal and that's like, you know, again, what the world is looking to us to prove that AI can really transform it.

s230
00:18:58.680 --> 00:19:06.840
So, to to end, I just want to say like go explore, make it real, and uh yeah, be less reasonable.

s231
00:19:06.840 --> 00:19:09.040
Thank you.

s232
00:19:25.913 --> 00:19:27.440
[music]

s233
00:19:27.440 --> 00:19:29.440
Woo!