WEBVTT

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

NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/bdHaOXZOhcM.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
00:00:13.280 --> 00:00:16.400
Hi everyone, thanks for coming.

s3
00:00:16.400 --> 00:00:17.680
Uh, I'm Hersh Agarwal.

s4
00:00:17.680 --> 00:00:21.840
I'm the CTO and co-founder of the browser company, makers of the Arc and Dia browsers.

s5
00:00:21.840 --> 00:00:28.400
Uh, and I'm going to talk about prototyping as leadership as you get to a manager schedule, how you can keep building.

s6
00:00:28.400 --> 00:00:32.000
So this is my actual calendar from last week.

s7
00:00:32.000 --> 00:00:35.520
Uh I imagine this is kind familiar to some of you in leadership.

s8
00:00:35.520 --> 00:00:38.000
Um that was my actual week.

s9
00:00:38.000 --> 00:00:40.480
I uh have a whole org that reports up into me.

s10
00:00:40.480 --> 00:00:48.399
So I have 15 plus recurring meetings a week uh standups, reviews, recruiting meetings, etc. Uh and seven direct reports.

s11
00:00:48.399 --> 00:00:55.280
And I've noticed something over the last several months, which is I've started to consistently ship, you know, two to 10 PRs a week.

s12
00:00:55.280 --> 00:00:56.320
Uh and this is new.

s13
00:00:56.320 --> 00:00:58.480
This was not possible several months ago.

s14
00:00:58.480 --> 00:01:00.719
It's really because of these new AI agents.

s15
00:01:00.719 --> 00:01:04.720
Uh, and notably, I also have a toddler at home, so I like cannot work 996.

s16
00:01:04.720 --> 00:01:06.720
You know, I'm working 40 50 hours a week.

s17
00:01:06.720 --> 00:01:10.960
So, I really have to fit all this in into a regular week.

s18
00:01:11.280 --> 00:01:17.840
And pre-I agents, uh, as you grew as a leader, you had more of the org reporting to you.

s19
00:01:17.840 --> 00:01:22.880
Uh, you would sort of the way you would influence the org was through communicating to people.

s20
00:01:22.880 --> 00:01:34.079
Uh so you'd uh write road maps, docs, meetings, and you'd sort of incept uh your context and what you wanted to build with your employees uh and your engineers.

s21
00:01:34.079 --> 00:01:44.320
And now, interestingly, in the last few months, as coding agents have become more autonomous and able to handle longer tasks, the manager schedule, as Paul Graham put it, is suddenly

s22
00:01:44.320 --> 00:01:46.399
usable as building time.

s23
00:01:46.399 --> 00:01:48.880
You can actually ship stuff.

s24
00:01:48.880 --> 00:01:52.000
And so I think building is part of the job now.

s25
00:01:52.000 --> 00:02:00.240
you can do it first of all as a leader but it's I think it's also becoming necessary as part of the job for two reasons.

s26
00:02:00.240 --> 00:02:11.760
First, uh, the technology world is changing where suddenly the core technology that's part of our products are these frontier models that change every 3 months, which is a new dynamic that's come out.

s27
00:02:11.760 --> 00:02:15.599
And as each new frontier model comes out, its capabilities change.

s28
00:02:15.599 --> 00:02:17.840
The contours of what it's useful for change.

s29
00:02:17.840 --> 00:02:21.360
It's, you know, how it actually reacts to prompting changes.

s30
00:02:21.360 --> 00:02:30.000
And there's so much noise on Twitter, even internally for us, and so many opinions with each new model release on what's good and what's bad.

s31
00:02:30.000 --> 00:02:37.440
And I found it is impossible to tell what a new model is good for unless you have your hands in it and you're using it all day long.

s32
00:02:37.440 --> 00:02:48.720
And so having an ability to slot in building time into your schedule means you can figure out the contours of what is this new model family capable of and both how do I direct my engineers

s33
00:02:48.720 --> 00:02:55.519
in terms of uh setting expectations on how they should be building but also getting a sense of how does this fit into our product?

s34
00:02:55.519 --> 00:02:56.560
How does this fit into our business?

s35
00:02:56.560 --> 00:02:59.599
You know how is this how is our strategy going to change because of this?

s36
00:02:59.599 --> 00:03:02.879
What's going to happen in three to six months when the open source models catch up?

s37
00:03:02.879 --> 00:03:07.440
All of that intuition comes from actually using the models and building.

s38
00:03:07.440 --> 00:03:14.879
I've also found even if you've built the intuition, it's actually tough to communicate that to other people who haven't played with the models.

s39
00:03:14.879 --> 00:03:17.120
Uh, and so you're like, "Oh, this new thing is possible.

s40
00:03:17.120 --> 00:03:18.000
It's going to be amazing."

s41
00:03:18.000 --> 00:03:20.000
And your engineers are like, "Okay, yeah, sure."

s42
00:03:20.000 --> 00:03:28.319
Um, so it's really helpful to have some time to actually build stuff so you can show them, you know, you can be like, "Hey, I built this prototype with this new model family.

s43
00:03:28.319 --> 00:03:29.599
It works in the product.

s44
00:03:29.599 --> 00:03:31.519
Here's an actual prototype you can play with."

s45
00:03:31.519 --> 00:03:39.120
And this is just so much faster and more efficient than trying to convince people every 3 months when a new model family comes out.

s46
00:03:39.200 --> 00:03:41.519
I also think leaders are really well suited for it.

s47
00:03:41.519 --> 00:03:54.000
You know, you all have uh context, more context than anybody else uh in the organization about the business, the strategy, the trade-offs, the decisions to be made, what needs to be done, what's uh what's coming up, what are imperatives.

s48
00:03:54.000 --> 00:04:02.879
And so the steering of a leader uh as you are prompting these models is to per token uh more impactful than an IC's.

s49
00:04:02.879 --> 00:04:06.400
And so you can really fit this into a little bit of time.

s50
00:04:06.400 --> 00:04:10.239
The delegation skill also transfers like you delegate to people.

s51
00:04:10.239 --> 00:04:14.480
That's sort of one of the core skills of being a leader and that transfers to agents pretty well.

s52
00:04:14.480 --> 00:04:18.079
You know it's setting goals, giving context, checking in, even coaching the agent.

s53
00:04:18.079 --> 00:04:19.919
You know what if you tried this?

s54
00:04:19.919 --> 00:04:24.720
Um, and then I've also found I I think this is maybe specific to now.

s55
00:04:24.720 --> 00:04:32.160
Uh, maybe this will change in several months as these models get better, but the models are really good at execution, but still not unbelievable at judgment.

s56
00:04:32.160 --> 00:04:37.600
You know, often the model will come back and be like, hey, that algorithm idea is not possible or like I can't do this.

s57
00:04:37.600 --> 00:04:40.479
And so you have as a leader have to be like, oh, have you tried this thing?

s58
00:04:40.479 --> 00:04:42.000
And the model's like, oh, okay, cool, cool, cool.

s59
00:04:42.000 --> 00:04:42.639
I'll try that.

s60
00:04:42.639 --> 00:04:48.240
And so it's a really nice dynamic you can have with these models if you have time slotted out for building.

s61
00:04:48.240 --> 00:04:52.240
uh that really leverages all the context you have.

s62
00:04:52.320 --> 00:04:56.000
So the great Julie Zo had a great tweet about this uh a month ago.

s63
00:04:56.000 --> 00:05:00.160
She pulled some Bay Area technical leaders on what can you actually build?

s64
00:05:00.160 --> 00:05:02.240
What should you be building in this new world?

s65
00:05:02.240 --> 00:05:03.840
And four categories emerged.

s66
00:05:03.840 --> 00:05:05.360
Uh you could build internal tools.

s67
00:05:05.360 --> 00:05:10.560
So internal efficiency uh quality of life improvements just like gardening around the codebase and the product.

s68
00:05:10.560 --> 00:05:11.759
This is really helpful.

s69
00:05:11.759 --> 00:05:12.880
Um I really like this.

s70
00:05:12.880 --> 00:05:14.720
I learned from this the celebration story.

s71
00:05:14.720 --> 00:05:17.520
You can build artifacts to celebrate people on your team.

s72
00:05:17.520 --> 00:05:31.759
And then arguably, I think the most important is the vision piece is really playing with the new model families, understanding what's possible viscerally yourself, and then trying to fit that into the business and building products that can show that off to your engineers

s73
00:05:31.759 --> 00:05:36.880
so you can really push uh the boundaries of what's possible with the product and business.

s74
00:05:36.880 --> 00:05:37.840
I she's right.

s75
00:05:37.840 --> 00:05:39.520
I would not take any critical path work.

s76
00:05:39.520 --> 00:05:50.400
the last thing you want to do is to have that dependent on you, especially you're going to be dragged into fires and recruiting calls and meetings, etc. Uh so really trying to do one of these four categories is is optimal.

s77
00:05:50.400 --> 00:05:53.039
So this is this is my how I work.

s78
00:05:53.039 --> 00:05:57.120
Uh and what two to three hours a day of coding can actually look like.

s79
00:05:57.120 --> 00:06:06.080
Uh so I usually have a morning coding block about an hour and this is when I look over last night's code, what the uh agent did last night, review it.

s80
00:06:06.080 --> 00:06:07.680
I'll talk a little bit more about this.

s81
00:06:07.680 --> 00:06:14.960
Uh, and then a few maybe steering blocks throughout the day interspersed between one-on- ones and reviews and meetings and everything else you need to do.

s82
00:06:14.960 --> 00:06:22.479
And then the most important block is that 5:00 p.m. block at the end of the day where you really set up whatever overnight run it is you want to run.

s83
00:06:22.479 --> 00:06:28.160
And I'll talk about that whether that's coding or training models or whatever you want the agent to do.

s84
00:06:28.400 --> 00:06:31.280
I'll go over sort of three examples of tasks you can do.

s85
00:06:31.280 --> 00:06:34.880
But really what I found work for me is this one overnight loop.

s86
00:06:34.880 --> 00:06:46.400
uh you first in that 5pm block gather context I'll talk about that uh you set up the run you know you get ask answer any clarifying questions uh and then claude code or whatever your coding agent is overnight

s87
00:06:46.400 --> 00:06:55.440
does the thing you know does the work for four six eight hours and in the morning you get this beautiful report uh and you figure out what to do with it and then you ship whatever it is so

s88
00:06:55.440 --> 00:07:06.400
I'll talk about building features which is sort of the most obvious one uh I found this worked really well with evals and hill climbing so optimizing AI features uh and then a new thing actually in the maybe more recent

s89
00:07:06.400 --> 00:07:12.960
uh generations of models is you can have these models train other ML models overnight too and this works really really well.

s90
00:07:12.960 --> 00:07:30.479
So building features um the big mindset shift for me that really helped was starting to think about building features not in terms of how do I build this feature and break it up and then give it prompting on how to do the individual task but rather

s91
00:07:30.479 --> 00:07:36.639
what is all the context this frontier model needs to be able to make decisions like I would make.

s92
00:07:36.639 --> 00:07:42.319
So trying to give it as much context as possible because if it's working for 6 hours, 8 hours overnight, you're not going to be there to steer it.

s93
00:07:42.319 --> 00:07:47.440
And so you want it to have as much context as you do about the business goals of whatever you're trying to get it to do.

s94
00:07:47.440 --> 00:07:57.199
So, uh, a little tip I found that has worked well for me, um, if you have a co-work agent, whatever your, uh, cloud co-work or codeex or I recommend DIA, I feel like that's the best one.

s95
00:07:57.199 --> 00:07:58.400
Not biased at all.

s96
00:07:58.400 --> 00:08:11.520
Um, whatever it is that's connected to your Slack and your Jira, Confluence, notion, the repo, etc. Uh, just I before a meeting or something, I'll be like at like 3:00 4 p.m. I'll be like, "Hey, I really want to build this.

s97
00:08:11.520 --> 00:08:27.280
go and do like 20 minutes of research and go dig through all of Slack and Notion and everything else and come up with a cloud code prompt that I can post paste into a cloud code and just give me as much context as possible the trade-offs what we tried before what works what doesn't why are we're doing

s98
00:08:27.280 --> 00:08:36.959
it what is like the business context of this feature I want to build or this model I want to train and that takes like 30 seconds to whisper flow into DIA or whatever your co-work agent is

s99
00:08:36.959 --> 00:08:44.080
and the nice thing is this comes back with a giant prompt that then you can paste into cloud code or cursor or codeex or whatever your coding agent is.

s100
00:08:44.080 --> 00:08:49.519
So this is an example of a 5mm prompt I'll have that I give it before I go to bed.

s101
00:08:49.519 --> 00:08:54.480
I'll say implement this whole feature and then just paste all of that context I got from my co-work agent.

s102
00:08:54.480 --> 00:08:57.040
Uh and then it's important to think about verification.

s103
00:08:57.040 --> 00:09:04.000
So as this uh agent is running overnight, you want it to verify how it's doing as it's building and testing the feature.

s104
00:09:04.000 --> 00:09:07.360
So first I'll be like hey write the tests first so they capture what you do.

s105
00:09:07.360 --> 00:09:12.240
This is really helpful because I I found of decoding agents if you if they write tests afterwards, they're a little sloppish.

s106
00:09:12.240 --> 00:09:16.240
Um, and then I'll tell it to test the endto-end flow with computer use.

s107
00:09:16.240 --> 00:09:21.519
You know, go around, click around, just make sure the flow works and matches the business context I gave you.

s108
00:09:21.519 --> 00:09:28.800
And so the modern models are really good at reasoning about, okay, does this flow actually solve the problem I was trying to solve?

s109
00:09:28.800 --> 00:09:35.440
And then uh I usually prompted to do as much work as possible so that when I wake up in the morning, the stack is ready.

s110
00:09:35.440 --> 00:09:47.680
So split this into reviewer friendly PRs with clear descriptions, manage CI and get CI green and just make sure you're you're monitoring uh just to make sure the the whole stack is passable.

s111
00:09:47.680 --> 00:09:49.279
AI code review is really important.

s112
00:09:49.279 --> 00:09:56.160
We have a bunch of internal AI code reviewers uh you know pre the previous talk was talking about codto and other AI code reviewer.

s113
00:09:56.160 --> 00:10:05.760
If you don't have any, I would look on Twitter for a bunch of AI code review skills and then just prompt it, hey, once you're done, run this AI code review skill in a clean sub agent, and then fix those things.

s114
00:10:05.760 --> 00:10:14.560
And then watch the PRs, fix every bot comment, every, you know, CI check that shows up, any sort of uh anything you need to do, resolve the threads, run this autonomously.

s115
00:10:14.560 --> 00:10:20.000
Uh, don't ask me questions, and then I'll like throw in a little encouraging something, you know, be like, you'll do great.

s116
00:10:20.000 --> 00:10:20.640
I believe in you.

s117
00:10:20.640 --> 00:10:21.360
It's going to be great.

s118
00:10:21.360 --> 00:10:25.200
Um, I don't know if that helps or not, but it's it's a I tend to do that.

s119
00:10:25.200 --> 00:10:26.560
Uh, and then I tell them I'm going to bed.

s120
00:10:26.560 --> 00:10:28.320
Just leave me a report in the morning on what you do.

s121
00:10:28.320 --> 00:10:31.839
I need the full stack ready and a report on what trade-offs you made, how you did.

s122
00:10:31.839 --> 00:10:33.519
Uh, and it does great.

s123
00:10:33.519 --> 00:10:37.440
Uh, it actually modern models, new Opus 4.8 or the new GBT.

s124
00:10:37.440 --> 00:10:44.240
They can handle what used to be, you know, weeks of work uh, in one overnight run and you come back in the morning with this uh, beautiful package ready for you.

s125
00:10:44.240 --> 00:10:49.040
It actually makes the mornings kind of nice because you have this little present ready for you.

s126
00:10:49.120 --> 00:10:51.680
Another example is optimizing AI features.

s127
00:10:51.680 --> 00:10:54.560
So, first you're at the AI engineering conference.

s128
00:10:54.560 --> 00:10:56.320
I assume you all are familiar with this.

s129
00:10:56.320 --> 00:11:01.920
When you first build an LLM call or agent into a into a feature, it's not optimized.

s130
00:11:01.920 --> 00:11:08.079
You know, you need to optimize it hill climb on some eval to get the quality, the latency, the cost to where you want it to be.

s131
00:11:08.079 --> 00:11:10.399
This is also all doable overnight.

s132
00:11:10.399 --> 00:11:12.560
Um, especially with the modern coding agents.

s133
00:11:12.560 --> 00:11:18.079
So, uh, usually if I'm prototyping a feature, I will, uh, put a little feedback button on it.

s134
00:11:18.079 --> 00:11:24.079
So after the LLM call or the agent runs, I'll be like, "Hey, uh just add a feedback button and a little text box.

s135
00:11:24.079 --> 00:11:30.959
And then in one of my 1 hour slots or 30-minut slots throughout the day, I'll just take the time to uh collect a bunch of feedback."

s136
00:11:30.959 --> 00:11:37.680
So I'll uh you know, run it a few times, collect a bit of feedback, be like, "Hey, this was bad or this was good or this could have been better."

s137
00:11:37.680 --> 00:11:40.880
And then just have it save JSON dumps on my downloads folder.

s138
00:11:40.880 --> 00:11:46.480
So it'll save a dump of each of the runs with the system prompt and the inputs and my feedback.

s139
00:11:46.480 --> 00:11:50.959
uh and you just collect a handful, you know, like even four, five, 10 are fine.

s140
00:11:50.959 --> 00:11:54.240
If you can get some other uh co-workers to run it, that would be great.

s141
00:11:54.240 --> 00:11:56.640
Collect like 20 or 30.

s142
00:11:56.640 --> 00:12:05.680
And then similarly, uh at 5:00 p.m. when you're setting up an overnight run, uh just say, "Hey, like here are, you know, 10 20 30 feedback JSONs.

s143
00:12:05.680 --> 00:12:07.680
This is all the information they have in them.

s144
00:12:07.680 --> 00:12:09.200
Turn this into an eval set."

s145
00:12:09.200 --> 00:12:12.639
Like just do it locally with SQLite or Markdown or whatever you want.

s146
00:12:12.639 --> 00:12:19.839
And then over the next few minutes, talk me through how would you design evals or scoring functions to optimize for this and let's do it interactively.

s147
00:12:19.839 --> 00:12:26.560
And then I want you to a build a harness that runs this call against the evals and hill climbs until the score goes up.

s148
00:12:26.560 --> 00:12:29.360
Like just do whatever you need to do to get these scores to go up.

s149
00:12:29.360 --> 00:12:32.959
And obviously with this kind of thing, if you have such little data, overfitting is a problem.

s150
00:12:32.959 --> 00:12:36.160
But even that I found these modern models, you can just tell it to not overfit.

s151
00:12:36.160 --> 00:12:37.680
And it actually does a pretty good job.

s152
00:12:37.680 --> 00:12:40.320
Uh and to just say don't overfitit, just keep it general.

s153
00:12:40.320 --> 00:12:47.519
uh run overnight autonomously until we align on a scoring rubric and then give me a full report uh in the morning.

s154
00:12:47.519 --> 00:12:53.279
And then a little tip, it's also nice to just say, "Hey, save this flow as a generic skill so I can reuse it."

s155
00:12:53.279 --> 00:12:57.040
Uh and that allows you to improve this flow over time and improve that skill.

s156
00:12:57.040 --> 00:12:58.720
And this works actually in the morning.

s157
00:12:58.720 --> 00:13:01.120
You wake up to ideally something hill climbed.

s158
00:13:01.120 --> 00:13:09.839
Uh and actually when using that feature when we we ship these features to employees to to further dog fooding to production it actually does improve the quality.

s159
00:13:09.839 --> 00:13:11.200
You know these are not overfit.

s160
00:13:11.200 --> 00:13:13.600
Um and we we use this flow quite a bit internally.

s161
00:13:13.600 --> 00:13:17.120
Uh we have an internal code reviewer similar to uh some of the external ones.

s162
00:13:17.120 --> 00:13:23.200
And same thing there we add that hill climb to pretty good quality just with these overnight runs.

s163
00:13:23.360 --> 00:13:27.360
And then the last example uh and this is again pretty new and really interesting.

s164
00:13:27.360 --> 00:13:32.000
uh you can train custom models with this mechanism overnight you know in one night.

s165
00:13:32.000 --> 00:13:36.480
So uh this is an example of a modern BERT PII classifier we trained.

s166
00:13:36.480 --> 00:13:43.839
We were trying Opus and Haiku for it and it was expensive and latency was not great and we just couldn't get the precision recall to be amazing.

s167
00:13:43.839 --> 00:13:53.839
And so uh we collected a bunch of training data and then overnight one night I was like here is a bunch of business context and a bunch of training data we collected.

s168
00:13:53.839 --> 00:14:04.320
uh a clean up the training data cloud just bolster it with synthetic data whatever you need to here's a bunch of open AI and enthropic keys uh use an ensemble of frontier models to push the quality

s169
00:14:04.320 --> 00:14:14.079
I don't know anything about what ML model will be good here so like you decide you know just give me the best model class give me a few options and in fact train two like don't even train one train two separate ones

s170
00:14:14.079 --> 00:14:20.880
uh and then I gave it AWS access I was like give me a provision sandbox do not give it prod that's not a good idea that's how you take prod down

s171
00:14:20.880 --> 00:14:26.880
um pick the right GPU and EC2 cluster like train it, test against eval examples, deprovision it, whatever you need to do.

s172
00:14:26.880 --> 00:14:28.639
Just give me back the train models.

s173
00:14:28.639 --> 00:14:30.720
Uh, and then give me a report in the morning.

s174
00:14:30.720 --> 00:14:36.800
And in fact, even put in how I would host this with my codebase on inference once this model is ready.

s175
00:14:36.800 --> 00:14:38.399
And again, ask me any clarifying questions.

s176
00:14:38.399 --> 00:14:38.880
You can do it.

s177
00:14:38.880 --> 00:14:39.519
I believe in you.

s178
00:14:39.519 --> 00:14:40.480
You're going to do great.

s179
00:14:40.480 --> 00:14:42.320
Uh, and in the morning, just have this ready.

s180
00:14:42.320 --> 00:14:44.160
And of course, it works beautifully.

s181
00:14:44.160 --> 00:14:48.800
In the morning, you have uh, two ML models trained, a full report, great results.

s182
00:14:48.800 --> 00:14:52.480
Uh, and we've, uh, pushed a bunch of these to production.

s183
00:14:52.480 --> 00:14:54.000
So those are three examples.

s184
00:14:54.000 --> 00:14:56.079
Um I would try try your own.

s185
00:14:56.079 --> 00:15:01.199
Uh but I think the main takeaway for me in doing these is really to push on task scope.

s186
00:15:01.199 --> 00:15:04.480
So this was a tweet by Simon Lass, one of the notion founders.

s187
00:15:04.480 --> 00:15:10.160
Uh and he he called out these modern models are just capable of so much more than we think.

s188
00:15:10.160 --> 00:15:23.279
uh and I think it's our jobs as leaders to really understand the contours of how much we can push these models because it affects both how we lead and how we influence our teams but also how they fit into our products and what these our new products are capable of in terms of utility.

s189
00:15:23.279 --> 00:15:34.959
So I'd really push you overnight try to think of how do I give it more and more uh scope so that we can do you know weeks of work months of work overnight and see what these models are really capable of.

s190
00:15:35.040 --> 00:15:39.680
I will say a caveat a lot of this works because of existing organizational scaffolding.

s191
00:15:39.680 --> 00:15:42.320
So you'll have to build this up in your organization.

s192
00:15:42.320 --> 00:15:44.399
You know, AI code reviewers, this really helps.

s193
00:15:44.399 --> 00:15:48.000
You have an internal one we trained, but also just external ones, whatever you need.

s194
00:15:48.000 --> 00:15:54.880
Um, agents.md hygiene, you know, cloud.md, agents.mmd, proper CI that you can trust.

s195
00:15:54.880 --> 00:15:59.279
And then we have a bunch of other things like uh really sophisticated feature flags.

s196
00:15:59.279 --> 00:16:03.920
Uh, we have a prototype branch that you can push to that goes to employees, but it doesn't go to production.

s197
00:16:03.920 --> 00:16:05.199
Just layers like that.

s198
00:16:05.199 --> 00:16:10.160
So you're not taking prod down basically as you're prototyping and showing these things off.

s199
00:16:10.160 --> 00:16:14.320
Not amazing for the CTO to take prod down.

s200
00:16:14.399 --> 00:16:16.959
I will say again code hygiene really matter really matters.

s201
00:16:16.959 --> 00:16:18.160
I've been humbled a lot.

s202
00:16:18.160 --> 00:16:20.079
You know my code has annoyed engineers.

s203
00:16:20.079 --> 00:16:21.199
It's caused sevs.

s204
00:16:21.199 --> 00:16:23.440
Uh coding agents are not perfect yet.

s205
00:16:23.440 --> 00:16:25.360
So just uh your mileage may vary.

s206
00:16:25.360 --> 00:16:34.720
you will get humbled, but I I still think it's worth it uh because you learn so much and you can model what taking on more and more scope even with a busy calendar looks like.

s207
00:16:34.720 --> 00:16:36.320
Some tactical tips on hygiene.

s208
00:16:36.320 --> 00:16:39.040
Uh I would test everything before the PR goes up.

s209
00:16:39.040 --> 00:16:43.120
That morning slot, that's a lot of what I'm doing is just testing what the overnight run did.

s210
00:16:43.120 --> 00:16:47.199
Uh even though there's computer use, etc., it's really important you test it yourself.

s211
00:16:47.199 --> 00:16:50.720
Um small readable PRs are really helpful.

s212
00:16:50.720 --> 00:16:54.240
I think they you're modeling to the rest of the team what good looks like.

s213
00:16:54.240 --> 00:16:59.920
So, if you're putting up three 5,000 line PRs, uh, other engineers on the team are going to start doing that, and that's not amazing.

s214
00:16:59.920 --> 00:17:01.440
So, uh, it's really important.

s215
00:17:01.440 --> 00:17:04.959
Your hygiene is really good because it's modeling for the rest of the team what good looks like.

s216
00:17:04.959 --> 00:17:07.760
Um, and then, ooh, this is so tempting.

s217
00:17:07.760 --> 00:17:11.520
It's so tempting to put other reviewers on code you haven't read yet.

s218
00:17:11.520 --> 00:17:12.480
Don't do it.

s219
00:17:12.480 --> 00:17:21.280
It's uh, mostly because you're going to look like an ass because there's going to be something in that code that is going to be so obviously bad and then some senior engineer is going to call you and be like, "Yo,

s220
00:17:21.280 --> 00:17:22.480
why didn't you read this?"

s221
00:17:22.480 --> 00:17:24.480
So uh I speak from experience.

s222
00:17:24.480 --> 00:17:28.959
Read the code, review it properly before you add anybody else to it.

s223
00:17:29.120 --> 00:17:31.200
So hopefully this convinces you.

s224
00:17:31.200 --> 00:17:32.720
Building is part of the job now.

s225
00:17:32.720 --> 00:17:43.600
Uh you can do it uh even with just one or two hours a day and you should uh you'll start to feel the models every 3 months as the new model families come out and you'll learn the skill on how to push scope

s226
00:17:43.600 --> 00:17:45.840
as you are uh instructing them.

s227
00:17:45.840 --> 00:17:49.760
And you'll be able to show the the the team and the org what's possible as well.

s228
00:17:49.760 --> 00:17:56.720
you know, just because you as you have this capability, you'll automatically think of ideas on what to slot in for these overnight runs and then you can model for the team.

s229
00:17:56.720 --> 00:17:58.960
Hey, this is what the future looks like.

s230
00:17:58.960 --> 00:18:00.762
Thank you.

s231
00:18:00.762 --> 00:18:02.762
[applause]

s232
00:18:15.827 --> 00:18:17.827
[music]