Learning Library

← Back to Library

Boris Chney on Latent Demand

Key Points

  • The pace of AI model development is so rapid that perspectives on tools and strategies can change dramatically within just a few months.
  • Boris “Boris” Chney, creator of Claude Code and former Meta principal engineer, emphasizes “latent demand” as a core product principle and warns against designing solely for today’s model capabilities.
  • He describes how “quad code” at Anthropic has tripled overall productivity—raising engineer output by about 70%—and stresses building solutions for the models that will exist six months from now.
  • Chney recounts his promotion to senior engineer at Meta, detailing how he led the “chats and groups” project that integrated Messenger with Facebook Groups, scaling the effort from a solo hack to a multi‑engineer team.

Sections

Full Transcript

# Boris Chney on Latent Demand **Source:** [https://www.youtube.com/watch?v=AmdLVWMdjOk](https://www.youtube.com/watch?v=AmdLVWMdjOk) **Duration:** 01:24:33 ## Summary - The pace of AI model development is so rapid that perspectives on tools and strategies can change dramatically within just a few months. - Boris “Boris” Chney, creator of Claude Code and former Meta principal engineer, emphasizes “latent demand” as a core product principle and warns against designing solely for today’s model capabilities. - He describes how “quad code” at Anthropic has tripled overall productivity—raising engineer output by about 70%—and stresses building solutions for the models that will exist six months from now. - Chney recounts his promotion to senior engineer at Meta, detailing how he led the “chats and groups” project that integrated Messenger with Facebook Groups, scaling the effort from a solo hack to a multi‑engineer team. ## Sections - [00:00:00](https://www.youtube.com/watch?v=AmdLVWMdjOk&t=0s) **Boris Chney on Latent Demand & Future AI Models** - In this segment, former Meta principal engineer and Claude Code creator Boris Chney explains the concept of latent demand, recounts cultural hurdles, highlights Quad Code’s productivity boost, and urges developers to build for AI models that will exist six months ahead. - [00:03:37](https://www.youtube.com/watch?v=AmdLVWMdjOk&t=217s) **Facebook's Generalist Engineer Culture** - The speaker describes how early Facebook encouraged engineers to combine coding, user research, and design— even soliciting cafeteria staff for feedback— fostering a generalist mindset that now drives hiring across product, data, and research roles. - [00:07:30](https://www.youtube.com/watch?v=AmdLVWMdjOk&t=450s) **Cultural Conflict Between Rapid and Reliable Teams** - The speaker describes how Facebook’s fast‑shipping product team and the reliability‑centric Messenger team clashed over metrics, organization, and values, causing a project to fail until a common objective bridged the divide. - [00:11:42](https://www.youtube.com/watch?v=AmdLVWMdjOk&t=702s) **Driving Adoption of an Internal State Tool** - The speaker explains how they recognized a shared Redux pain point, created the internal Unux solution, and boosted its adoption by analyzing issue data, personally contacting team leads, and delivering dozens of targeted tech talks. - [00:14:47](https://www.youtube.com/watch?v=AmdLVWMdjOk&t=887s) **From Broken Arms to Functional Programming** - After a motorcycle accident broke both arms, the speaker shifted from JavaScript to low‑keystroke languages like CoffeeScript, sparking an interest in Haskell and later Scala functional programming. - [00:18:42](https://www.youtube.com/watch?v=AmdLVWMdjOk&t=1122s) **From PHP to JavaScript: Promotion Journey** - The speaker reflects on the pitfalls of job hopping and recounts how taking ownership of revamping Facebook Groups' clunky PHP site into a modern JavaScript interface led to their promotion to staff/E6 at Meta. - [00:21:47](https://www.youtube.com/watch?v=AmdLVWMdjOk&t=1307s) **Optimistic UI Mutation Queue** - The speaker explains a bug with rapid button presses causing out‑of‑order responses, and how they solved it by queuing mutations for optimistic updates to preserve UI consistency. - [00:25:22](https://www.youtube.com/watch?v=AmdLVWMdjOk&t=1522s) **Three-Option Tactic for VP Reviews** - The speaker explains a tongue‑in‑cheek practice of presenting three choices so senior leaders tend to select the middle one, highlighting how this aids decision‑making and reflects the dynamics of reporting to high‑level management. - [00:29:01](https://www.youtube.com/watch?v=AmdLVWMdjOk&t=1741s) **Earning Trust and Scaling Leadership** - The speaker discusses how, regardless of past accomplishments or managerial titles, one must earn trust and authority in a new company, noting that the lack of titles can simplify this process, and explains the challenges of orchestrating large‑scale engineering scopes within a rapidly expanding organization. - [00:33:22](https://www.youtube.com/watch?v=AmdLVWMdjOk&t=2002s) **Launching a Tech Design Competition** - A leader explains how they swiftly organized a technical design contest to front‑load risk, balancing immediate action with building consensus, and shares quick‑scoping tips for tech leads. - [00:36:28](https://www.youtube.com/watch?v=AmdLVWMdjOk&t=2188s) **One-Line Change, Big Ripple Effects** - A developer explains how a seemingly minor database field adjustment for Facebook group membership sparked extensive debates over the semantics of “join” vs. “follow,” data‑model restructuring, and costly migration work, largely influenced by a senior engineer’s strong stance. - [00:40:00](https://www.youtube.com/watch?v=AmdLVWMdjOk&t=2400s) **Undoing a Massive Code Migration** - An engineer led a year‑long effort to insert conditional logic across hundreds of call sites, then reversed the change to confirm the original data model was correct, showcasing technical leadership and the willingness to challenge senior decisions. - [00:45:40](https://www.youtube.com/watch?v=AmdLVWMdjOk&t=2740s) **Automating Tedious Code Review Tasks** - The speaker describes tracking repetitive code‑review comments in a spreadsheet, converting frequent issues into custom lint rules for automation, and cautions that over‑automation can detach engineers from the code. - [00:49:06](https://www.youtube.com/watch?v=AmdLVWMdjOk&t=2946s) **Testing at Scale: Large‑Data Unit Tests** - The speaker describes how missing automated tests for massive Facebook entities (groups, profiles, events) caused production failures and how they initiated infrastructure to generate unit tests for large‑scale data sets, rallying engineering effort to prevent similar issues. - [00:52:40](https://www.youtube.com/watch?v=AmdLVWMdjOk&t=3160s) **Instagram Culture vs Feature Overload** - The speaker contrasts Instagram’s product‑centric, human‑focused engineering approach with Facebook’s, highlighting the concept of “unshipping” to avoid cluttering apps with rarely used features. - [00:56:11](https://www.youtube.com/watch?v=AmdLVWMdjOk&t=3371s) **Time Zone Freedom Revives Coding Passion** - Being in a remote Japanese time zone eliminated meetings, letting the engineer refocus on hands‑on development, which alleviated burnout and restored their deep, emotional connection to coding. - [00:59:47](https://www.youtube.com/watch?v=AmdLVWMdjOk&t=3587s) **Delegating Leadership in Engineering Projects** - The speaker describes handing a major initiative to a trusted engineer, stressing that delegation should only involve tasks you enjoy and understand. - [01:03:20](https://www.youtube.com/watch?v=AmdLVWMdjOk&t=3800s) **Engineer’s Motivation vs Corporate Portfolio** - The speaker describes their personal drive to create delightful, problem‑solving products, frustration with company‑driven portfolio decisions, and recounts early rejected pitches that later became reality. - [01:06:30](https://www.youtube.com/watch?v=AmdLVWMdjOk&t=3990s) **Choosing Anthropic for AI Safety** - The speaker explains why they left Meta to join Anthropic, emphasizing a commitment to rigorous AI safety and alignment work as an engineer to prevent escalating risks from advanced models. - [01:09:39](https://www.youtube.com/watch?v=AmdLVWMdjOk&t=4179s) **Mission‑Driven Shifts & Cloud Code Edge** - The speaker contrasts the purpose‑filled work at change.org and Anthropic with a less motivating experience at Instagram, then explains how Cloud Code broke through early AI autocomplete tools by providing genuine coding assistance instead of mere Q&A. - [01:12:57](https://www.youtube.com/watch?v=AmdLVWMdjOk&t=4377s) **Setting Standards for AI Code** - The speakers discuss Karpathy’s caution about “vibe coding,” emphasizing that AI‑generated code should be held to the same quality bar as human code and used wisely, especially for prototypes and non‑critical tasks. - [01:16:24](https://www.youtube.com/watch?v=AmdLVWMdjOk&t=4584s) **Non‑technical Teams Adopt Cloud Code** - The speaker explains how employees across roles are using Cloud Code tools like Quad Code to write SQL, DBT pipelines, and integrate with Salesforce, and asks about its advantages and stickiness compared to competitor Codeex. - [01:19:36](https://www.youtube.com/watch?v=AmdLVWMdjOk&t=4776s) **Automation Over Manual Coding** - The speaker stresses learning practical tools first and now boosts productivity by orchestrating “quad” code pipelines and AI assistants like Claude to automate engineering work. - [01:23:00](https://www.youtube.com/watch?v=AmdLVWMdjOk&t=4980s) **Trust Your Common Sense** - The interviewee advises newcomers to rely on common sense and self‑trust to cut through corporate or startup noise, avoid misaligned incentives, and focus on what users and the market truly need. ## Full Transcript
0:00The models are moving so quickly. If you 0:02ask me this question in 3 months or 6 0:03months, my answer will be totally 0:05different. This is Boris Chney. He's the 0:08creator of Claude Code and former meta 0:10principal engineer. And we talked about 0:12everything that shaped his career. Can 0:15you explain latent demand? 0:16>> Latent demand I think is the [music] 0:17single most important principle in 0:19product. 0:20>> You said that there were some clear 0:22cultural differences and that was 0:24difficult. 0:25>> Oh my god, difficult is such an 0:26understatement. It was a nightmare. We 0:28also talked about quad code and what's 0:30actually [music] happening in enthropic 0:32right now. Even though enthropic has 0:34tripled, productivity per engineer has 0:36grown like almost 70% [music] 0:37because of quad code. Don't build for 0:40the model of today. Build for the model 0:426 months from now. The one technical 0:44book I would recommend to everyone that 0:45has [music] had the greatest impact on 0:47me as an engineer is what are your 0:49thoughts on the competition with codeex 0:52and openai? 0:54Here's the full episode. 0:59I want to start at the beginning of your 1:01story with you getting promoted to 1:03senior engineer at Meta. What's the 1:06story behind the projects that got you 1:08promoted and where were you at the time? 1:11If I remember right, the project was 1:13chats and groups and this was a project 1:15to bring uh Messenger and Facebook a 1:17little bit closer together. And I 1:19actually the first few projects that I 1:21worked on at Meta, it was about 1:23Messenger and Facebook. And I think the 1:26first one was like Zach had this idea 1:28about like syncing Messenger chats and 1:30Facebook groups. Um, but there's a few 1:32of these projects just trying to bring 1:33like uh Messenger and Facebook closer 1:35together. And I think the motivation was 1:38there was this feeling that this kind of 1:40public space social product was 1:42disappearing and that things were moving 1:44a little bit more into chat and these 1:46kind of more casual realtime spaces. And 1:48so we tried a few versions of the 1:50product and chats and groups is the one 1:52that worked. I think it was like number 1:53three or number four at the time and I I 1:55was in the Facebook groups or in um 1:58Facebook at the time and I was working a 2:00lot with messenger that was like 2:01organizationally very distant. And this 2:03is a idea that I think Steve who was a 2:06PM at the time he sort of had this idea 2:08this is a thing we should build and I 2:09just picked up on that and I was like 2:10yeah hell yeah let's let's do this. Uh 2:13and so I started hacking on it. Uh and 2:15then pretty soon there were some signs 2:17of life. So I asked for more engineers 2:19and uh there were three engineers that 2:22joined. There was uh Shatambri, uh 2:24Crystal and Chaang. They were like the 2:27first three engineers that joined this 2:28and then we got some uh we got some data 2:30science support, some design support. Um 2:33and it started just on web. Then we also 2:36moved to mobile a little bit and uh 2:39yeah, I think we just kind of proved out 2:41this idea that you can have chats inside 2:43of Facebook groups and this kind of 2:44product can work. And there's just like 2:46a lot of stuff honestly that didn't work 2:47at all about it. Um it was like a super 2:50jank experience I think by modern 2:52product standards. Like back in the day 2:53everyone was building on web and all 2:55sorts of bugs were totally okay. 2:57Nowadays, I think the standard honestly 2:59like the visual standard like quality 3:00standard is a lot higher and yeah so 3:03like the the product grew and we were 3:04such a small team like everyone had to 3:07do everything and I remember like we 3:10didn't have a user researcher so I would 3:12go to the cafeteria during lunch and we 3:15would have a new feature and we would 3:16show the cafeteria workers the feature 3:18and be like hey can you figure out how 3:19to open a chat and you know like 3:21sometimes they would find it sometimes 3:22they wouldn't be able to and this is 3:24just like an observational user research 3:25study So you kind of see how people like 3:28in a particular situation can do a task 3:30and you don't prompt them that much. So 3:32you you don't want to give too much away 3:33and you kind of see where they struggle. 3:35You see what they get. Um and so we we 3:37did this and then like I I kind of 3:38taught the team how to do this. So then 3:40pretty soon we would all go to the 3:41cafeteria at lunch and start like 3:42bugging cafeteria workers to you know as 3:45just kind of like representative users 3:46to be like you know does this make sense 3:48or not? It's interesting how the early 3:51Facebook culture that you were operating 3:53in let engineers do so much outside of 3:57just like the code. For instance, you're 3:58doing UXR. It sounds like in some of it 4:02I remember in your story you did some 4:04design as well and you were coaching 4:05people to do design. So I think that's 4:07pretty interesting unique thing in 4:10Facebook's culture. I think this is so 4:13important and I think to this day, you 4:15know, on the quad code team and this is 4:17the team that I'm on right now. We 4:19really prioritize generalists. So I love 4:21working with generalists. If you're an 4:23engineer that codes, but you can also do 4:25product work. You can also do design. 4:27You have product sense. You go, you want 4:28to go talk to your users. Like I love 4:31this kind of engineer to work with. And 4:33this is actually how we recruit for all 4:35functions now. So like our product 4:36managers code, our data scientists code, 4:39um our you know user researcher codes a 4:41little bit. So like I just love these 4:43generalists and I think this is really 4:44like the way that I grew up like from 4:46the beginning when I was running you 4:47know my first startup when I was like 18 4:49I had to do everything and uh up until 4:53Facebook I worked at smaller companies 4:54where you had to do everything and I 4:56kind of feel like at big companies you 4:58get forced into this you know particular 5:00swim lane but it's just sort of official 5:02cuz like what is engineering? It's like 5:03it's a very narrow skill set, but really 5:05the thing that you're doing is you're 5:07building product or you're building 5:08infra and there's just so much more that 5:09goes into doing that end to end besides 5:11just writing code. It it was just really 5:14cool being at a place that um I think 5:16Facebook uniquely kind of rewarded that 5:18at that time. And I I think actually at 5:20the end of that half I got promoted and 5:22then I think the half after every single 5:24one of the engineers got promoted too. 5:25In those early products, there was this 5:28concept 5:29um latent demand that you mentioned a 5:32few times, which uh it sounds like was 5:35the impetus for a lot of those product 5:37directions. Can you explain latent 5:39demand? 5:40>> Latent demand I think is the single most 5:42important principle in product and I 5:45think if you look at especially at 5:47Facebook successful products, every 5:49single one has an element of latent 5:51demand. So for example, a marketplace, 5:54it came from this observation that if 5:55you looked at Facebook groups at the 5:56time, 40% of the posts were buying and 5:59selling stuff. And so Facebook groups 6:02were not designed for commerce, but 6:04that's what people were using it for. 6:06And so it's kind of cool like you design 6:08this product in a way that can be 6:09hacked. It can be abused by users a 6:11little bit, and then you look at the 6:12data, you see how they're abusing it, 6:14and then you build a product around it. 6:16And so you know like there was Facebook 6:18groups and then there were buy sell 6:19groups and then that succeeded obviously 6:21because people already wanted to buy and 6:22sell and do commerce on Facebook groups 6:24and then marketplace was next. It was 6:26just a natural extension of the same 6:27intent that people had. Um I think 6:30Facebook dating was pretty similar. I 6:32think the observation was something like 6:3460% of profile views were people of the 6:36opposite gender that were not friends 6:37with each other. Um, so you know this 6:39kind of like traditional like kind of 6:41like creeping on each other like and and 6:44I think for like um like Nate and like 6:45FMS at the time like this this was 6:47evidence that this would work. And I I 6:49think the the principle in product is 6:51you can never get people to do something 6:53they do not yet do. The thing you can do 6:55is you can find the intent that they 6:57have and then you can steer it to let 7:00them better kind of capitalize on that 7:02intent um and kind of do the thing the 7:04thing that they want more easily. I 7:06think also at this part of your story, 7:08you mentioned that you worked across 7:10orgs. You worked because you were 7:12bridging the gaps between messenger and 7:14a lot of the group's um engineering 7:17work. I'm curious, you said that there 7:19were some clear cultural differences and 7:22that was difficult. Uh do you have any 7:24advice for working across uh very 7:27different culture orgs? 7:29>> Oh my god, difficult is such an 7:30understatement. It was a nightmare. Like 7:32for Facebook at the time, we wanted to 7:34ship like we just wanted to go fast and 7:36ship awesome product as fast as we 7:39could. And then Messenger was all about 7:41reliability and performance. That's all 7:43they cared about. It was just polar 7:44opposite values. Um and this isn't just 7:47cultural. It's not just like an engineer 7:49to engineer thing. It's like the 7:50engineers on that team were suspicious 7:52of us because we would affect their 7:53performance metrics. And 7:55organizationally, their or was set up in 7:57a way to ship slowly without regressing 7:59the metrics and we were set up to ship 8:01quickly. So it's like and then the goals 8:03were totally different you know like 8:05they had SOA up times and for us it was 8:06just about daily active users and 8:08engagement. So I think for me the 8:11running was these kind of cultural 8:12values go super deep. It's not just a 8:15thing people talk about but you can 8:17actually see this in like in org design 8:18and in in goal design and in every part 8:21of everything. And honestly, I think one 8:24of the reasons that project uh failed 8:28was uh and eventually it evolved 8:30actually into something successful, but 8:32that version of the project failed was 8:34because of this difference in values. 8:36So 8:38I think that fundamentally if you want 8:40to get companies with really different 8:42values to succeed and kind of work 8:44together, you have to find some kind of 8:46shared goal or kind of shared interest, 8:48shared belief, some kind of hypothesis 8:50that they want to test together that 8:51would be really interesting for both of 8:53them if it worked. Um, and I think this 8:56like chats and groups thing 8:57fundamentally was really cool for 8:58Facebook, but it's not that cool for 9:00Messenger for a lot of reasons. So, 9:02knowing what you know now, how would you 9:06change things going back to that kind of 9:08project? 9:08>> I think I probably would have gone to 9:10Zuck and just been like, if you're 9:11really serious about this thing, we we 9:13should move Messenger into the Facebook 9:15or and I think this has since happened 9:17and it's actually happened like a few 9:18times like Messenger was in the or then 9:20it moved out and then it moved in then 9:22moved out. It's a big company like this 9:24happens. But I think fundamentally for 9:26this kind of thing to succeed, the the 9:27common report can't be, you know, the 9:30common manager can't be like Chris Cox. 9:32It has to be like a little bit lower 9:33down. And so you can structure the orgs 9:35to be a little bit more collaborative. 9:37>> I see. To align the incentives so you 9:39don't get that kind of constant 9:41struggle. 9:42>> Yeah. Exactly. at this point in your 9:44career, I saw there were a bunch of 9:46really interesting side projects that 9:48you had and I'm kind of curious like 9:50what what's the butterfly effect of 9:51those kinds of projects? So, for 9:53instance, um even before you got to 9:55meta, you worked on Undux, the state 9:58management framework for React. Um I'm 10:01curious like how did that impact your 10:04career if at all? 10:06>> Yeah, I mean for me side quests are so 10:07important and for me like when I hire 10:09engineers this is definitely something I 10:11look for. I want people with side 10:13quests, like cool weekend projects, cool 10:16side projects, like even someone that's 10:18like, you know, just really into like 10:19making kombucha or something. Like you 10:21want people that are generally curious 10:22and interested in stuff outside of their 10:24main work. These are kind of 10:25well-rounded people. These are the kinds 10:27of people I enjoy working with. I think 10:29for me, this is where a lot of my growth 10:31came from is um working on these kind of 10:35side projects. So something like Unex 10:37honestly where it came from is uh React 10:40state management is honestly 10:41unnecessarily complicated and at the 10:44time the state-of-the-art was there 10:46there was like flux and then there was 10:48this other thing called Redux and I just 10:50couldn't wrap my head around Redux. I 10:52was just you know I I consider myself 10:53kind of average engineer like I build 10:55product. I'm like I'm not one of these 10:57like incredible systems engineers. And 10:59so for me like Redux at the time I had 11:01these concepts of like you know like 11:02reducers and this kind of like just like 11:05this like very complicated flow you had 11:07to go to to just like update a little 11:08state and I just couldn't wrap my head 11:10around it. So I I built a simpler thing 11:12that seemed to work. I used that um I 11:15was volunteering at a nonprofit at the 11:16time and uh they started using it and 11:18their engineers liked it. And then when 11:20I joined Facebook, I saw a lot of kind 11:23of frustration around Redux usage 11:26because there was a internal group for 11:28people that use Redux and there were all 11:29these questions where people were asking 11:30the same questions I did. Um, you know, 11:33like when you as an engineer or, you 11:35know, as a product person are running 11:36into a problem, sometimes it's just you. 11:39Often it's other people too. And I think 11:42it's important to build the spidey sense 11:43for like when this problem might be 11:44shared by others. And so this is a 11:46problem that definitely was shared by 11:48others. And I could kind of see this in 11:50support posts and by the difficulty my 11:52team had using Redux. 11:54And so I launched uh Unux internally and 11:57Unex is like it's fine. It's like not 11:58that great of a product, but at least 12:00it's better than Redux. And um at 12:03Facebook I didn't actually know how to 12:04get adoption. So I kind of posted about 12:05it. A few people started to use it. Um I 12:08remember there was like Jeff Case on the 12:09notifications team was a big early 12:10adopter and we spent some late nights 12:12debugging some like gnarly like 12:14notification related bugs due to it. And 12:18um I wanted to get more adoption. And so 12:20what I did was I wrote a little script 12:22and I scraped the group of uh people 12:24reporting issues and I just tallied them 12:27by team. And then I reached out just 12:30over chat to the tech lead and the 12:31manager for every team and I scheduled 12:33uh like a tech talk just for that team. 12:36And I think overall I did maybe 20 30 40 12:39tech talks something like that over the 12:41course of like a few weeks. Yeah. 12:43>> And I I remember just like biking around 12:45the meta campus and kind of doing these 12:47talks and it was so fun cuz people were 12:49so engaged and they were just so excited 12:50that someone cares about solving this 12:52problem that they they really have and I 12:55think at some point on was like the most 12:57popular state management framework at 12:59Facebook and then I think it got pretty 13:01quickly replaced by recoil and kind of 13:02more modern alternatives and nowadays 13:05it's like relay and and things like 13:06that. Does that kind of side project 13:08appear in your like performance review 13:11or is it like help you in some way? So I 13:13think it was in my performance review. I 13:15think by meta standards it's kind of a 13:17cherry on top. It wasn't really you know 13:19something that kind of gets you to the 13:20next level in itself. Um but I had a lot 13:22of other side quests around uh around 13:24that time too. Like at some point I got 13:26really into TypeScript actually and this 13:28was like at the previous company I was 13:29at. We were using it. There weren't a 13:31lot of good resources. And so I just 13:34like started writing a book about it cuz 13:36I was like someone like someone should 13:37do this. It's crazy it doesn't exist. 13:39This language is just like magnificent. 13:41It's it's just this like really really 13:43shockingly 13:45shockingly good design. It has all these 13:47ideas that no other language had at the 13:50time. Um you know things like eventually 13:53like at the time there were no 13:54conditional types but like conditional 13:56types like literal types for everything. 13:58Uh map types these are just like 14:00absolutely insane things. like even the 14:02like the gnarliest Haskeller is going to 14:04be impressed by this kind of language 14:05feature, but no one was writing about 14:07this stuff and so I just got like super 14:09into it and like wrote this book and it 14:11it just sort of like ate up like a year 14:13of my life. [laughter] 14:14Would not recommend it. Um but it was 14:16really fun to go really deep on it and 14:18then I also started I think like the 14:20world's biggest TypeScript meetup at the 14:21time in in San Francisco and that was a 14:24really cool chance to meet uh there was 14:25like Ryan Doll who created Node.js. 14:27there was um just like all all these 14:29like famous JavaScript celebrities and 14:31it just sort of made me realize like all 14:33these people are just people and um you 14:37know like everyone just like builds cool 14:39stuff and some of it's cool some of it's 14:41cool at a particular time but it's all 14:43just people and anyone can do this 14:44stuff. 14:45>> Did you end up using TypeScript or that 14:47technical depth later in your time at 14:49Meta or or maybe even in anthropic? 14:52Yeah, I think there was, you know, it's 14:54funny. I actually like I used to not 14:56care about languages and then at some 14:58point, this was maybe like 10 years ago, 14:59I used to ride a motorcycle and I got in 15:01a pretty bad accident actually. I broke 15:03my arms. 15:03>> Oh my god. 15:05>> Both of them. 15:05>> Both of them. Yeah. I had like two 15:06slings on. 15:07>> Oh my god. How'd you code? 15:08>> Um, so that was the hard part. So like I 15:10actually like couldn't code for like a 15:11month and then you know my hands like 15:12still kind of hurt and so like I 15:14couldn't write JavaScript which is what 15:16I used to write at the time. And so I 15:19actually had to branch out and learn 15:20other languages because they literally 15:22used less keystrokes. 15:24And so I started with like coffees 15:25script because there was like less 15:27parenthesis and stuff. I don't I don't 15:28think that language even exists. No one 15:29uses it nowadays. Um but that's also how 15:32I got into Haskell and kind of 15:33functional programming. Um cuz it was 15:35kind of you can do the same thing with 15:36fewer keystrokes and that that was like 15:38literally the motivation at the time. 15:40And then at some point I was working in 15:43uh at a hedge fund and this was like 15:45before uh before Facebook and I had a 15:48co-orker Rick who was really into Scola 15:50and I really didn't understand Scola and 15:53uh he he kind of really got me into it 15:55and he got me into this like functional 15:56programming side of the house and this 15:58is still like the one technical book I 16:00would recommend to everyone that has had 16:02the greatest impact on me as an engineer 16:05is this book called functional 16:06programming in Scola and you're probably 16:09never going to use schol today, but the 16:12way it teaches you to think about coding 16:14problems is just such a change from the 16:16way that most people were encoding 16:18either practically or in school is just 16:21it's incredible. it's going to 16:22completely change the way that you code. 16:24And so for me it was it was kind of like 16:26Scola was like there was kind of like 16:28Haskell and kind of coffecript these 16:30kind of few key languages that was like 16:32a first step then Scola and then 16:35Typescript and I think this kind of this 16:37changed the way I think because now I 16:39think in types when I code the thing 16:41that matters in your code the most is 16:42the type signatures. This is more 16:43important than the code itself. And so 16:45like getting this right leads to very 16:47clean code. And so even at Facebook 16:50where I was writing mostly kind of flow 16:51and and hack and then later at Instagram 16:53Python um it it was very helpful here at 16:57anthropic I mostly write typescript and 16:58python um so it's actually quite 17:00relevant but I but I think the the 17:02bigger lesson is just think in think in 17:04types 17:05>> at this point in your career uh you 17:07mentioned that you came in underleveled 17:09you came in as a mid-level engineer even 17:12though you had a lot of experience and 17:13you said um in hindsight you were lucky 17:16to be underleveled and I'm curious is 17:18what's the the thinking behind that? 17:21>> Just lower expectations. 17:23Um yeah, I feel I feel like at a big 17:26company there's all these kind of um 17:30you know like at every level there's 17:32certain expectations in terms of kind of 17:34the project impact and people impact and 17:37all this kind of stuff and the the 17:39specific criteria are kind of different 17:40across companies. Um but there's um I 17:44think a lot of it is about either 17:46project impact or kind of checking a 17:48bunch of check boxes and all this just 17:50takes a lot of time. Um and so I think 17:52coming in under level it just gave me 17:54the space to explore and uh just like 17:57build cool stuff for the sake of 17:59building cool stuff. Definitely. And I 18:01wonder if it also helps with building 18:05momentum. Like what I mean is if you 18:07came in as a mid-level or E4 and then 18:11you you crushing it. Everyone's saying 18:14Boris is amazing. What this is crazy as 18:17opposed to you came in at your whatever 18:20I don't know expectations and you did 18:23good. I think there can be this uh 18:26effect when you come in and you really 18:28wow everyone. you have such a strong um 18:31you know first impression I think can be 18:34helpful for building a good reputation 18:36that gives you more credibility more 18:38projects and stuff like that in the 18:40future. 18:40>> Yeah, I think that's totally true and I 18:42I think actually this is probably good 18:43advice for any company is like I think a 18:46lot of times engineers switch jobs and 18:47they really push you know like I want to 18:49go to a different company and I want 18:50like a level plus one or whatever 18:52[snorts] and actually there's a lot of 18:53downsides to that like you said. Going 18:55on to the thing that got you promoted to 18:57staff or E6 at Meta, I'm curious the 19:00story behind um you know where you were 19:03at the time and what got you promoted 19:05into more of that leadership position. 19:07So what was happening was chats and 19:09groups that was launched and that that 19:11was going and there was kind of a team 19:12working on this. Um, and uh, I actually 19:18had a I I'd done a lot of JavaScript 19:20before I joined, but at Facebook, I'd 19:23never actually written JavaScript cuz it 19:25was all PHP. 19:27>> And so I really just wanted to write 19:28JavaScript. And we had this like web 19:30interface. And for Facebook groups in 19:32particular, a lot of people use web as 19:34opposed to mobile because, for example, 19:36for like being a group admin or 19:38whatever, it's just easier to do on a 19:40big computer with a with a keyboard and 19:41stuff. And at the time the site was 19:44really janky. It was like a static site. 19:46It was all PHP. There's these little 19:47bits of JavaScript that are injected a 19:50little bit in different places. There's 19:52all sorts of inconsistent state like all 19:54these problems that come out of it. It's 19:55just it doesn't feel like a good UX. And 19:58so I wanted to rewrite it in JavaScript 20:00and I got a lot of push back from the or 20:01at the time. And I think the big reason 20:04was that the info just wasn't really 20:05ready for it. 20:08Luckily, at the same time, Comet was 20:09starting and Comet was like the rewrite 20:11of facebook.com on desktop. And this is 20:14like Tomcino, who's who's now at Versel, 20:16this was Jing Chen. Um, there there's a 20:19bunch of these kind of core people that 20:20that were working on this. And I just 20:23really wanted to be involved. So, I 20:24reached out and asked how I can help. 20:27And I offered Facebook groups as the 20:28guinea pig for it. And I didn't ask 20:30anyone. Just [laughter] kind of just 20:32kind of did it. And then uh later I kind 20:34of went to my leadership in Facebook 20:35groups and I was like, "Hey, comments's 20:37coming. It's going to be a bunch of 20:38work. We can get ahead of it." Kind of 20:40set the standard for everyone, build 20:41relationships with these other teams. 20:43And I still got a bunch of push back 20:45that was like, "Hey, you know, you can't 20:46put 20 engineers on this." And uh after 20:49a bunch of reviews and kind of haggling 20:50for engineers, I think we put we got 20:52like 12 engineers or something like that 20:54cuz it was a pretty big migration. You 20:56know, it's going to take like a year. 20:58groups is the single biggest product 21:00surface in all Facebook, which is 21:02actually kind of surprising. 21:04Um, yeah, and the the migration kind of 21:06worked and and I think something that 21:07was pretty fun about it besides just 21:09like building relationships and 21:10friendships with this infra team I never 21:12would have worked with otherwise, which 21:14was in itself so rewarding and so fun. 21:16Um, I think a lot of it was we got to 21:19influence the direction of Comet. And 21:21it's kind of weird because for an info 21:22project, a product team often cannot 21:25influence the direction. and they're 21:26more seen as a customer of it. But what 21:28happened here was because we helped 21:30co-build it, we built a lot of the 21:32abstractions that were then used by 21:35other teams that were also building on 21:37Comet. Um, and you know, for example, a 21:39particular one I remember was like relay 21:41mutations. So like you send API requests 21:43and you need some sort of consistency. 21:45Um, but there's actually this bug where 21:47like let's say there's like a button and 21:49you press the button. Every time you 21:50press it, you send a post request. And 21:52uh every time you press the button, it 21:54toggles the state of that button. For 21:56really nice UX, what you want is as soon 21:58as you press the button, the state 21:59should toggle, which means you need an 22:01optimistic update. But also, uh when the 22:04network request comes back, you need to 22:05also update the local cache to make sure 22:07it's consistent. And if you're just like 22:09mashing that button, what can happen is 22:10that the responses come in out of order 22:13and you might end up with a different 22:14state than what was in the UI. Um and so 22:17I wrote a system to kind of queue up 22:18mutations. So it was like consistency at 22:20the cost of reliability and this was 22:21kind of the right trade-off at the time. 22:23Uh and everyone ended up using this and 22:25this is how I met like Joseona and a 22:27bunch of the relay team that was working 22:29on the the data stores. Um, and it was 22:31just really fun. And this is something 22:33that since then and before then and you 22:36know whenever I work with engineers, I 22:39just love when people go a layer deeper 22:41and uh, you know, just try to figure out 22:43like what's going on and like just 22:44because you're a product engineer 22:46doesn't mean you can't build infra. Just 22:47because you're an infra engineer doesn't 22:49mean you can't go talk to users like 22:50just be curious about these other parts 22:52of the stack. 22:53>> Definitely. and in your agency and 22:56getting ahead of comet or this big 22:58JavaScript rewrite. You mentioned in 23:00your in your writing that you know 23:02getting ahead of that actually gave you 23:04a lot more control and also dibs on 23:06opportunities. So when you talk about 23:08opportunities there is is this what 23:10you're kind of talking about like 23:12building these fundamental pieces of 23:13prod infra that are impactful for 23:16everyone that's going to take on the new 23:18platform? 23:19>> Yeah. Yeah, that's an example of it. Um, 23:21and then maybe you know like a different 23:23kind of example is Comet was a lot 23:26higher quality than the thing that came 23:27before because it, you know, it's like a 23:28single page web app. Um, so it can just 23:30feel a lot more polished. But we hadn't 23:32yet figured out like what exactly 23:34quality means on the product side. And 23:36so I wrote a bunch of notes trying to 23:37define that and then did a bunch of tech 23:39talks trying to just like teach people 23:40on other teams like here's what we 23:42learned about quality. Um, and just kind 23:45of like setting up the conversation 23:46about that. 23:48you mentioned a big headcount ask for I 23:51guess this migration to comment you know 23:54I feel like I'd be curious what that 23:57would look like in today with these new 24:00tools like cloud codeex etc. I'd be 24:04curious like knowing what you know now 24:06about cloud code and you let's say you 24:08were in charge of doing that same 24:10scoping for that same job. How many 24:12engineers do you think it'd take to do 24:14that 12 engineer job? 24:16>> Yeah. So I think overall to move 24:18Facebook groups it it started with 12 24:20engineers but I think at the end it was 24:22maybe like 20 or 30 engineers or 24:23something for about two years. So it 24:25turned out to be a pretty big project. 24:27Um I think nowadays it would be maybe 24:30I don't know like five engineers for six 24:32months something something like that. 24:34>> So a fourth of the fourth of the time 24:37and um like more than a third or less 24:40than a third of the engineers as well. 24:42>> Yeah. Yeah. cuz you just like everyone 24:44would just have a bunch of quads running 24:45in parallel and you know just like let 24:47it cook for a couple hours and then it 24:48comes back with a PR and then you give 24:50it like puppeteer or something so it can 24:52kind of see the UI and and adjust and I 24:54I think that's pretty much all it would 24:55be and then I you know nowadays the 24:58world we're in is so different from a 24:59coding point of view because the models 25:03are moving so quickly that you know if 25:06you ask me this question in 3 months or 25:076 months my answer will be totally 25:09different in 6 months the answer might 25:11be this is actually one engineer 25:13um it's just moving so quickly now it's 25:15really hard to to do these estimates or 25:17to predict how they're going to change 25:18in the future. 25:20>> At this point in your career you you had 25:22mentioned something maybe it was 25:23tongue-in-cheek I'm not sure. You said 25:25this was when I learned to always 25:27present three options in VP reviews 25:29since 80% of the time they'll just pick 25:32the middle option and then it says you 25:34your VP picked the middle option in 25:36frenies. Um what's the thinking behind 25:38that? 25:40Yeah, this is this is very much tongue 25:42andcheek. Um, but maybe this is actually 25:44kind of true at Meta at the time. 25:45[laughter] 25:47Um, I think like decision makers that 25:49are far away from the work want to know 25:52that you did the due diligence of 25:55finding the right options and the right 25:57trade-offs and that you did the work, 25:58but they also want to contribute somehow 26:00to the decision. Um, so, you know, the 26:03middle option is kind of the easy way to 26:04do that. It's a little tongue-in-cheek 26:06because I think not all leaders are like 26:08this. A lot of leaders will do their 26:10work themselves. They trust their teams 26:12more or less. There's sort of there's so 26:14many different ways to operate. Um, but 26:16at the time I remember we had like a 26:18pretty non-technical leader and this was 26:20kind of the way to to help her make make 26:22decisions. I think at this point in your 26:24career you had the most proximity you've 26:27had to senior management. It's you said 26:30you were reporting to uh a senior 26:32director at some point and you were 26:34involved in a lot of huge scoping 26:36conversations. I'm curious what's the 26:38downstream effects of reporting to 26:39someone so senior like that. 26:41>> Yeah, I think it kind of depends on the 26:43engineer and it depends on the company. 26:46Um, so for example, like you know, now 26:48I'm at Anthropic and I think at 26:50anthropic it doesn't matter. Um, it 26:52doesn't it doesn't matter which level 26:53you report to. There's some of the most 26:55senior people at the company report to 26:56line managers. Um, a lot of the line 26:59managers are like XCTO's and things like 27:01this. So um, it actually doesn't matter. 27:04So I think this is kind of like a meta 27:05it's a very meta-pecific cultural um 27:08cultural observation. I think there's 27:11sort of like two things going on. So one 27:13is you want at meta you needed uh as an 27:17engineer you always needed to find 27:19scope. Some of this you can find 27:21yourself and then some of it your 27:22manager helps you find or you know your 27:24tech lead or the people you surround 27:25yourself with and you know like the PSC 27:29process is like growing like famously 27:30growing at meta and so you just have to 27:33constantly talk about your impact and 27:34like scope is like the biggest 27:35contributor to that like if you have 27:37enough scope and you executed well 27:38that's impact that's the formula I think 27:41the other part was at meta no one had 27:45titles so even the most senior engineers 27:47their title was software engineer which 27:49I actually really love. And um you know 27:52like Bell We L We L We L We L We L We L 27:52We L We L We L We Labs had this with 27:53like member of technical staff and this 27:55is true at anthropic too, but we 27:56actually go even further here. 27:58Everyone's title is member of technical 28:00staff. It doesn't even matter if you're 28:02an engineer or a PM or a designer, it's 28:04all the same title. Um, and I actually 28:06really love it because back to this 28:09point of working outside your lane and 28:11doing stuff that just should be done 28:14and, you know, like are just good things 28:15to do regardless of what you are 28:17personally expected to do. Um, I think 28:20this kind of culture just sets that up. 28:22I mean, I I I see a lot of the benefits 28:24of the no titles. I could also see a 28:27case though where um and maybe this is 28:29only true for big companies where you 28:31reach out to someone across the company 28:33and you say hey I'd like to I don't know 28:36do this collaboration and if your title 28:38said director or whatever it kind of is 28:42like a shortcut for them to understand 28:44how seriously to take you or how to 28:47interact with you like if you're a 28:48designer or some other role. So I mean 28:52now anthropics got a bit bigger at this 28:54point. Do you see uh any of that? I 28:56mean, people probably all know you, so 28:58maybe you don't see it as much. 29:00>> Yeah, I think I think this is definitely 29:01the downside. I think the upside 29:03outweighs it, which is you have to earn 29:05trust. And I I think this is true. Like, 29:08you know, regardless of what company 29:10you're at, you got to earn it. And just 29:12because you did a cool thing before, 29:14doesn't mean that you have you should 29:16deserve respect. Well, everyone deserves 29:18respect. Doesn't mean that you should 29:19deserve authority at a new company in a 29:21new setting. Um, so I think even for 29:24people coming in with manager titles, 29:25you kind of have to earn it. And in some 29:26ways, having a manager title makes it a 29:28little bit harder to earn this kind of 29:30trust. Um, so as an IC, you got to do it 29:33either way. And I I think just the lack 29:34of titles makes it a little easier. At 29:37this point in your career, you were kind 29:38of like becoming more and more of a tech 29:41lead or Uber tech lead. And I think you 29:44had a few stories where you scoped out 29:46work for hundreds of engineers. And I 29:48was just thinking about how do you do 29:51that if there's so much to scope and you 29:54know you're one person. How do you go 29:56about doing such massive scoping 29:59requests for leadership? 30:01>> Yeah, this was a totally insane time. So 30:03I worked a lot with uh Tina Shutchman 30:04who's uh she she's now at Microsoft um 30:07but she was she was my manager at the 30:08time and then uh Ephe who's who's my 30:10manager after and there was a lot more 30:13investment going into Facebook groups at 30:16the time. So I think the org was maybe 30:19150 or 200 people when I joined and by 30:21the time I left to Instagram I think it 30:23was like 600 or 800 people something 30:24like that. So there's this feeling from 30:27Zach that Facebook app should be all 30:30about communities and he just wanted us 30:32to go like faster and faster to make 30:34that a reality and you know as an 30:36executive your kind of biggest way to do 30:38that is to put the right people in 30:40charge of decisions and then uh to give 30:42them resources and so like in you know 30:44in the case of meta it's it's just 30:45engineers um you don't need like GPUs 30:47for this you need like engineers to do 30:49stuff and so we pitched this project to 30:52uh to Zach and it was called communities 30:54as the new organizations that was like 30:55the internal name and uh he grin just 30:58like a a bunch of headcount to go 31:00towards this and so we just had to 31:01figure out what these people will do and 31:04you know for him I I get it it's like if 31:06the thing is important you got to put a 31:07bunch of people on it in hindsight what 31:09I would have done differently is I I 31:11would have put way less people on it 31:13because what matters is like solving 31:15people's problems and building awesome 31:17product and this actually has to kind of 31:19be bottoms up and you kind of want to 31:21like slowly dial this up as you find 31:23product market fit for new product lines 31:24you can't just do it all at once. And uh 31:28yeah, we just had to like scope out all 31:30the stuff. Like there were weeks where I 31:31had to, you know, do like a scoping dock 31:33for like, okay, we're going to put 30 31:34engineers on this. Here's like three 31:35technical options. We're going to pick 31:36this one. Next project, we're going to 31:38put 20 engineers on this. Here's three 31:39options. We're going to pick this one. 31:41Next project we're going to do. And just 31:42like doing this like over and over again 31:44just to have like, you know, some some 31:46sort of confidence that this thing isn't 31:47totally crazy. We did some baseline 31:49technical scoping roughly matching the 31:51number of engineers to the project. And 31:54there there's actually some pretty fun 31:55stuff like I remember we were trying to 31:57merge Facebook groups and uh pages at 31:59some point like in the in the data model 32:01side and this was this like very gnarly 32:03migration it would have been you know to 32:06fully do it this is like many years and 32:07like probably hundreds of engineers to 32:09fully do it because you have to do it 32:10across like the data model the product 32:12layer integrity systems ad systems 32:14there's just all sorts of stuff that has 32:16to get merged and at the time Ysef 32:18Carver uh he just joined I think he came 32:21from either profile or events like a 32:23different or that that joined forces 32:25with groups to make this happen and he 32:28was working on it but he was kind of 32:29struggling with a with a decision at the 32:30time and I think he was even more senior 32:32than I was but he just like wasn't 32:33making the decision on the data model 32:35and so I just took a bunch of people and 32:37I was like all right the tech leads 32:39across the entire org we're going to 32:41spend the next like 3 hours on this day 32:44and we're going to do this like 32:45essentially like game where we get to do 32:47architecture and so I split everyone up 32:50into two teams I think it was like blue 32:51team and green team or I I forget what 32:53these were. And we gave everyone this 32:55like this problem of like how do you 32:56merge these data models? Here are the 32:58requirements. And then everyone had 3 33:00hours in a whiteboard and they had to 33:01come up with a design. And what was cool 33:03is that going into it, we had no idea 33:06how we would do this because it just 33:07seemed too crazy of a problem. But the 33:10going out of it, we had two designs that 33:12were 80% the same. And so it was really 33:15obvious what we could execute on. And 33:16then the 20% where the differences were, 33:18it was very obvious where the risk was. 33:20And so we could kind of front front 33:22frontload a little bit of that risk with 33:23a little bit of technical spikes. Um but 33:25also we can just start execution right 33:27away because we knew exactly what we had 33:28to do. 33:29>> Yeah, that was really interesting when I 33:31saw that it was like a technical design 33:34competition with all the senior 33:36engineers and you just put people in 33:38separate rooms to come up with um I've 33:42never heard anything like that. When you 33:43proposed that idea for this design 33:45competition within the org, were people 33:48excited about it or was it like kind of 33:50a crazy idea? 33:51>> Yeah, it was sort of crazy. I mean, with 33:52this sort of thing, you just have to 33:53kind of do it. So, I just I just kind of 33:55told everyone, "Hey, we're doing this." 33:56And then I just put it on everyone's 33:58calendar and um it just seems fun, you 34:01know? So, like as an engineer, you would 34:02want to do it. But I think this is the 34:04sort of thing where like sometimes you 34:05need consensus and sometimes you just 34:07have to act. And in this case, because 34:09the path was unclear, it was important 34:10to act. But at the same time, I didn't 34:13know how to proceed. So, we had to kind 34:15of get everyone together to build 34:16consensus. And so, I think it's like as 34:18a leader, you're kind of always juggling 34:19these kind of two things. 34:20>> After that experience, just giving being 34:23given hundreds of engineers and scoping 34:25things out, do you have any tips for 34:28someone who's like a tech lead who's 34:30needs to do quick, you know, scoping? 34:32Anything that worked well for you? I 34:34think the biggest thing I think the 34:36biggest foe that I've seen is people 34:37just taking too long and getting too 34:38into the weeds. there's always an 34:40infinite number of details. Just start 34:42with a high level. You know, most 34:44technical scoping you can do within like 34:4530 minutes very very roughly. Um and if 34:49you don't know the systems like nowadays 34:50you would just use quad code run in the 34:52codebase and just ask it to like you 34:54know like what are all the systems 34:55involved? They can actually just do this 34:56for you. And this is another just 34:58totally insane change. You know I when I 35:01was doing this stuff I never would have 35:02expected that AI could do this for me 35:04now. Um but now it does. in the past. I 35:09think that would have been my biggest 35:10advice though is uh just time box it. 35:14Spend maybe 30 minutes, maybe like 35:15couple hours max. If you have to like 35:17dig through code and stuff, um 35:19definitely reach out to experts and just 35:22make a list of experts. Talk to all of 35:23them. Run the design by them. Don't just 35:25ask them for input. Give them a straw 35:27man cuz then they can actually like give 35:28you feedback on it and it's something to 35:30go off of. Continuing with your career 35:32story, I think the thing that got you 35:35promoted to senior staff or IC7 was um 35:38public groups on Facebook. So I'm 35:41curious like the story behind your 35:42involvement in that and you know 35:44anything interesting that happened at 35:46that point. Yeah. So public groups was 35:48one of these projects that came out of 35:49this uh the scoping for um you know like 35:52making Facebook groups more about 35:53communities. There's this like one very 35:55narrow change that we wanted to make 35:56that seems so simple on the surface, but 35:58it was so complex under it. And it's 36:00just funny like explaining this to 36:02anyone that wasn't there. They're like, 36:03"Wait, this is like a oneline change." 36:04And I'm like, "No, it's not." It's like 36:06it was very difficult to pull it off. 36:08And so the change was in order to 36:10participate in a public Facebook group, 36:12you no longer have to join first. 36:14>> So you're saying um you can just view 36:18like you have read access for all all 36:20groups essentially or public groups? 36:21read access for all groups and for some 36:23groups even comment access. So you can 36:25comment without joining first. 36:27>> Interesting. 36:28>> And this is the thing you know it feels 36:29like a oneline change and it actually 36:31was a oneline change but there's all 36:33these downstream implications that were 36:35so tricky. So one is um you know in the 36:38data model there's essentially a field 36:39in the database that was like group 36:40member and we had this like really 36:43intense technical debate about like 36:45these people that are commenting in a 36:46group are they group members and the 36:49model also changed where before to join 36:52a Facebook group an admin had to approve 36:55you so there's kind of a vote of 36:57confidence that you can be in this group 36:59and then after we switched to this model 37:00where to join a public Facebook group 37:02you just you just essentially press like 37:03follow and we actually went back and 37:05forth should it be join or follow like 37:06what's the right verb to describe this 37:08but it was essentially follow cuz 37:09there's no reciprocal action you know if 37:12you follow a group are you a member like 37:14should you be stored in that same part 37:16of the database and we we just went back 37:19and forth on this for a while and I 37:20remember at the time there was this like 37:22really senior engineer Bob he was kind 37:24of the most senior engineer in the in 37:25the or at the time and he felt very 37:27strongly that it should not be the same 37:29thing and he kind of pushed us pretty 37:30hard um even though it would be a ton of 37:33engineering work to migrate stuff to 37:35make at a different thing. And so we did 37:37this work um because he was actually one 37:39of the early engineers on Facebook 37:40group. So he knew it really well. Um and 37:42he felt pretty strongly. There's a bunch 37:45of these other like downstream changes 37:46around uh like moderation and different 37:48new like admin tooling that admins would 37:50need to handle kind of the influx of 37:52spam and things like this. And I 37:54remember at the time thinking like if 37:56anyone can make a comment, the comments 37:57are just going to be like filled up with 37:58spam. And I had to hard I had a hard 38:01time kind of convincing people of this. 38:02And so at some point I built this like 38:04Monte Carlo like visualization of how 38:06this would work. And it was just like 38:08this like really simple kind of like 38:09scratch pad of you know like a comment 38:11comes in there's a certain probability 38:12of it being good or bad and then like 38:14what actually happens to comments. And I 38:16think that actually did a pretty good 38:18job of convincing the integrity teams to 38:20jump in and help with this. And so at 38:21the time the pages integrity team jumped 38:23in and they helped with a comment 38:25ranking because kind of ranking spam 38:27comments lower was the main technical 38:29mechanism to make it so people don't see 38:31these comments. So there's a bunch of 38:33these like pretty gnarly downstream 38:34implications of uh letting people 38:36participate. There's also this data 38:38model migration that we're doing. And so 38:40to do all this, we had to staff a big 38:42team to um to kind of make this happen. 38:44And so we hired a new director, Yammen, 38:46who um hired a bunch of engineers. 38:49There's a bunch of internal transfers. 38:50So some of the most senior engineers 38:51from the org like uh there was like 38:53Henry Henry Long uh Joe Cham there was 38:57like a few other engineers and they were 38:59all working on this and I was the same 39:01level as them. I was like a you know an 39:03IC6 at the time and so were they and I 39:07remember just feeling this kind of 39:08imposter syndrome of having to kind of 39:11direct them and kind of point them at 39:12work knowing kind of in my mind that 39:15we're the same level even though levels 39:17are hidden. you kind of you know you 39:18know through like rumors and stuff who's 39:20what you know in hindsight I think this 39:22is sort of like misplaced imposttor 39:24syndrome because levels don't matter at 39:27all this is my current view and you know 39:29some people that are very junior can 39:31shoot way higher than that and just give 39:33you amazing results some people that are 39:35very senior can give you terrible 39:36results and so the level actually 39:38doesn't matter that much but at the time 39:41I remember just like really thinking 39:42about this and it it was just kind of 39:44hard to step into this role 39:47And eventually I did it. And it's funny, 39:50eventually the thing that got me the 39:51promo to IC7 39:54was reversing this decision that Bob did 39:56cuz he wanted to do this big migration. 39:58And we did it. And it was just like, 40:00dude, it was so much work. It was like 40:02it was like 6 months or a year of work 40:03or something just migrating just 40:05hundreds and hundreds and hundreds of 40:06call sites to to do this correctly. And 40:09then technically I felt like actually 40:12what we did is we essentially just added 40:13an if else at every single one of these 40:15call sites in the process. We audited 40:17all the call sites. So we kind of knew 40:19that it was safe but uh we didn't 40:21actually change the logic. And so 40:22actually what we learned is that yes 40:24member is the right field to model both 40:26followers and group members. This was 40:28the right decision. And so I pushed the 40:31same engineer that did this to then undo 40:32it. it was the right thing to push this 40:35engineer because it showed maturity on 40:37his part that he said yes and was able 40:38to do it. He also had the most context 40:40technically so he could do it the best. 40:43And I think for Bob it made he he felt 40:47better about me as a technical leader 40:49because he knew that I was wishing I was 40:51willing to pull back to to push back on 40:54um decisions that even senior folks 40:57make. Um and in the end this was the 41:00right thing. So we reversed the 41:01migration. It also took a long time to 41:02do it, but in the end it made it so 41:05everyone building on this info could do 41:06it and everyone wasn't always constantly 41:09bumping into this like should I use this 41:10field or or this field. 41:13>> Yeah, I'm curious about that part cuz 41:15you had a strong technical disagreement 41:17with Bob or senior TL. Um, but the 41:21outcome at the end is actually it seems 41:23like it strengthened the relationship. 41:25He was a champion for you in your your 41:28promotion. So, I'm curious, how would 41:30you recommend going about strong 41:32technical disagreement in a way that 41:34doesn't hurt the relationship? 41:37>> I think the biggest thing is you have to 41:38earn it. Yeah. You just have to earn 41:40trust and it could be as simple as, you 41:43know, like what I did at the beginning, 41:45which is just disagreeing and committing 41:47and showing that I'm willing to do that 41:48and I'm willing to just execute if 41:50someone else thinks it's a good idea and 41:52I kind of look up to them. 41:54But also you have to kind of show that 41:56you have good technical judgment. Um but 41:59you can't really do that until you earn 42:00trust. So take the time to get that 42:02trust first. 42:03>> And then on the imposttor syndrome, um 42:05leading those engineers that were also 42:07very strong. Um do you have any advice 42:10for overcoming imposter syndrome? 42:12>> Yeah, just don't overthink it. You know, 42:14no one really knows what they're doing, 42:16you know, at any level. No one really 42:18knows. We're all just trying to figure 42:19it out. 42:20>> That's easier said than done. Was there 42:22like a an aha moment where you realize 42:25actually maybe I do got this or this 42:28isn't that big of a deal? You know, I 42:30don't I don't think so really. There 42:31wasn't a single moment. It just it kind 42:33of goes away over time. And I think at 42:35every at every level, it doesn't matter 42:37what level you're at, you should always 42:38feel a little bit of imposter syndrome 42:40cuz if you don't, then you're not 42:42pushing yourself hard enough. 42:43>> At this point in your career, you were 42:46like more and more of a tech lead and 42:48and therefore you were writing less and 42:50less code. Um and you mentioned that um 42:55you know at Meta especially there's 42:57cases where other functions are underst 42:59staffed and you view that as an 43:02opportunity for engineers so to be you 43:04know more product minded and maybe help 43:06out with uh you know the PM 43:08opportunities. I'm curious when would 43:10you say that you should go that 43:12direction as opposed to escalating and 43:14say hey we need more PM support um and 43:18you know trying to write more code 43:19instead. Yeah, you have to understand 43:22the trade-offs. I think this is the this 43:24is the thing that I think a lot of 43:26people don't really get when they push 43:27for stuff or, you know, I think a very 43:30common failure mode is an engineer will 43:31push for an idea and then get gets 43:33frustrated when no one else buys into it 43:35or wants to fund it. Um, or the 43:37organization just like doesn't listen or 43:38their leader doesn't listen. But what 43:41you have to do is understand the 43:42trade-offs. And whoever it is that 43:44you're trying to convince, think of it 43:45from their point of view. What do they 43:46care about? What are the projects 43:47they're working on? What is this 43:48trade-off against? if they do this 43:51thing, are are they going to see their 43:53work as a as a success? Um, 43:57so I I think I think that's really 43:59important and for some orgs at sometimes 44:02they might not have PMs cuz it might 44:04just not be a very sexy project and so 44:06it might be really hard to hire and 44:07maybe the leader is already feeling that 44:09pain. Maybe for some orgs uh they are 44:12trying to hire PMs but there's actually 44:13just much much more important things 44:16those PMs should go to. For other orgs, 44:19they might they might actually have like 44:20too many PMs. And so actually, if you 44:22ask, that's the right thing to do. Uh 44:25because they could just take a PM off a 44:26less important project and put on your 44:28project cuz it's more important. So I 44:30think it's really important to kind of 44:31be situationally aware, understand the 44:34context you're in, and understand how 44:36your decision makers think about it. at 44:39this point and this is kind of the end 44:40of that part one of your story. You 44:43credit a lot of your success to again 44:46the side quests and like having these 44:48side projects or running list of you 44:50called them 20% time ideas and I'm 44:54curious do you have any tips on how to 44:56find opportunities for engineers? Yeah, 44:58I think at some point there was probably 45:00I forget the exact numbers, but there 45:01was probably like 5000 engineers or 45:03something like this that were just like 45:04working on these like side quests that I 45:06like scoped out and like spun out of 45:07various points. And so pretty much like 45:09every week I I'll think of like some 45:11project, you know, just like on a run or 45:13something or maybe like while I'm coding 45:15I think of some idea, I'll just do some 45:17basic validation and then I'll just ping 45:19an engineer that I know and be like, 45:20"Yo, are you interested in this?" And 45:22then I'll connect them with a couple 45:23other engineers that might be 45:24interested. And this kind of added up 45:26very quickly. I think the for me one way 45:29that I really think about my work is how 45:32can I do less of it? And as an engineer, 45:35our superpower to do this is automation. 45:38The most tedious stuff you can automate. 45:40And this is something that's like really 45:41hard actually for other fields, but for 45:43us it's this incredible thing that we 45:45can do. And as it's something a lot of 45:47engineers don't really do for whatever 45:49reason, but we should all be doing it 45:51all the time. It's so important. It's 45:52leverage. It's like free leverage. And 45:55so a thing I often did was every time I 45:56did a code review, if I was commenting 45:59about a particular kind of issue, maybe 46:01like a stylistic issue or something, I 46:03literally had a spreadsheet where I 46:04would tally up that issue and I posted 46:08kind of the link to that poll request 46:10and then I would do this for every code 46:11review and then when I commented about 46:13the same kind of thing more than a few 46:14times, I would just write a lint rule 46:16for it to just automate that. Um, so 46:19this is kind of an example of leverage. 46:21So, and at some point I automated most 46:23of my code reviews cuz like I just had a 46:25you know like a flock of lint rolls that 46:27were just like doing all this work for 46:28me. And I think this is actually kind of 46:30similar because all these side quests it 46:32was improving prod infra and dev infra. 46:34And these are things that slowed me down 46:36in my day-to-day coding. And this is why 46:38like when I was doing west coding this 46:40was actually very dangerous because as 46:42an engineer you need to be anchored to 46:43reality. You need that intuition and if 46:45you're not in the code anymore then you 46:47lose it very quickly. It's it's a very 46:48dangerous place to be in. And so for me 46:50when I was in the code a lot there was 46:52all these really cool ideas that came 46:54out of it and it was leveraged not just 46:56for me but for the whole edge team 46:57because again of this principle that if 46:59you have a problem probably other people 47:01have it too. And you know I did like YC 47:04back in the day and in YC they teach you 47:06that first you build for yourself. You 47:08have to build like awesome stuff. You 47:10have to build stuff people love. But if 47:12you're trying to find a market to build 47:13for you start by building for yourself. 47:15And that's a pretty good indicator other 47:16people probably have that same problem. 47:18>> Yeah. There was a quote that you wrote 47:19that I thought was really good. You said 47:21better engineering is the easiest way to 47:23grow your network and gain influence as 47:25an engineer. So I could totally see um 47:28you had like your 47:31scope of influence was so much further 47:33than just the code you're writing 47:35because you're passing people all these 47:36great ideas and you know overseeing 47:39them. It's the the leverage is really uh 47:41insane. 47:42>> Absolutely. Yeah. And and it's also just 47:44like an example of being contextually uh 47:46was it situationally aware 47:48>> um because you know at meta at the time 47:50engineers were evaluated in the 47:52performance cycle uh we looked at 47:54project impact people do you remember 47:57the other 47:58>> direction 47:58>> direction 47:59>> and edge excellence 48:00>> and edge excellence yeah and the edge 48:02excellence is a thing that a lot of 48:03engineers struggled with and so you know 48:06I was you know one of the people that 48:08came along and was like hey if you want 48:09to do edge excellence here's a project 48:11and people are already incentivized to 48:13do it. So, they see it as an 48:14opportunity. And I think this is just 48:16like I don't know. I think it was a 48:19chance for me to kind of hone my skills 48:20about working with people where you 48:23never ever want to tell anyone what to 48:24do in any in any context. In a personal 48:27context, in a work context, everyone 48:28hates being told what to do. But if you 48:30understand what a person wants, then you 48:32can go to the right person with a red 48:34opportunity and they see it as an 48:35opportunity. Um, and this just always 48:37works better for everyone. When I think 48:39about these 20% time ideas, I mean, 48:42there's the there's the top of funnel, 48:44finding the ideas, and then there's 48:46actually, you know, executing on them, 48:48the, you know, getting someone to do it 48:50or whether it's yourself or someone 48:52else. Um, the thing I'm interested is 48:54the top of funnel, like how how do you 48:57source so many ideas um as an engineer 49:00for these side quests that are 49:02impactful? 49:03>> Just common sense. [laughter] 49:06I don't know. Maybe maybe spidey sense. 49:08I I don't know the right word sense like 49:09how so like what's a what's a concrete 49:11example? 49:12>> Yeah, like a really concrete example is 49:15um you know like I think lint rules are 49:17a good one. Maybe maybe another one is 49:19there were all these cases where we had 49:21sevs because uh Facebook groups were not 49:25being tested with very large sets of so 49:29um and so like for example and so kind 49:31of like this is kind of like a Facebook 49:33way of saying like you know rows in a 49:34database. So you could imagine like a 49:36Facebook group with like 10 million 49:37members. Like no one's ever tested this. 49:39There's no like unit test for this. You 49:40only see it in production. And when I 49:43looked across the org, I started seeing 49:45similar cases of this. There's, for 49:46example, like if you have a profile with 49:48like 20 million followers, a lot of 49:49stuff breaks. But obviously like no one 49:51tests this in an automated way just cuz 49:54it's kind of annoying to write a unit 49:55test with this much data. And so there 49:57there's a bunch of instances of this. 49:59And then um I pitched an engineer to 50:01build a way to write unit tests for 50:04large data sets. So you know like a 50:06really big object like a group with a 50:08lot of members, a profile with a lot of 50:09followers, an event with a lot of 50:10attendees. And uh I think this infra 50:13still exists. Um and it's you know it 50:16prevents a lot of issues and this is 50:17something where you can scope it and 50:19then he brought in a bunch of other 50:20engineers to do the work and and help 50:22him out with it. So I guess just think 50:24about the problems that you actually hit 50:25day-to-day. 50:26>> Got it. Okay. So think about the 50:28problems and if you're experiencing that 50:30problem repeatedly then it's time for 50:33automation and that's like a great uh 50:35better engineering project. 50:36>> Yeah. Exactly. Exactly. Like if you hit 50:38the same problem like two or three times 50:40you should probably kind of look around 50:42see if other people are hitting that 50:43project hitting hitting that problem 50:44too. 50:45>> The last leg of your career at Meta um 50:47this is where you got the E8 promo. I 50:49know that um you moved orgs. So, you did 50:53all of your growth in Facebook groups 50:55and then you moved to Instagram. I'm 50:57curious, uh, what's the story behind you 51:00moving orgs to Instagram? At the time, 51:04um, I was dating, my wife and I were 51:06still, uh, dating. And, um, she was 51:10living in Berkeley, I was living in SF. 51:12And at some point, she's like, I found 51:15my dream job. And I was like, sweet, 51:17awesome. And then she was like, we're 51:18gonna have to move. And I was like, 51:20okay, great. we we've been dating like 3 51:21months at the time and uh we were kind 51:25of deciding like should we like keep 51:26dating and um and so she was like yeah 51:29we we would have to we'd have to move if 51:31you want to like keep dating and I was 51:32like yeah okay I do let's let's do it 51:35and then so the job ended up being in 51:36like rural Japan like it was sort of 51:38like middle of nowhere and I was trying 51:40to figure out like how do I do it 51:42because I really like the work that I 51:43was doing 51:44>> and so first I talked to uh like 51:46Facebook groups leadership and tried to 51:48set up like a Japan 51:50out there for for Facebook groups. That 51:52didn't really work for, you know, a 51:53bunch of kind of organizational rules. 51:55Um, then I tried to do this with the VR 51:58org and it was actually working, but 52:00then the person that was sponsoring it 52:02left to go to, I think like YouTube or 52:03something. Uh, and then at the time Will 52:05Bailey reached out and he was in the 52:08Instagram Tokyo office. He was part of 52:09this like landing team for for Instagram 52:12and he was like, "Hey, I kind of want to 52:14grow this office. Do you want to be part 52:15of that?" And I was like, "Yeah, let's 52:17let's do it." And I didn't know 52:18anything. I I didn't even have Instagram 52:20installed at the time. I'd never used it 52:21in my life. And um I so I I said yes and 52:26then I immediately downloaded Instagram 52:27and then like I moved like I think like 52:29the next week or something. Um so pretty 52:32much or or actually you know I think it 52:34was like a few weeks that that I had in 52:35the US. 52:37But I I moved out pretty quickly. Um and 52:40I actually really fell in love with the 52:43Instagram culture. Um it was very 52:46different than Facebook culture. big 52:48emphasis on building awesome products, 52:50on on shipping stuff that people don't 52:52use, on thinking about things not just 52:55from a data point of view, but also from 52:57a human point of view and an experience 52:59point of view. And you can see this in 53:01the app and in the craft that goes into 53:03it. It's just completely different 53:04engineering and product and design 53:07cultures between the two companies. Um, 53:10so I learned so much being being on that 53:12team and that that was such a fun 53:13journey. 53:14>> You mentioned the the un shipped part. 53:16Um, what what is that? 53:19>> Unchipping is the idea that if you just 53:21add features to an app, it's cool for 53:24some small percent of users, but it's 53:26actually bad for most users that don't 53:28use the feature. And so, you can think 53:30of an app where you only add features to 53:32it. And over time, the features 53:34accumulate and they add up. And you 53:36know, if every feature is used by like 53:3810% of people, the average user sees a 53:40bunch of features that they don't use. 53:41And so, it seems cluttered and 53:43confusing. And when they open the app, 53:44they don't know what to do. And uh you 53:47know like with software fundamentally 53:48the screen is a limited size. That's the 53:50limited real estate. Uh there it's a 53:52limited resource that all the different 53:54features are competing for. And so by 53:56adding a feature that's taking the 53:57opportunity away from you know a 54:01different feature the person could have 54:02used. Um and so unshipping is the idea 54:05that you have to meet some sort of usage 54:07bar and if a feature doesn't meet that 54:09bar then we just delete the feature. And 54:11a small percent of users are going to be 54:13pissed, but it's actually great for the 54:15majority of users. And on average, it's 54:18really great for everyone 54:20>> at this point in your career. Um, I 54:22mean, you moved, you didn't just move 54:24orgs, you moved across the world to work 54:27at Instagram. And I think when you're 54:30such a senior tech lead with a lot of 54:31credibility in your existing org, it's 54:34much easier to get things done or at 54:35least influence others cuz they say, 54:37"Oh, I know Boris and I know his past 54:40work." But I'm curious, how did you 54:42build up credibility uh at Instagram uh 54:46when you were so far away from everyone 54:48else? 54:50>> I think a lot of the credit early on 54:52this goes to like Nam who was Nam Guian, 54:55he was the he's still the VP of Avenge 54:57at Instagram and and Jeff Hang um who 55:00was my director at the time but he now 55:02he's a VP. Um and you know will I think 55:06there was a lot of connections made by 55:08these people. So you know for example 55:10Nam was like hey you really like doing 55:12you know working on code quality and 55:13like tech debt reduction you know which 55:15we call you know better engineering at 55:17meta and he kind of connected me with 55:20the people working on it and this was 55:21like Lucas Camera and um Gabe and a 55:24bunch of this bunch of these other folks 55:26that that were working on this stuff. Uh 55:28so so those connections were really 55:30useful. Um, and then I think a lot of it 55:33was I just had to earn the trust again. 55:36And honestly, this is a healthy thing to 55:38do. And I think this is one of the 55:40really awesome things about meta 55:41engineering culture that there are not 55:44titles. And so you kind of have to 55:45constantly reearn your trust. And you 55:49know, even if I was a great engineer in 55:50the past, I may not have been a great 55:51engineer at Instagram. And if I wasn't, 55:53then I don't deserve influence. I don't 55:55deserve to have a really loud voice that 55:58people listen to. So I had to earn it 56:00along with everyone else. And so my 56:03first few weeks I spent a lot of time 56:04like meeting people, mapping out the 56:06org, mapping out goals, writing a lot of 56:08code so I can get to know the codebase. 56:10But then in Japan it was totally 56:11different because you know like 400 p.m. 56:14Tokyo time was like or it was like 9:00 56:18a.m. Tokyo time was like 7:00 p.m. New 56:19York time. There was just like no time 56:21zone over. 56:23>> It was rough. Yeah. Um, but it was also 56:25great because I think in the few years 56:27before I was doing so many meetings and 56:29docs and all this stuff, I just wasn't 56:30coding. Um, so I just started to feel 56:33pretty unhappy cuz like as an engineer 56:35we code. That's that's what we do. Like 56:37that's that's the reason we pick this 56:38job is you like for me like when I write 56:42code I have an emotional relationship 56:43with a code and it's something that I 56:45think about when I'm really deep in a 56:46problem. I dream about it. So it's just 56:48so important for me to code. And when I 56:51wasn't doing this for years, it was sort 56:53of rough. And I think I was starting to 56:55burn out a bit. And so, actually, it was 56:57it was a gift to be in this time zone 56:59where I literally couldn't do meetings 57:00cuz people weren't awake or didn't want 57:02to like, you know, do 9:00 p.m. meetings 57:04just to talk. Um, so I didn't do any 57:06more one-on- ones. And this is actually 57:08still something I I don't do. So, I 57:10still don't do any standing one-on- 57:11ones. And I just could spend a lot of 57:13time coding. And what I realized is I 57:16was one of a few engineers at Instagram 57:18at the time that was coding this much. 57:20Um, you know, people code, but people 57:23don't code that much because there's all 57:25this stuff that fills up your time 57:27there. There's meetings and, you know, 57:29docs and all these other obligations. 57:31And 57:33I was able to do a lot of stuff that I 57:35think everyone else wanted to do, but 57:38just didn't have time. Um, and this was 57:40kind of a superpower in in that org. And 57:43pretty early on, Nam connected me with 57:46uh with Joel Pamer, who's still a good 57:47friend and and and mentor and um he's 57:50he's at Google now. And we just started 57:53talking about like you know at the time 57:54the codebase was written in Python and 57:57uh there it was sort of rough for a lot 58:00of different reasons and really the 58:02codebase should have been moved over to 58:03hack which is the main Facebook monolith 58:06you know and this is where all the 58:07language support is. There's so much 58:08infra HHVM is just this absolutely 58:11phenomenal web serving stack. There's 58:13nothing else like it in terms of like 58:14efficiency. If you're using GraphQL, you 58:16absolutely have to use it because it's 58:18just so optimized for this stuff. And 58:20Instagram just wasn't using any of this. 58:22And engineering was suffering. Like in 58:23the really early days, you know, like 58:25when like Mikey was at Instagram, really 58:27basic decisions 58:29were the the basic principle for 58:31decisions was do the simple thing that 58:33works. And this worked really well. But 58:36then at some point it stopped working 58:37once you get to like a thousand 58:39engineers, 2,000 engineers working the 58:40codebase. and you know many many years 58:42of techad and products built on top of 58:44each other you kind of have to do you 58:47have you have to make slightly different 58:48decisions than you would have made at 58:49the start and so even if Python was 58:51absolutely the right decision at the 58:53beginning it was not the right decision 58:55by the time I was there and this was 58:56painfully obvious as an engineer and I 58:58think a lot of other people saw this but 59:00what stopped them was just the amount of 59:01work it would have taken to to move the 59:03stuff over and so I just started like 59:05scoping this and kind of figuring out 59:07what it would take and so I started by 59:09finding the people that would disagree 59:10the 59:11And there's a bunch of these like info 59:12old-timers that just thought this was 59:14like a terrible idea and would never 59:16work. And so I went to talk to them 59:18first at like food in New York and you 59:19know like we got a bunch of beer and 59:21just got to know them as people before 59:22we even talked about the technical 59:23problem. You have to build trust. So I 59:26had to kind of get to know them as 59:27people and this was so valuable and this 59:29is still a lot of my friends today. Um, 59:32and after building this trust, I also 59:35learned there was a bunch of other 59:37people that actually did want to do this 59:38and were kind of afraid to say it. And 59:40so these people came out of the woodwork 59:42too. And eventually we started scoping 59:44this and this project kind of spun out. 59:46And it it's actually still going today. 59:47And there's, you know, many engineers 59:48working on it. But it but it was it's 59:51funny because at Facebook I think this 59:52kind of problem rarely happens because 59:54the org is so engineering driven. At 59:56Instagram there were many problems of 59:57the shape because the org is very 59:59product driven. So there isn't just a 1:00:00lot of time for those engineering driven 1:00:02initiatives. this project at some point 1:00:05I mean you got it off off the ground 1:00:07kind of this bottoms up initiative and 1:00:09then at some point it became high pry 1:00:11enough where it needed that uh in-person 1:00:14support of someone that wasn't in Japan 1:00:16and I understand that Jake Bolum is 1:00:19someone that you helped onto the project 1:00:21and he kind of took more of a a lead 1:00:23role um but locationbased and you know 1:00:27close to everyone else so he can help 1:00:28shepherd it along. Um, I'm curious your 1:00:31thoughts on 1:00:33that point of delegation. Like when do 1:00:35you decide when to delegate something so 1:00:37big and when do you decide, oh, I I need 1:00:40to still be around and how do you 1:00:42navigate that trade-off? Jake is 1:00:43amazing. We're we're, you know, we're 1:00:45friends. Every time I go to Seattle, we 1:00:46we hang out and he's just one of the 1:00:48best engineers I know. So, it was 1:00:50obvious that he would be a good owner 1:00:53for this. The same rules of delegation 1:00:55apply as always. So, you know, you 1:00:57delegate. You never delegate the thing 1:00:59you don't want to do. That's kind of the 1:01:00most important rule. You always delegate 1:01:02the thing you do want to do and that you 1:01:04know well because then you can monitor 1:01:05the progress and make sure it's going 1:01:06well. Um and there's this really great 1:01:09book uh high output management by uh 1:01:11Andy Grove. He was a you know old Intel 1:01:13CEO and it's just like the most boring 1:01:15sounding book ever, but it's just like 1:01:17the best. And this is one of the pieces 1:01:19of advice is delegate the thing that you 1:01:21like to do so then you can monitor 1:01:23progress. And so yeah, it's it's kind of 1:01:25the same thing. You kind of delegate a 1:01:26little bit. you check in. The more trust 1:01:28you have, the less you have to check in. 1:01:30And with Jake, he's so good technically 1:01:32and so proactive. There's just very 1:01:34little I had to do. It it was very much 1:01:35on track from the start. And so I think 1:01:37this coupled with um some other work uh 1:01:41large migration to kind of GraphQL or 1:01:43modernize some of Instagram's data model 1:01:46ended up getting you uh promoted to this 1:01:48principal level before you left Meta. 1:01:51Um, what was the story behind the 1:01:53promotion or anything that you might 1:01:55share 1:01:56>> that promotion? I think Will in in a 1:01:58one-on-one that I had with Will, my 1:02:00manager, he was like, "Hey, like I think 1:02:02you should like we should put you up for 1:02:04IC8." And I was like, "Cool." 1:02:06>> And that was pretty much it. I I hadn't 1:02:08really thought about it. Yeah, it's not 1:02:09something I really asked for. 1:02:11>> Um, I think Will was just like does a 1:02:13great job of recognizing people and kind 1:02:15of advocating for his team and he felt 1:02:17that I was ready and yeah, that was 1:02:19that. at any point in your journey 1:02:20because it sounds like you were impact 1:02:24and impact only and growing and your 1:02:26leverage and your credibility everything 1:02:28was growing and then the promos were 1:02:30this lagging thing where they just oh 1:02:32they kept happening as a byproduct. Um 1:02:35I'm curious though to structure your 1:02:38thinking about how to um get more 1:02:41leverage or kind of have more impact. 1:02:44Did did you ever think about the levels 1:02:46or would you say that it's not good to 1:02:49think about like oh what's the next 1:02:51level more think in terms of just I 1:02:54don't know leverage or impact or 1:02:55something else you have to think about 1:02:57what levels are for right levels exist 1:03:00so that the company can communicate to 1:03:02an engineer what it is they expect the 1:03:04engineer to do it also exists so that 1:03:07there's some accountability so for 1:03:09example in performance reviews the 1:03:11engineer at a particular level you can 1:03:12compare them to another engineer at that 1:03:14same level and also sometimes on the 1:03:15finance side different engineers or you 1:03:17know the pit costs a different amount. 1:03:19So you can kind of think about what kind 1:03:20of portfolio you want. So you know 1:03:23levels it kind of exists for company 1:03:24reasons and not for engineer reasons. 1:03:27And for me it's just never the way that 1:03:29I thought about any of this. Like the 1:03:31thing that I like to do is work on 1:03:33interesting projects. I like to figure 1:03:35out the problems and solve them. I like 1:03:36to um just make the things that people 1:03:40use delightful like the products that 1:03:41they use delightful. this this is what 1:03:43really motivates me. So for me, this was 1:03:46just never really the way I thought 1:03:47about it. And I I don't know like my my 1:03:49first week at Facebook, I was like an 1:03:50IC4 and I was I had like all these ideas 1:03:53for stuff that we should build. And so I 1:03:54just started writing like product 1:03:56briefs. And then I remember I went to 1:03:57like the VP of like connectivity and 1:03:59like pitched him some idea and he just 1:04:01didn't understand it at all and thought 1:04:02it was terrible. And then like that 1:04:04didn't work. And then I had another idea 1:04:06for messenger and like I went and like 1:04:07pitched that and like they also like 1:04:09didn't do it. But then they actually did 1:04:11end up building it a couple years later, 1:04:12this particular idea. So yeah, for me 1:04:15it's just like, you know, what are what 1:04:17are the cool ideas and like how can I 1:04:18help and who else is interested in this 1:04:20stuff and how can we build build cool 1:04:22stuff? You left Meta to go to Anthropic 1:04:25and I'm very curious what was your 1:04:27thinking about going to anthropic. 1:04:30>> I remember using chatt for the first 1:04:32time when it came out. This was uh you 1:04:34know this was like year years ago and I 1:04:36remember I was I was in Japan. I was the 1:04:39only engineer in my town. I was the only 1:04:40person that spoke English in my town and 1:04:43there's just no one that I could talk to 1:04:44about tech stuff and I remember every 1:04:46every morning I read Hacker News and I I 1:04:49remember using Chat GPT and I was just 1:04:53like blown away by by the product and 1:04:56just this this feeling that it gave me 1:04:57of just nowadays we take it for granted 1:05:00but you know LMS are just magic. It's 1:05:02just this absolutely incredible 1:05:04technology and I think now my view of it 1:05:06has shifted. It's like to me it's LMS 1:05:09are this kind of alien life form that we 1:05:11get to nurture and we get to bring into 1:05:13existence. It's not just a technology. 1:05:16And I'm also a really big reader. I I 1:05:18read a lot of sci-fi. That's kind of my 1:05:20my big genre. Um and so at the time I 1:05:23was like, "Oh my god, I have to work on 1:05:24this stuff." And you know, what are what 1:05:26are the labs that are that are working 1:05:27on it? So I went to talk to friends 1:05:29working, you know, at various labs. And 1:05:31I feel like when I came to Anthropic, I 1:05:33remember my first lunch. This was with 1:05:34Ben man, who's uh one of the founders 1:05:36here. And we were sitting at lunch with 1:05:38like him and a bunch of other people. 1:05:40And I mentioned this like weird like 1:05:42sci-fi book that I like. Uh it was I 1:05:44think it was like by Greg Egan or 1:05:46something. It's like hard sci-fi author. 1:05:47And I've literally never met anyone 1:05:49that's read this book. And at the table 1:05:52I kind of gave an anecdote from it. And 1:05:54everyone around the table was like, "Oh 1:05:56yeah, yeah, that book was good, but like 1:05:57what about this other book?" And so it 1:05:59was just like this group of people of 1:06:00these like intense sci-fi nerds, these 1:06:03people that think so deeply about these 1:06:04same problems I care about. And I I 1:06:08think the other part for me was how 1:06:10you're reading a lot of sci-fi, you kind 1:06:12of get a sense of, you know, in a very 1:06:14speculative way how this thing can play 1:06:16out. AI is just so transformative to 1:06:18society. I think it's hitting 1:06:20engineering first and it's going to hit 1:06:22every part of society. It's going to 1:06:24affect everything. And we're seeing the 1:06:26very first waves of this right now. 1:06:28And I think I just, you know, like I I 1:06:30read enough that I kind of knew the bad 1:06:32ways that this can play out and there 1:06:33can be so many ways this thing goes 1:06:35wrong. And so for me, anthropic was just 1:06:38the obvious choice cuz I wanted to be at 1:06:40a place where in the tiniest way I could 1:06:43make sure this thing goes well, which is 1:06:45as an engineer, this is all I can do. 1:06:47Um, and it it's funny. I think like 1:06:50before I joined, I sort of took safety 1:06:51seriously. Like you know, it's like a 1:06:53thing that can happen. And you know at 1:06:54Meta it was always seen as this kind of 1:06:56tax where integrity teams and you know 1:06:58and so on they kind of like they get you 1:07:00to do stuff but it's not really the 1:07:02thing anyone's really excited to do 1:07:03because it's not the product. And so 1:07:05this was kind of my view of safety 1:07:06before but at the same time I kind of 1:07:08speculatively knew that it could be kind 1:07:10of a very different thing. And now being 1:07:12at Anthropic, 1:07:14I know it's completely different and I 1:07:16see every model that comes out and kind 1:07:18of the new risks that come with it and 1:07:20how as a company we put our money where 1:07:23our mouth is, you know, in terms of like 1:07:26how much compute goes to safety research 1:07:28and alignment research in terms of like 1:07:30how many people work on this. We've held 1:07:32up model releases in the past because we 1:07:34did not know that they were safe and so 1:07:36we had to make sure that they were safe 1:07:37before. Um and I think also like with 1:07:40Opus 4 the risks just went way up like 1:07:42if the model can design you know like 1:07:44bioviruses and can do these things that 1:07:46are like really really dangerous and 1:07:49it's not just like you know like the 1:07:51baseline risk is something like election 1:07:53manipulation. This is something that you 1:07:54know like at was a really big deal. This 1:07:56is sort of a risk for kind of a 1:07:58low-level model. As the models get more 1:08:00dangerous, the risks get higher and very 1:08:02quickly you get into this territory 1:08:04where people can use the model to build 1:08:06things that are actually dangerous for 1:08:08humanity. Like not not just like you 1:08:10know like uh you know politics in a 1:08:12country but like literally the existence 1:08:14of humanity. Um and these this is not 1:08:16sci-fi. This is a real risk today that 1:08:18we actively have to fight. Um and so for 1:08:21me just like getting to be a part of 1:08:22this and getting to contribute a little 1:08:24bit this this is just what put me over. 1:08:27What about when you joined? When you 1:08:29think about the engineering cultures 1:08:31that you came from compared to 1:08:33enthropic, were there any really jarring 1:08:35differences? 1:08:37Yeah, I would say the two things. One is 1:08:40still being a startup, there's a lot of 1:08:41common sense. Um, and it's funny. I 1:08:44think this is something every big 1:08:45company loses and it's sort of this 1:08:47thing you have to fight because over 1:08:50time the decision makers get more 1:08:51distant from the impact of their 1:08:53decision, whether it's the product or 1:08:55the people or whatever. you have to come 1:08:57up with all sorts of processes to bring 1:08:58them closer and to improve the quality 1:09:00of decision-m but still being at a 1:09:02startup uh everyone just has common 1:09:04sense and generally just does the right 1:09:06thing. So I don't have to spend a lot of 1:09:08time convincing people to do stuff. If 1:09:10we should just do something it's it's 1:09:11obvious and everyone just does it. I 1:09:14think the second thing is for me 1:09:16personally something that I learned over 1:09:18time motivates me the most is mission. 1:09:20It's so important and it's the thing 1:09:23that keeps me going to work every day 1:09:25excited to go to work every day. It's 1:09:27the thing that makes me like code on the 1:09:29weekends because I want to do it, not 1:09:30because there's a deadline. Um, and so I 1:09:33I felt this a lot actually in Facebook 1:09:34groups. It it was very missionoriented. 1:09:36And um, for a long time, Jen Dolski was 1:09:39the VP and she she used to be the CEO of 1:09:40change.org. And so she ran all of 1:09:43Facebook groups like a nonprofit. She 1:09:44had like a theory of change and this 1:09:46theory about how you want to connect 1:09:47people to like-minded people to form 1:09:49communities and it was just so 1:09:51motivating to work on that. And I feel 1:09:53like at Instagram, you know, maybe 1:09:54because of the geographic distance or 1:09:56maybe something else, I just never quite 1:09:58felt that same mission. Um, but I feel 1:10:00like at anthropic I feel that so 1:10:01strongly and um that's probably the most 1:10:04exciting thing for me. I know that you 1:10:07know you're you're credited as the 1:10:08creator of Cloud Code and I think you've 1:10:11you've told that story in many places um 1:10:14but I'm kind of curious uh I was talking 1:10:17with a friend about what the environment 1:10:19was like when cloud code came and there 1:10:22were actually a lot of competing 1:10:24existing tools that hooked into the 1:10:26model and I'm curious what is it that 1:10:28was different about cloud code that kind 1:10:32of won out and just caught like wildfire 1:10:34internally 1:10:36At the time coding looked very 1:10:37different. If you thought about AI 1:10:39coding, people thought about like 1:10:40autocomplete. That was essentially it. I 1:10:42think there was like some very early 1:10:44agents, but it was kind of a secondary 1:10:46thing next to the auto autocomplete. And 1:10:49often times it was used for Q&A. Um it 1:10:52wasn't really used for coding. And so 1:10:54when people thought about AI for coding, 1:10:56it was just a completely different 1:10:57product that you imagined. You know, it 1:10:58was like tab to autocomplete. 1:11:01And I thought that's what it was. And I 1:11:04thought that was kind of the state of 1:11:06it. And then um Ben who um was my 1:11:10manager at the time uh he pushed me to 1:11:13think a little bit bigger and he I think 1:11:16really internalized cuz you know he was 1:11:17there from the beginning of of Enthropic 1:11:19and you know he was like at at other 1:11:21labs before and so I think he really 1:11:23understood the scaling laws kind of 1:11:25internally about how quickly the models 1:11:26are improving and so he actually pushed 1:11:28me really hard to be like don't build 1:11:30for the model of today build for the 1:11:32model 6 months from now. Um, and so 1:11:35honestly for a long time quad code was 1:11:37not a great product and even when it was 1:11:39used internally I used it for maybe like 1:11:4110% of my code or something like this. 1:11:43You know use it sometimes but it really 1:11:44just can't do most things cuz the model 1:11:46is not capable enough and then at some 1:11:48point we released uh set and opus 4 and 1:11:51I think this was maybe March of this 1:11:53year and the product just worked and we 1:11:56saw this in kind of the usage data and I 1:11:58saw this in my own coding. I started to 1:12:01be able to use it for probably like half 1:12:02of half of my code. And this was totally 1:12:05borne out because this was actually like 1:12:07literally 6 months after starting the 1:12:08project. This was the timeline. And you 1:12:11know at this point most of quad code is 1:12:13written using quad code. I think it's 1:12:14like 80 or 90%. If you look at teams at 1:12:17anthropic, there's some teams that you 1:12:19know like 90% of their code is written 1:12:21using quad code. Um this is not just our 1:12:23team. And if you look at the impact on 1:12:26productivity even though anthropic has 1:12:28tripled I think since the start of the 1:12:30year uh productivity per engineer 1:12:32measured in you know like poor cost we 1:12:34were talking about this before has grown 1:12:35like almost 70% per engineer because of 1:12:38quad code and so you know like as a 1:12:42product person I usually think a step 1:12:44ahead and I think being at a lab you 1:12:46have to actually think kind of 1:12:47differently you have to think a step 1:12:48ahead um not two steps ahead um but also 1:12:51you have to be really aware of the model 1:12:53and kind of the exponential potential 1:12:54that we're on. 1:12:55>> Did you see the recent interview with 1:12:57Karpathy? 1:12:58>> I haven't had a chance to watch it yet. 1:12:59>> Okay. So, one thing that he said in the 1:13:02podcast was kind of like a you know 1:13:05pushing back a little bit because 1:13:08um basically vibe coding although it has 1:13:11a lot of miraculous results because of 1:13:13what it can generate there's also a lot 1:13:15of like he said like slop or you know 1:13:18there's like some drawbacks. So I'm 1:13:20curious like how do you think about that 1:13:22when the model produces um a lot of code 1:13:26but maybe it's not exactly how you'd 1:13:28like it or maybe the end result has some 1:13:30non-obvious problems with it. 1:13:34AI coding is a tool like every other 1:13:36tool that we use and you have to learn 1:13:38how to use it. So I think one of the 1:13:40most you know Karpathy obviously like he 1:13:42knows how to code. I think a lot of 1:13:44people that are new to this kind of 1:13:46tool, they tend to just ask it to do 1:13:48stuff that's a little bit too big or 1:13:50they hold a different bar for the 1:13:51model's code versus their own code. Um, 1:13:53and so something that I do for the team 1:13:55is for the quad code team, we have the 1:13:57same exact bar regardless of whether the 1:13:59code was written by the model or by a 1:14:01human. And so if the code sucks, we're 1:14:03not going to merge it. It's the same 1:14:04exact bar and you just ask the model to 1:14:06improve the code and and make it better. 1:14:08I think there's also kind of different 1:14:10ways of wielding these tools. Sometimes 1:14:12you want to vibe code. Uh, and this is 1:14:14really important for throwaway code and 1:14:15prototypes, code that's not in the 1:14:17critical path. And you know, I do this 1:14:19all the time, but it's it's definitely 1:14:21not the thing you want to do all the 1:14:22time. Um, because you want maintainable 1:14:24code sometimes. Um, you want to be very 1:14:26thoughtful about every line sometimes. 1:14:28So the way the the problem depending on 1:14:31the problem, you want a a different 1:14:33approach. And so, you know, there's like 1:14:35a set of approaches that I that I'll 1:14:37use. Like sometimes I'll vibe code 1:14:38stuff. Um, this is actually quite rare. 1:14:40It's actually mostly for like prototypes 1:14:42and things like that that are throwaway 1:14:43code. Usually I pair with a model to 1:14:46make to to write code. And so first we 1:14:48align on a plan, you know. So this is 1:14:50like shift tab in quad code to get into 1:14:52plan mode. And first the the model will 1:14:55make a plan, then I'll see the code, and 1:14:56then I I might ask it to improve the 1:14:58code or clean it up or so on. But it's a 1:14:59very involved process. I'm pairing with 1:15:01the model. We're working together to to 1:15:03write this code. And then sometimes I'll 1:15:06still write the code by hand. So you 1:15:08know there's some parts of our core 1:15:09query loop where I have very strong 1:15:11opinions about like you know things like 1:15:13the names of parameters or kind of like 1:15:15on which particular line of code is um 1:15:17some some code is and so for this I'll 1:15:19still write it by hand. I think the the 1:15:21thing though is 1:15:24the models I think are still overall not 1:15:26great at coding. I think there's still 1:15:29so much room to improve and this is the 1:15:32worst it's ever going to be. But it's 1:15:34just insane to me to think a year back 1:15:36where the state of AI coding was, you 1:15:38know, like type ahead and it's just a 1:15:41completely different world. And you sort 1:15:43of think about like where this is headed 1:15:45and kind of what's about to happen and 1:15:47what this looks like over the next few 1:15:48months and few years. And yeah, I think 1:15:51for me what keeps me really excited 1:15:53about it is just having the context 1:15:55about this trajectory that we're on. 1:15:57When people hear cloud code, they think 1:15:58about coding, but um there's many use 1:16:02cases outside of just software 1:16:04engineering like querying data for like 1:16:07data scientist. What are your thoughts 1:16:08when you think of cloud code for 1:16:10everything? It was the craziest thing 1:16:12when I walked in. I remember walking 1:16:14into the office like maybe it was like 6 1:16:16months ago or something and our data 1:16:18scientist Brandon had cloud code up on 1:16:20his computer. He like sits next to us 1:16:22and I'm like, "Dude, what are you doing? 1:16:23Are you like trying it out?" And he's 1:16:24like, "No, no, I'm like it's like doing 1:16:25my work." I was like, "What?" So he like 1:16:27he figured out how to like use a 1:16:28terminal and like how to install Node.js 1:16:30and then he installed quad code and then 1:16:32it was writing a bunch of SQL and like 1:16:33doing analysis for him. And now when I 1:16:36walked by kind of the the the data 1:16:38scientists that sit next to us, every 1:16:40person has like a bunch of cloud codes 1:16:42up at the same time. It's not just one 1:16:44anymore. And they're doing all sorts of 1:16:46stuff. They're like writing SQL and you 1:16:49know kind of crunching data. Um they're 1:16:51writing DBT pipelines and they're 1:16:53writing code. Um, so I think there's all 1:16:55these applications outside of coding and 1:16:58it's just so cool to see how people are 1:17:02are using this for all sorts of stuff 1:17:04and there's even like totally 1:17:06nontechnical users like half of our 1:17:08sales team at Anthropic uses quad code 1:17:10to do their work like they they can like 1:17:12connect it to Salesforce and they can 1:17:14connect it to different data sources and 1:17:16you know like they can do their work 1:17:17this way. So it it's just so cool to see 1:17:19this is not how we designed it. This is 1:17:21not the intent. When I hear cloud code, 1:17:24I also think of Codeex, one of the 1:17:26biggest competitors. And I'm curious, 1:17:29what are your thoughts on the 1:17:30competition with Codeex and OpenAI? What 1:17:33does Cloud Code do better? Um, and also 1:17:36I'm curious like the stickiness of these 1:17:38AI products like, you know, what keeps 1:17:40people in cloud code versus um Codeex 1:17:43for instance. 1:17:44>> You know, I'm not totally sure. I don't 1:17:46personally, I don't really use the other 1:17:47products. [laughter] 1:17:48>> Okay. Okay. For me, you know, like the 1:17:51thing I tell the team is like it's so 1:17:53easy to get sidetracked by, you know, 1:17:55looking at competitors. And I think this 1:17:57is something that I saw that big 1:17:59companies fall into this failure mode a 1:18:00lot where 1:18:02because there's so many competitors and 1:18:04it's so easy to see the thing you could 1:18:06build by just, you know, copying it. 1:18:09It's a little bit harder to come up with 1:18:11novel ideas and things that solve the 1:18:15user need better. And so the thing that 1:18:17I try really hard to do and the thing 1:18:19that I nudge our team to do is don't get 1:18:22sidetracked by all these other you know 1:18:24products there's always going to be a 1:18:25lot and the more there are the bigger a 1:18:28sign of success that is for us and the 1:18:31thing we stay laser focused on is 1:18:33solving our problems and solving 1:18:34anthropic researchers problems and 1:18:36solving our users problems. Coming to 1:18:38the end of the conversation I just want 1:18:41to ask a few career reflections. Um, one 1:18:43thing I was curious about when I was 1:18:45looking is that um, you didn't have a CS 1:18:47degree or a computer science degree and 1:18:50you've, yeah, you've become such a 1:18:52strong software engineer and so I was 1:18:54curious is there any part in your career 1:18:57where you know it might have held you 1:18:59back in any way or do you think it's not 1:19:01necessary or relevant? 1:19:03>> Yeah, I studied economics and I actually 1:19:05dropped out to to startups. 1:19:08Um, you know, for me, anything that you 1:19:11do, you learn on the job and I think 1:19:14programming is just such a practical 1:19:16skill. I couldn't imagine, you know, the 1:19:18kinds of things you learn in school. And 1:19:20like, you know, if you're if you're like 1:19:21in a data structures class or something, 1:19:23you're like running this, but you 1:19:24haven't like built a product, how is it 1:19:26relevant at all? Um, so I don't know. I 1:19:30think my recommendation would be to 1:19:32people like learn coding practically. 1:19:34It's a very practical skill. And then if 1:19:36you want to go back and learn the theory 1:19:37after, go do that. Um, but personally, 1:19:39I've never felt like it it's held me 1:19:40back at all. 1:19:41>> What about productivity tips? So, I saw 1:19:44that you said you work roughly 9 to 6 1:19:47every day and you only type with two 1:19:50fingers, [laughter] but your output is 1:19:53ridiculous. I mean, you have all these 1:19:54side projects and of course, your main 1:19:56stuff's no joke. So, what's your top 1:19:59productivity tips or how do you maintain 1:20:00that? Yeah, I think I think nowadays my 1:20:02tips are very different than what I 1:20:04would have said a couple years ago. 1:20:05Nowadays, the tip is just learn how to 1:20:07use quad code and learn how to run a 1:20:09bunch of quad codes to do stuff like um 1:20:11you know we launched plugins like a 1:20:13couple weeks ago and Daisy who's one of 1:20:16the the the engineer that built it she 1:20:18just had her quads like set up a sauna 1:20:20board make a tasks and then she had a 1:20:22swarm of like 20 clouds just like build 1:20:24plugins over the weekend and uh you know 1:20:26she ran it in like a docker container 1:20:28dangerous mode and it was done after a 1:20:30couple days. Um and this is the kind of 1:20:33thing that is the future of engineering. 1:20:35Um, and I think nowadays when we talk 1:20:37about tips for productivity, it's learn 1:20:40how to use claude in order to automate 1:20:42toil and also how to use a bunch of 1:20:45quads together to to do uh work. So 1:20:48you're kind of orchestrating instead of 1:20:50manually writing code. Um, I think years 1:20:54ago my tips would have been really 1:20:55different. It would have been a lot more 1:20:56prosaic than that, you know, like about 1:20:58like blocking off time and and things 1:21:00like this. 1:21:02>> Yeah, it's it's interesting with cloud 1:21:03code. I wonder what that does for you 1:21:05know the famous um like maker schedule 1:21:08versus like manager schedule. It feels 1:21:11like what you just said is that software 1:21:13engineers becoming more like a manager 1:21:16style of working where you got a fleet 1:21:19of these cloud codes and you're not you 1:21:21don't need deep focus to move forward. 1:21:24You need context switching across like 1:21:2620 different things. Um so do you no 1:21:29longer have big focus blocks for coding 1:21:31or what are your thoughts on that? not 1:21:33as much. I tend to code on the weekends 1:21:36also. Um so I love the quiet time. Um 1:21:39but yeah, otherwise I'll kind of start 1:21:41every morning I open Quad Code and um 1:21:43you know like the mobile app. It's like 1:21:44there's like a code tab in Quad now. We 1:21:46we just launched this this week, but 1:21:48we've been kind of using it for a while. 1:21:49And so every morning I wake up and I 1:21:51start a few agents to just like start my 1:21:52code for the day. And it's crazy because 1:21:55I if you'd asked me like 6 months ago if 1:21:57this is how I would code, I would be 1:21:58like, "No, are you like are you crazy? 1:22:00Like how can you code in this way?" But 1:22:01it actually works. like this is here and 1:22:03it works and this is how I write a lot 1:22:04of my code now. And I'll I'll start a 1:22:07few agents then you know when I get to a 1:22:08computer I'll kind of check in on the 1:22:10status. Sometimes I'll just merge it if 1:22:11the code looks good. Sometimes I'll like 1:22:13pull it locally and kind of teleport to 1:22:15to edit a little bit. Um but yeah and 1:22:19and I think like you're going to talk to 1:22:20Fiona later and I think for her from 1:22:24what she told me she hasn't coded in you 1:22:26know like a decade but she's writing 1:22:28code like multiple times a week now 1:22:30because as a manager even though her 1:22:32schedule was insane uh she can still use 1:22:35the mobile app and she can use web and 1:22:37you know she can still open a terminal 1:22:38to just kind of write write code for 1:22:41her. Um, so yeah, it's it's just crazy 1:22:45like we get to live through this 1:22:46transition where the thing that we do 1:22:50and like the thing that I grew up on, 1:22:51it's just totally changing and yeah, 1:22:53it's just so cool. Like it's becoming 1:22:55accessible to everyone. 1:22:57>> And then last question for you is um 1:23:00knowing everything that you know now in 1:23:02your career, if you could go back to 1:23:04yourself when you just entered the 1:23:05industry and give yourself some advice, 1:23:08what would you say? 1:23:10Just use common sense. 1:23:12[laughter] 1:23:13[gasps] 1:23:14I think there's a lot of stuff in a 1:23:16especially in big companies that pulls 1:23:18you away from common sense. There's a 1:23:20lot of this like organia. Things are 1:23:22this way because they have been this 1:23:24way. There's a lot of misaligned 1:23:26incentives. There's also a lot of good 1:23:27things, but there's the these things 1:23:28also. So, it it's just really important 1:23:31to use common sense for this. And you 1:23:33know, early on in my career, I was kind 1:23:35of starting a bunch of startups and 1:23:36worked at a lot of startups. And I think 1:23:39there too, it's the same thing. Kind of 1:23:42use common sense to figure out what the 1:23:43market wants and what users want to 1:23:44build it. So yeah, just like trust 1:23:48yourself and and develop your common 1:23:50sense. 1:23:51>> Awesome. Well, thank you so much, Boris, 1:23:53for your time. Really appreciate it. 1:23:55>> Yeah. Thanks. Thanks for listening to 1:23:57the podcast. I don't sell anything or do 1:24:00sponsorships, but if you want to help 1:24:03out with the podcast, you can support by 1:24:06engaging with the content on YouTube or 1:24:09on Spotify. If you want to drop a 1:24:10review, that'll be super helpful. And if 1:24:13there's any guests that you want to 1:24:14bring on to, please let me know. I feel 1:24:16like sourcing very senior IC's, there's 1:24:19no wellstudied list out there on Google 1:24:22that I can just search this up. So, if 1:24:24there's someone in your org or at your 1:24:26company who you really look up to and 1:24:27you want to hear their career story, let 1:24:29me know and I'll reach out to