Do Resumes Still Matter? AI, Job Hunting, and the Future of Work

Transcript generated automatically from the episode audio and lightly cleaned up (filler words removed, misheard names corrected where clear). Speaker labels were added afterward and may contain errors.

Anna Kirk: Hey and welcome to another episode of AI and NYC. I'm Anna Kirk, founding sales lead at Thread AI.

Rob May: And I'm Rob May, the CEO of Neurometric. And the AI in NYC show talks to lots of different guests doing interesting things around the city that are AI related. They they're not all technical. They don't have to be. So, if you know somebody that should be on the show, let us know. Ryan is actually off today, so we're going to make fun of Ryan. Stuck with me. Yeah. Should we should should we make some like Eagles suck? He's a big Philadelphia sports fan, so some Eagles

Anna Kirk: We don't actually need to make that joke. It makes it for itself.

Rob May: Yeah. Okay, [laughter] fair enough.

Anna Kirk: Today's guest is Joe Perrotta, CEO and founder of SimpleApply AI. Joe, welcome to the show and thanks so much for joining us.

Joe Perrotta: Awesome. Thank you for having me.

Anna Kirk: Why don't you kick things off by telling us about the background of SimpleApply and the idea behind it?

Joe Perrotta: Yeah. So with simple apply the idea was my co-founder and I we were looking for jobs and it was very timeconuming to not only find but also apply to relevant jobs. So took a look online. There was a couple people that claimed to be able to find and automate job search and apply to jobs. Signed up for a bunch of the tools, even paid for some. None of them really worked. So I knew based on our our history and our technical ability, this is something that we could do way better. So back in January we built a proof of concept. By February we had our first paying customer and now we have over 10,000 users and we're able to find and apply to thousands of jobs every day.

Anna Kirk: So how exactly like walk me through how the product works if I'm a user and I want to apply to a job and have your platform do that for me? Like how would I go about doing that kind of from start to finish?

Joe Perrotta: Yeah, so like the name implies we try and make it as simple as possible. So, you go out onto our website, you sign up, fill out an an onboarding page, basically like you would do an application. The idea is it's the last sort of application you would fill out, upload your resume, and we will return relevant jobs for you based on like not only what's in your resume, but also some of the skills that you input, and then we'll automatically apply for them for you in just one click.

Rob May: Have you seen any companies catch on and try to make it harder? Like are are you fighting a battle of like people are making it harder to auto apply or they're putting little games or whatever in place or what are you seeing out there in the market?

Joe Perrotta: Yeah, I'm yes and no. Right. So it's sort of like web scraping, right? They want to sites don't want to be scraped like Reddit will prevent you from scraping them. Facebook especially once they s they saw how valuable some of the data was on there. So we have a little bit of downtime on some of that but there's lots of different ways around it, right? Like solving captures, proxies, there's only so much they can do without making it difficult for an average person to go on and apply, right?

Rob May: Oh, I've had some captures I couldn't solve [laughter] every once in a while and you're just like, wait. Or you go to one of these sites that's got like six layers of captures where it's like find all the buses in these pictures for, you know, you spend 30 seconds doing it. Like that's rough. At that point, you're just training Gemini for them.

Joe Perrotta: Exactly. Yes. I mean, it's definitely it's definitely a cat-and- mouse game. We'll have some downtime, but luckily there's a variety of different ATS's and if we have a downtime on one, we'll make it up on the other. And it's my I mean, I can't speak highly enough of my co-founder. His technical ability is incredible. So, if there is downtime, it's usually not more than a couple days.

Rob May: Gotcha. And how do you have any metrics yet? Have you guys been around long enough or seen enough at scale to know that like are people are they getting more jobs? Are they getting more responses? Like are you able to compare it to anything else or Yeah.

Joe Perrotta: And that's a that's a question we get all the time. Especially we try we're very large we have a very large presence on Reddit. And that's one of the questions we get all the time is like how effective is this? Like doesn't people don't care if you apply to 100 500 jobs. What we guarantee is an interview. So we have a it's a premium model. If you sign up it's you get five applies per day, one click. If you pay us $35 a month we guarantee you an interview. And you also get access to full auto apply. So what that means is you don't have to do anything. Will just automatically apply to jobs that are above a certain threshold and it happens in your sleep. You don't have to do anything and you just get the interviews and you prepare for those.

Rob May: Still just five a day even with the auto apply or not.

Joe Perrotta: With with the paid full auto apply it's unlimited. Okay. So but again the goal is on interviews, right? People don't really care so much about like is it 300 500. What people care about the end of the day is we're making the process more efficient in the sense that we're finding relevant jobs. It's also not in our interest to apply to a crap ton of jobs, right? We want to find relevant jobs, apply to those, make sure we get the interviews. So we want to obviously reduce our cost in that sense. So we have an incentive there not to send out a thousand.

And as far as the metrics, what one of the things I'm most proud of is like our interview the unique users that have received an interview is going up over 10% every month. So it it shows our ability to not only find those relevant jobs, but also make sure we're getting applies in for those jobs for those users.

Anna Kirk: And I guess from your position helping people apply to jobs, what are you seeing in terms of the overall market general trends? Like is it is it growing? Is it shrinking with AI these days?

Joe Perrotta: Yeah, it's rough out there, right? So like you don't have to spend too much time on the internet or maybe it's just my feed since I'm I'm doing stuff like this. But if you go on Instagram, TikTok, Reddit, you'll see people complaining that they have to apply to a 100 jobs, 200 jobs just to get one interview. And that takes a lot of time, right? So in our case, applying to 100 jobs, our AI can do that. Our AI AI agent and it takes basically no time out of their hands, right? So it's definitely a rough market. I think the main thing if you want to boil it down to like metrics, the interview to apply rate has just gone up, right? Right? So, you have to apply to more jobs, maybe 100, maybe 75. Depends on your your industry, your experience, all that stuff. But it is a lot of you have to apply to a lot of jobs these days.

Anna Kirk: And I imagine this is becoming more of the standard for people across the market to have to use tools like yours in order to stay competitive. Have you seen that?

Joe Perrotta: Yes and no. I think like super our tool is trying to buy back like time for users, right? So our our our core target demographic right now is like remote people, tech people that have a job, high like medium to high income, and they're trying to buy back their time, right? Like so obviously we don't want to be charging people that don't have a job or on unemployment. Those types of people shouldn't be paying if you don't have an income.

Anna Kirk: I'm curious what it looks like in this kind of AI native future when everyone is applying to, you know, 75 80 jobs using these types of tools. Like how do you as as a corporate hiring someone, how do you screen out candidates if everyone is using these types of tools going forward?

Rob May: And I and I'll just add we talk about this on the investing side. We've talked about this as markets that face AI doom loops where you make it easier for people to apply, people get more applications, so they make more screening tools and all that and you just they they they feed on each other and yeah, you know, how do you how do you avoid that or is that inevitable and we just have to accept it and you need the best you need the best agent?

Joe Perrotta: Yeah. So I think it it's these automations create like temporary inefficiencies, right? So, like the idea being if it's if you lower the barrier to entry to apply to a bunch of jobs, it lowers the value of like the applicant like the application going in. I our long-term goal here is to just create efficiency in the hiring process, right? So, right now, some of what we're doing and like it appeals to people applying jobs is like we automate applying to the those jobs. I think an underrated aspect is finding relevant jobs. It's very hard and I don't think people give enough emphasis on that part, right? Because if you're a project manager and you're applying to like to you type in the word project manager, you're going to get a thousand roles. But what it doesn't show you is like if you have industry experience, if you have the relevant skills. So really important part is finding those relevant jobs. And then also yeah, I don't think it'll create a doom loop. I think it'll creates a temporary inefficiency that is like looking to be solved.

Rob May: Do you think do you think there's going to be you're going to gather new types of data somehow? And let me give you some context around this question. It years ago 2016 2017 my co-founder at a startup that that we were building that we were we were building chat bots and we were considering the HR space and wordtovec had come out which was an algorithm that one of the first algorithms that basically encoded natural language in a way that made it really good for AI models prel right there was a and we did a CV to vec And the idea was if if I've hired you and you're a good employee, I want to find another employee like you. So I just run your CV through that. I get CV to vec and I say show the you know the word embeddings closest to you.

The problem we ran into was I didn't necessarily get somebody who worked like you. I got somebody who wrote a resume like you. So there's a little bit of a difference between the what somebody puts on their resume, how they write about their experience, and who they actually are. Right? It's a it's a it's definitely deeper. And so I'm I'm curious like how you think about that and do you think this market is going to like gather more data that makes this matching problem better or how's this going to play out?

Joe Perrotta: Yeah, I think you hit the nail on the head, right? So résumés are inherently like just an extremely inefficient way to transfer information, right? To summarize like somebody's skills, experience on one piece of paper, I think it's archaic and the future of like being able to identify somebody's skills and whether they're a good match is way beyond that, right? So we'll take into consideration what people upload. We can take a look at their LinkedIn their GitHub things like that. Take into account and another one would be like a portfolio like the creative industry. So how do you take into account those sorts of things that's like a person is not just a two dimensional sheet of paper like you said. You could get as fancy as you want with like vectors but if the input is not good you're not going to get great results there. So that's what I was saying like our our ultimate goal is to make that efficient and that means going beyond just resumes.

Rob May: Yeah. I will tell you like I don't I never look at rums. I haven't for years. When people you know when when I have to interview somebody and people send in rums, somebody else normally like screens them and decides who to interview. But then I don't look at their resume just because I just don't think it's that useful. And I'm like, particularly if they're senior, I'm like, you've had somebody write this for you and you're really good at it. And I just want to talk and see how you think and ask you some hard questions at some point.

Anna Kirk: And I'm speaking to my own personal experience with like, you know, white collar entry level jobs. It does seem like, you know, once you get to a certain level, everyone's resume is expected to be good and there's not necessarily like a standout. Kind of taken as as a given. So, you know, I'm I'm curious to see in the next 5 10 years are is the hiring process going to become more interpersonal so people can screen out and screen for authenticity or do you think we're going to have two different types of application processes? Like how are you seeing this shaking out in the next 5 years?

Joe Perrotta: Yeah, I mean interesting you say that, right? Like actually the highest rate of our paid users are people with over 10 years experience. So I believe that's more of a reflection of the current job market. But, yeah, I mean, it's no secret if you know somebody, if you have a reference, right? I think somebody who's been in the job market for 10 years plus, you can apply to jobs, but more often than not, you've met people, you've worked with people, you've proven your skills to somebody else, and if you're on the job market, you call somebody or somebody sees on LinkedIn and they go, "Hey, I know this person like can deliver. They can do X, Y, and Z. Let's bring them on here." so I think that's I mean, it's not going to be where we evolve to, but maybe that information is stored not to like play buzzword bingo, but like in a blockchain or something like that. Like, how do you how do you democratize or have access to that information more globally, more democratically versus like just I know this because I worked with that person.

Anna Kirk: Yeah. Yeah. And I think there's an interesting there's probably an interesting use case for that. And I have no idea for the life of me why LinkedIn isn't already doing this and monetizing it to some extent, but it does seem like you could be leveling up those insights a ton. Like say that you know a third degree connection of Rob has a co-orker who has a perfect job opening for you and you never hear about it because the connectivity mapping isn't that great. You know, like there there's got to be an opportunity there.

Joe Perrotta: Yeah. I just think for LinkedIn some of their incentives are not aligned with that necessarily, right? So like I the way they make money is like job postings you pay per day. I don't think that the incentives that they have created for themselves. I think they kind of created this like social network for professionals concept. And like they're obviously the biggest players in the game. So I think their current incentives don't actually encourage like efficient job matching like that. Yeah.

Anna Kirk: Is that something you'd consider expanding into?

Joe Perrotta: Yeah. I mean, well, we have a partnership with a company called Talent Labs, and they've taken on the B2B side of this, right? So, we're hyperfocused on the jobseker side. So, we want to focus on on that core concept. What is the workflow? What is the what is the day-to-day of somebody who's actively looking for a job where as our partner focuses on the hiring aspect? So, a lot of other companies, one just got, raise money with YC. They they have their founders working on this concept and they're like me where they have experience like hey I know what it's like to apply to 100 jobs not get a response how do I automate this how do I do it better and I think we have an advantage there because we've partnered with a company that has 30 plus years of experience in recruiting whereas they have the disadvantage yes they raised some money with Y cominator some other large firms but they have to learn from scratch and it's really hard to replace that 30 years of experience on the recruiting side

Rob May: how you know since New York focused. Are you able to break down some of your data or do you see any patterns geographically? Like do you do you see anything happening? Is New York a hot market or a cold market or what's going on here?

Joe Perrotta: As far as users, as far as applications?

Rob May: Yeah. Yeah. Just about like any just if you have any insights, I don't know.

Joe Perrotta: Yeah. I mean generally when looking at a map, it generally boils down to like a population map, right? So our top user like the highest state that we have users in is like California, New York, Texas. So I don't have anything to that granularity. It's a better way I guess that I think of it is like industry and like especially since remote jobs are so prevalent. So like everyone you see in the news about like how it's hard for software engineers to get jobs. The fact of the matter is we have software engineers on our platform and we get to see their resumes and helps feed our information but they get hundreds of interviews. So if you have the relevant experience and the relevant field, there's plenty of there's plenty of jobs out there that are hiring. Just the skills that people are looking for is what's shifting.

Rob May: Where do you think AI is going next with this? Is it going to is it going to eat more of the hiring stack? Like is it going to are we going to see tools where you interview with a bot and that bot asks you questions, dynamically adjusts based on your answers to those questions, screens your facial expressions. They already have that, do they? Wow.

Joe Perrotta: They they absolutely already have that. Yeah, there's a couple companies that do that that I've seen. I actively use my tool just to see what it's like and I've gotten invited to do some of those. So, sign in. And it's just very I think there's a lot of hype in those. I don't know how effective they are. I don't believe they're effective right now. I think it's a long time before they get to actually being useful.

I think what companies are trying to do, it's a little bit of a back and forth, right? So, like right now was on the job seeker side, they're like, "Hey, we're going to apply to hundreds of jobs." So, if you look on LinkedIn or some of these other places, it's very common for jobs to get a thousand applications. So, their response is, "All right, well, what if I start to screen them what using like those virtual LLM or AI interviews?" what I think they're going to see is that they're not actually very effective. A better way to do it is to find those find those matches before the actual interview. How do you get a good candidate before you get a human in there? I think it's going to be pretty irreplaceable for a while that in order like I think you me mentioned this before, you don't even look at the resume. You want to be able to but let let the AI let the algorithm look at the resume determine who's relevant and then when is it worth your time to speak with someone?

Rob May: So, you know what I do when I when I hire people at I started doing this at my first company, so we've done it a few times now. I make the hiring manager interview 10% 10% of the people they interview have to not meet the job requirements. And I the reason I do that is to introduce some intentionally some variability into the process both both because you had this it started at my first company so it's probably like 2010 we looked around and we were 20 people and we were like 19 white men and one white woman as you can imagine and so then somebody goes like hey we should we should have some diversity and because people hire from their networks and everything else and people like we should have diversity and you have to make a big push for it and then some people raise their hand and go no don't we want to hire the best person and you're like I had a executive who was really savvy about how to get around that.

He's like, you don't change the bar for who you hire ever, right? You don't lower it for somebody who's you know, not like the rest of the company. But what you do is you might change the bar for who you interview and you might find out that sometimes like like I'll give you an example. If you know I think I think if if you were a woman minority who grew up in Boston and went to UMass first generation college student and your parents you had a you know job like it at a local restaurant chain or something like your parents just think it's amazing that you have a corporate desk job probably and like you might be brilliant you might be born with just a mind where you're like I spend my weekend writing graphics processing algorithms cuz it's just fun and whatever and like you would never get a job at most tech startups because you didn't have the right schools or the right experience on your resume and so we were always very concerned with like it's like I'm just I'm a big fan of randomness and I know sometimes it works against you and you're like well that was a waste of time but also it it it's sometimes how you find the diamonds in the rough.

Anna Kirk: So, I mean it's it it begs the question like what and to whatever extent you can answer this is fine, but what types of demographics of people are attracted to your platform? Do you find it's homogeneous? Is it kind of elevating visibility for people who might not typically apply for those types of jobs?

Joe Perrotta: Yeah, I mean our we skew a little bit on the male side as far so like the way that we collect the information about like demographics is the same way that you would like collect it when you apply. So people opt in not to like answer that information be like what's your race, what's your gender, stuff like that. So a large percentage of people opt not to give that information but the one the information that we do collect skes a little bit towards male. We work all across 50 different different countries. So we also are heavily US-based. We have a lot of users in Canada, in Great Britain, a ton in India. Unfortunately, they don't like to pay. But on the on the demographic side, I think people as far as like applying to jobs, since we automate some of that and we give them relevant jobs, I think they wouldn't find those jobs necessarily because of their demographics. I think it's just that jobs are so desperately spread across all these different platforms.

Rob May: Yeah. Is that is it like does it vary by sector? Is like more white collar, more blue collar?

Joe Perrotta: Yeah, I mean to a degree it's mostly white collar. So most of the jobs we our our core right now it works best for remote and tech users. We just added the ability to search by distance which is a little bit of a headache to solve but again co-founders great super just give them a couple days solving this technical problem. And now we're able to because that was one of the main things right we work I'm able to look at the stats when somebody signs up we're able to see how many jobs they got returned how quickly they're using it when they're getting interviews. And I noticed like, hey, we work really well for these remote people, but we're not working well for these people looking for on-site roles. And that's like these days there's a lot of people.

Rob May: Interesting. Do you guys have a favorite interview question that you like to ask people? I'm just curious. Popped into my head.

Anna Kirk: There's one I'm just going to like repeat exactly what Simon Sinek has already put out there on one of his podcasts, but one of them that I that really resonated with me was I'm going to be your new manager when you join this job. I hate surprises. What might I be surprised by with us working together? I thought it was really fresh and a nice reframe on the classic like what are what are your weaknesses? Because it's not necessarily you don't want to get a canned response like, "Oh, my weakness is that I'm too much of a perfectionist, right? Like you want to get something authentic."

Rob May: somebody was telling me the other day they hired a new woman at the company. It turned out she had an only fans and they were like, "We got to let her go because they were the kind of company that couldn't tolerate that." But, that say depends where she's applying. So, yeah.

I'll tell you mine and I got this from a guy who used to work for me. Likes to say, when I go to do your references, I'm going to hear a lot of things back. What's the thing I'm likely to hear back that you're going to tell me, "Oh, that's actually not true." But sometimes people say that about me. What was your do you have a response? It get it gets for me personally.

Yeah. Yeah. I can tell you what mine is. I tell everybody that works for me this way. I come across as impulsive, but I'm not. I'm the I'm the opposite of that. So when you when you work for me, something happens that was unexpected and then I go, "Here's what we're going to do." And people like, "No, no, we should talk about it." And I think the people that have worked for me for a long time now say, "No, Rob already thought about that. He thought about the fact that that might happen." That's just like because that's just the way my mind works because I think about like all the permutations on everything. It's just like a weird habit that I have. And so yeah, I probably already thought about it. And then it people are like, "Oh, we got we gota have a war room and whatever." And I'm like, "No, here's what we're going to do." so that's mine. Most of the most common one you get is something people some people give really really good answers. But sometimes people just say stuff like, "Everybody says I'm late, but I'm not. I'm just Yeah, you'd be surp

Joe Perrotta: when interviewing people. I kind of I leave a open-ended question. So, first thing I look I look to see how prepared they are, right? So, if somebody shows up, the people that I've hired at at the company so far are people that show up prepared. They're like, "Hey, I researched this. Here's this idea. I'm looking for someone who's like self-starting and I don't have to micromanage. And then also just like you said, right? Like give them an open-ended question. Are they going to dig themselves a hole? Yeah. So if they dig themselves a hole in that question, it's probably going to be rocky.

Anna Kirk: You mentioned this earlier and I would love to double click on it, especially since you mentioned a lot of the applicants in your platform are technical. There are a lot of engineering jobs out there. So it's not a quantity problem, but it maybe is a type of skills problem, maybe it's an education problem. Like what are what can you speak to in terms of those dynamics because we're getting, you know, people hired by OpenAI or Meta who are getting paid millions a year. But let alone that's not that's probably not indicative of the broader kind of technical job seeeking market. So what what are you seeing there in terms of Yeah.

Joe Perrotta: I Well, so just I think it was just today at Meta, they actually laid off all 600 of those people. Yeah. Yeah, those super intelligence labs people. So I think part of it is it's a little bit well it's very reactionary some of that but on on the technical side there's a lot of skills if if there's a good technical person some a lot of times those skills will translate right so you do you have to have an experience with like a SQL database or elastic search or like this sort of data construct generally that will like translate if you know if you understand the concepts of databases it'll translate but that's where some of the inefficiencies come in for some of those jobs, right? So, I gave the example of project manager before. A lot of times job titles will just be software engineer. And what people don't see when they apply is they won't necessarily see like some of those keywords, what frameworks you need to use, what packages. And that's where our algorithm comes in, doing that research ahead of time, looking at at those keywords, comparing it to what the user has, and letting you know like, hey, you're 70% match, 50% match, whatever it may be. So that way users can not waste their time when they apply.

Anna Kirk: Is there a way for users to tailor that more customizably on the platform? Like could I in a in a free text box say I am an extrovert. I really want to work in roles where I'm already ready to apply to 100 new jobs. Not not yet. [laughter] Don't tell my current employer that.

Joe Perrotta: Yeah, 100%. That's actually one of the really important parts we have there. Right. So people are used to prompting an LLM. So we have this open text box at the bottom and our suggestion that we put there is like put the equivalent of a cover letter like a generic cover letter. But a lot of people because we were able to see what people put in they'll put in stuff like that like I want this type of job that is rewards people that are innovative or rewards people that take initiative things like that. So it's able to take that into account because we do use LLMs in the back end for some of that.

Anna Kirk: And yeah, I think that's an interesting take and really powerful on how people can find jobs that are not just like a match on paper, but a match on like their personality. Yeah, this brings up an interesting point and it's kind of similar to a conversation we had on dating last week, which is people tell you what they think they want, not what they actually want. So, I'm curious.

Rob May: Oh, I can tell you I've hired so many people. When you hire for startups, one of the things that you tell people is that you have to be you have to be comfortable with failure because we're going to try a lot of things. We're going to have a lot of projects that don't work, a lot of features that don't work because you're you don't have millions of users to put stuff out against and test and interview and survey.

And even those big companies like you know Coke and Pepsi and Procter & Gamble like they spend millions of dollars on these surveys and they still have failed product launches, right? Like it's hard and across the board people be like yes that's the kind of place that I want to work where failure but it turns out they don't when you when you try to push them to do something that they are worried might fail they don't want to do it and so it's hard it's hard to be really innovative unless you find that person that's just like very comfortable saying like well that didn't work like let's move on to the next thing so I think there's a lot of a lot of problems like that in the job market and I think it's I would love to see AI tools that start to help like with team chemistry and figuring out who will work where to get well together and particularly as a from a manager perspective where you need the team to go because sometimes you're like I just need this team to work together and maintain the status quo and other times you're like I need them to push forward.

So sometimes you want to build a team where you're like I need somebody on the team who's a little bit of an aggre aggressive but I need somebody on the team who's also a little bit of a like let's pump the brakes. I need them to be I need the aggressive person to win, but I need somebody to put a little bit of a break on them, but I need them all to work well together and not get mad and hate each other. And like that's a hard thing as a manager to build together.

Joe Perrotta: Yeah. When I when I had a day job, I think that goes into a little bit of like personalities, right? So when I joined my my former employer, they had us take like this personality test and that sort of like it gave a the output was like, hey, this is how I respond to criticism. This is how I whatever. Basically essentially personality test. I think it gets a lot to your point there of like how do you understand those types of people? It lets you understand how to build those teams. Because yes, the skills might be there, but if they can't work together cohesively, it's not going to work most likely.

Anna Kirk: So, I'm always curious about that like when you I've I've never taken a personality test as part of a job application. So, when you apply to a job and you think that this personality test is going to be incorporated into your evaluation of your performance, like are you honest? Because I, you know, I could see some people in especially in like a say it's in like a sales and trading job and you're an introvert, but you want people to think you're an extrovert because that's a like characteristic of people who apply to those jobs and are successful. Like that's a good

Joe Perrotta: So, I haven't seen them before hiring. I've only taken them after being hired, but I mean, I think you're right, right? Like the incentives are like I what what do they what do they want me to say? Right. If you got hired and you already have the job, maybe you got your first paycheck, you're like your guard's probably down a little bit. You're like, "All right, I'll be a little bit more honest." But if you're if you're doing it as part of the interview process, I would probably be incentivized to be like, "Well, I really need this job. What do they want me to say?" Yeah. What do I think the right response should be?

Anna Kirk: Rob, what what's your take on personality tests as part of the application?

Rob May: I have I've never used one as part of the application process. I've been given them years ago as part of the application process. I never got hired for any of those jobs, which probably tells you how I came across. Probably a little too in independent minded in most things. To use a yeah, to phrase it a nice way, I use them a lot to build team chemistry. So at some of my companies, we've used I've asked people their MyersBriggs sometimes. We've used one called disk, the ISC. That's pretty interesting. And it really helps how you map. I mean, these things all have versions of truth and then and then people are people are different in different situations. So like disk, what I like about disk is disk gets at the fact that you might be different in different situations and even if like D is for dominant. If you have a dominant personality, you might not be in every room dominant, right? Or in every situation or someplace where you don't have as much confidence. So, there's a lot of things to think about that.

I actually think about that a lot because the advice that I give new CEOs is always that cuz everybody wants like a management trick that they use that makes everybody do what they and what they want. You're like that trick doesn't exist. What you have to do is you have to tell people the thing they need to hear to get them to do the thing you need them to do. And some people need to be smacked and some people need to be hugged. And like there there's a type of person that if you're that if you're you know they they they really want to do something ambitious and you're like I just I don't think you can do that. Some people are like yeah you're right. And other people are like I'm going to show you. And you need to know which kind of person you're dealing with so that you know whether to say that and whether you crush their spirit or whether you motivate them. And you know you can extrapolate that down a hundred different personality traits. It's really it's really hard and even people who are great great managers don't get it right all the time. Yeah. It's just one of those it's one of those things.

Anna Kirk: I don't think you could smack people anymore, but

Rob May: Well, yeah, but you know what I mean. Metaphor,

Joe Perrotta: but yeah, I mean I 100% I' I've been experiencing this myself, right? Like how do I give people hard love? How do I give people feedback? What I've noticed is like I just have to accept the inefficiency in the short term, right? So like as I'm learning how people respond to certain things, I will not necessarily I'll be like, "All right, let me pump the brakes. Let me one of my favorite features in Slack is I'm going to schedule this message for tomorrow. Yeah, let me think on it. Let me let them do this other thing. So at least from my perspective without without doing any like personality tests with the people I work with is like just giving people space and one of the values at my previous companies was like grace. Like always assume the best of the other person. Assume that they're trying to get to the same place you are. And it's just going to take a little bit of time.

Rob May: Yeah. My first conversation with somebody who's not working out is always to assume like it's my fault, right? And you go in and you say, "Look, you're not really delivering what I thought. It's probably my fault. I'm not managing you well. Can you help me understand what I need to do better, be clearer, whatever it may be." And then you see if you give them an out, you give them a chance to criticize you to do whatever. And then if you're a manager, you have to learn to just take it and not care if somebody says, "Yeah, actually you don't do this right. You're not clear in the instructions you give me." Or something like that. But I think I think you I believe in that grace principle as well on that first one. And then if it happens again, then you go, well, now it's different.

Joe Perrotta: Yeah, for sure. I mean, you can only give people so many chances, but yeah, I always try and reflect internally. Am I giving the right directions? Was I clear on the goal? A lot of times I don't really care about how something's done just as long as the goal is met. But yeah.

Rob May: Yeah. Well, let's let's switch a little bit to some of the news of the week. I mean, you brought this up earlier that Metallaid off 600 people in their AI division. What do we think this means for there's been some criticism of their llama models, which I can tell you having worked with them, the Llama version 4 models actually are pretty good. They're pretty competitive with other things out there across a lot of use cases. But what does that mean for Llama? What does it mean for their AI ambitions? I mean, Zuck just went through a thing four months ago where he called 50 of the top AI researchers in the world and I've talked to some people that were on that list who didn't go to Meta for this reason because they were worried this would happen. I mean, they got a nice payday for a little bit. So, I mean, the money probably wasn't too bad.

Joe Perrotta: I'm not exactly sure what their strategy is longer term to be quite frank. A lot of the other models they like so their model is open source and a lot of the other models have made it so easily accessible and also as a founder we use a lot of their startup credits so I haven't seen Meta offer any credits like that but the reason that we use the models that we do they are for the most part same very similar across the board we're looking for cost and speed and whoever gives us credits I haven't seen Meta offering credits like that I don't know they seem to be going in circles

Rob May: Yeah, they haven't really set up llama like on a comput infrastructure that that you can just use. You have you have to go to one of the other infrastructure. Well, it's you know in AI it's a little bit weird because if you're a lab like that you have to or if you're doing a lot of AI you sort of do have to open source something to attract talent. People have to think they're working on cutting edge stuff and you need an external signal. I think it's part of the reason that Apple is perceived as having struggled there because they don't have anything like that in the market. Now they have come out recently with some small language models for mobile devices and stuff that are pretty interesting. But yeah, so I think maybe that's why they open sourced the llama models.

Joe Perrotta: Yeah, I think I mean Apple I think has dropped the ball the most. Like autocorrect on my iPhone should not be this bad still. It's kind of crazy how terrible it is.

Rob May: Yeah. If you get a text from me with with, "Oh, I don't have an iPhone, but still, I always, for some reason, instead of typing maybe, it's m a y ne." Like, that's not a word. I will never use it, like, why can't it figure that out?

Joe Perrotta: You probably typed it wrong one time. Yeah. And sent it and the AI is like, "This is what he means." [laughter] Yeah. I Yeah, I think Apple out of all the major tech companies, I think Apple's dropped the ball the most there, especially they have all these users. They could be using them to I think their last iPhone launch last year was to be was supposed to be like what was it like Apple intelligence or something like that they were calling it and I think they just fumbled that horrifically.

Rob May: Yeah. We'll see. I The interesting thing about Meta from a strategic perspective is they can benefit more from vertically integrated AI than almost any other company. They have they they run their own data centers. They've got some of their own they're working on some of their own hardware. They have a lot of their models. They have use cases. I saw a presentation at a conference a couple of months ago that they're working on custom hardware for recommendation engines, right? Like I don't think you could build and sell a recommendation engine AI chip broadly because there's probably not enough big of enough market and probably 80% of that market. I mean the market would be like Netflix and Meta primarily or YouTube but Google has their own probably.

Yeah. Yeah. I'm sure Google's probably even more advanced than those guys but you know and they have all the data, they have the network, they know all the stuff about people. They have the ad platform. You can add AI at, you know, who you should connect with, what content you should see, what ads you should see, like all the way up the stack. And they monetize it themselves as well. And so that's they control their own monetization. So it's really really a chance for deep AI integ. So in some sense, it's a good place to go. But I think like one of my friends who got a call to be on that team was like, there's no way this sticks. This is like I mean, you're talking about a company that went all in on metaverse. It was a really bad decision. And I think that people think about that and the people in Silicon Valley particularly in AI like they talk so much about the other places, right? Like why you shouldn't go there and I talk you out of it.

Anna Kirk: So well it does seem like other companies are moving in that direction speaking of vertical integration. So maybe maybe that's a good segue into one of our other articles we wanted to talk through this week. ChatGPT Atlas OpenAI's new browser. A big step for the company to try to append Google Chrome theoretically. So it's first set to roll out on Mac, followed by Windows. But the key feature here that really wowed me is that there's basically a side panel with an agent mode that automatically has context of anything you're browsing on your screen at any given time. It has full access to your browser history. So going back to your point around vertical integration, do we think OpenAI is going to start monetizing on ad revenue?

Joe Perrotta: It's probably inevitable. I've seen also some I mean on the monetization side I've saw their they're getting into like we'll say exotic videos in by the end of the year.

Rob May: That was very diplomatic of you Joe.

Joe Perrotta: Yeah. Yeah. So yeah I feel like it's inevitable. You have to create some revenue somehow. And I think that's what like Sora 2 is kind of trying to get at right. Like so how do we get this platform? How do we generate some revenue? I think they're feeling the pressure from all the money that they've raised. I don't even know what it is at this point. But I think it is a play to start serving some of those ads some products because from what I've seen you can in chat now if people will search for products they're trying to when you go in there you don't have to ask a question necessarily right so say here's your news of the day here's these products so they're it's not built into the answers yet but they definitely seem a little bit I wouldn't say desperate but almost desperate to monetize I will say I've used the Atlas browser a little bit And it's very similar. They were doing something similar with agent mode inside of chat already. So access to a browser, access to like a Python session.

So I think it's going to be tough, right? It's just from what I've read, it's basically just built on top of Chrome, so on top of an open source like Google framework. One of the companies that did try and build their own browser, Perplexity, I believe it's called Comet, something like that. They actually had a ton of security issues. Yeah. So, I think they they sort of half went into this browser play of like, all right, well, let's see what sticks. I think it's just one of three or four bets that they're making to try and monetize.

Rob May: Yeah. I'll I'll take the contrarian position on the monetization thing, and I'll say two two more reasons, right? One is that they they want to get data on how people do stuff and use that data set to train more models. And so, living in the browser like that helps them gather more usage data. I think on the monetization side, whether they're serious about it or not, they have to sell a big monetization story because they owe they have a trillion dollars in commitments to build data centers and they have 15 billion in revenue. So, they're two orders of magnitude short of where they excuse me, where they

Anna Kirk: don't do too much math, Robin. It makes the problem go away.

Rob May: And so I think you have to sell that story by saying, you know, well, we're going to get $12 billion from ads in our browser. You know, you can run that math. I think they have to do a lot of things like that. But like I said on a previous show, I actually I put my short on on Oracle under the idea it's up about 5% which is good. But it's it was under the idea that at some point OpenAI is going to struggle to raise money and that's going to ripple through the chain and I think Oracle stock will be hit the hardest. But not going to leave that on very long because I'm that makes me nervous. Still they're way overlevered and that's not going to that's not going to go away any but that that's why I think they might be doing some of it.

Anna Kirk: Interesting. Yeah, I've been putting my money into gold which has been working out well for the past couple days. But there is an interesting I mean I think there's an opportunity here for OpenAI to kind of double dip and monetize on both sides if they can also go get into the aspect of content creation around marketing for some of their paying customers on the ad side.

Rob May: Well, what's interesting about AI is that it's breaking down a lot of what used to be traditional strategic barriers because if you think about the SaaS world, you used to be able to write code and as you got ahead of people just on your codebase and the maturity of your product, that was a barrier for 6, 8, 10 years until the market matured and people got to feature parity and you got to that point in the market where all the features are marginal and not really that needed. And with AI making data one of your core assets, that data may be applicable to multiple markets that you play in. And so you might play in two markets that historically might not have made sense that now do because data is your asset, not say the customer relationship or you know the production workflow to get there. And then you have this idea where you know code's getting easier and easier to write faster and faster and so that's not a barrier anymore. So yeah, so it's weird to think about strategy in an AI world like from a traditional Michael Porter's competitive strategy sense and where where you demarcate what your company's good at. Like what are the lines of demarcation for what you should do? What do you think that is for OpenAI? Because they're they're getting into enterprise consulting ads.

Anna Kirk: I think their goal is just get too big to fail.

Joe Perrotta: Yeah. So yeah, I view like OpenAI has quoteunquote the best science out of the models, right? They're usually the ones innovating a little bit more. Like they have like agent mode as that's accessible in like the chat. It's one of the only ones. So I think their science is best. I would say like claude is some of the best that code. Hence cla code and then Google it's just an insane advantage that they have. Everyone has like Gmail, Google Calendar, all these things. Like their ability to integrate is I mean I already use it across a bunch of different things. YouTube has a data set. They have their own they're they're on like their sixth version of TPUs. So they've actually figured the chip thing out and it's like very low power compared to GPUs like we well so they've run out of text data allegedly to train on and like the next areas to train on are going to be audio podcast like this. Crazy to think we'll be on the training set and then also video. So but yeah, Google has a massive advantage there. The expertise, the ability, forget forget like training the model, like how do you store all of that data? Like that's how does OpenAI is going to have to resolve that problem.

Rob May: That's cool. And another thing we haven't here, but like the, you know, OpenAI is talking to Broadcom about co-designing chips and becoming more vertically integrated, which I think does make sense because I think the vertical integration on the AI side, I think people have started to realize that these general purpose chips are fine for a lot of common use cases, I don't think it's going to hurt Nvidia, but as there's more specialized use cases for these AI workloads, I think these ASIC make more sense and you know the workloads that you need. I there's going to be a ton of innovation there. I think over the next 10 or 15 years. So that's exciting for me to see. And I think you'll see I think you'll start to see more and more companies building their own chips like that. So drisking them from a supply chain perspective.

Joe Perrotta: Yeah. You've got part of that too, right? Which is I mean everybody's trying to get more GPUs and it it's a it's a long-term strategy. I it takes years Yeah.

Rob May: To offset that. If you if you have your I think I got a friend at Microsoft that said like three years ago their goal was to take to make themselves 50% of their dependence on Nvidia go away within like five years or something like that. It's a long-term play play if you have to do stuff like that.

Joe Perrotta: Yeah. I mean they're building I don't even know how many fabs here but yeah I mean strategically it's a it's a security issue national security issue as well right. So that's why reason we had the chips act and trillion dollars into building all these fabs here.

Rob May: Yeah. Do you know what people don't understand? They they with the with the fabs coming here we still don't do the packaging here. You still have to ship them over to Taiwan to get them packaged even after they're fabricated here. So yeah. We're going to have to have to figure something else out about that. But I think like if if I was looking for a place if I was like a physics or material science person and I was looking for a place to like do a PhD and think about I would figure out like a new way to make these chips. There's got to be something. I mean there there are non-silicon versions. People have tried 3D printing. It doesn't work very well.

Like but there's there's got to be some better ways to do this without using like cutting. Like I don't know if you ever watched a video on how they do this, but it's like these machines heat up a drop of 10 to like 10 TIN, right? To like 2,000 degrees or whatever, and then they drop it and that causes some things and they blast it with a laser and that etches creates a light wave that etches. It's like it's something like that. It's an insane process to hit that drop with a laser as it's the whole thing is crazy. Like there's got to be a cheap way to do this. I mean, these machines cost $500 million or whatever.

Joe Perrotta: Yeah, ASML is one of the ones that makes them, I believe. And like I think we're hitting the limits of phys physics on like how to how close we can get some of the circuits there. Like the electrons will like quantum tunnel over. I think one way area that they're looking to make gains is like stacking vertically some of those things. But I think quite literally we're hitting the limit of phys physics on a lot of this.

Rob May: Yeah. So we will see. What else do we have Anna on the articles news of this week?

Anna Kirk: And so anthrop Anthropic took a bunch of crap, right, for particularly from David Sacks. David Sacks is saying that anthropics what is it called? Regulatory capture for sort of coming out and talking about what policy should be and American AI leadership and then Jack Clark wrote this sort of like doomsday piece on his newsletter. And I just think all this is like weird positioning. I don't really have a strong opinion of it on where any of it should go. I don't know. Dario's letter that came out this week was incredibly diplomatic. And of course they're going to want to advocate for a federal standard instead of state byst state, but we'll we'll see if that sticks. I'm skeptical.

Rob May: I think they've I think Anthropic has played the I think them staying out a lot of the government stuff compared to how much the other companies do it has has benefited them. I think they've built a good business and I'm kind of surprised because when they first raised money I just thought like well this that Frontier model game is over they've done a really good job differentiating and getting out there and I you know I think Sam Altman I think has a mixed reputation for being so focused and out there on the government and policy stuff particularly given some of the decisions that they made. I mean their initial thing was was OpenAI right like make this available to everybody. That's where the that's where the nonprofit started and they're nowhere near that now.

Anna Kirk: Yeah. And ironically, I feel like Anthropic has taken more of that stance than OpenAI and Dario has been very adamant about advocating for observability.

Rob May: Yeah. They called it a const I think constitutional model from the beginning or whatever that it had some rules built in and all that. That was kind of their initial Yeah. Play. We're going to have a guest on at some point who lives here in New York who's done some work for them. So we'll we'll see what she says on the ethics side. Put a pin in that. Yeah. Yeah. Cool. We have a few minutes left. Do you want to jump into our question?

Anna Kirk: Yeah. So, we're gonna close this week out with Joe, we do this every single week where we have one New York specific question. So, it'll be like favorite bodega in New York. Favorite bagel place. And this week's question is favorite New York City meat cute. If you meat cube, do you know what a meat cute is?

Joe Perrotta: What is that?

Anna Kirk: Most men don't. Okay. [laughter] Oh, damn it. All right. So for female listeners, you might know what this is, but it's basically like is there a story or anecdote you have about spontaneously meeting someone, whether that's a friend or someone you ended up dating in New York and kind of how that happened. If you don't want if you don't have one, we can pick another question for this week.

Joe Perrotta: I mean we could give the example how we met, right? So, I went to a VC event, on Tuesday and at the end, Rob was one of the speakers. At the end, everyone was lining up to talk to the speakers saying, like giving their, elevator pitch. So, I went up to Rob and I was like, "Hey, I don't know if we'd be in match like we're funding all this stuff, but I would love to go on your podcast."

Rob May: So, and we just had a cancellation, so it [laughter] worked out.

Joe Perrotta: So, I get an email like a few hours later, and I'm like, "Wait a minute, is this the same guy?" And that's the beauty of New York, right? Like, you can have so many spontaneous interactions like that.

Anna Kirk: Like Rob and I met, well, I guess technically we met in Los Angeles. It's not a good example. But no, my one of my best friends, who I can guarantee you will end up being one of my bridesmaids, she met my roommate in the laundromat that our two buildings shared. And we've stayed in touch ever since. She's like one of my best friends.

Rob May: So, a lot of my best New York I guess if I had to say meet cutees would would actually not happen in New York. So, I didn't meet you in New York. I didn't meet Ryan in New York. I met Ryan in Boston at a conference. Stephen Messer, who we had on as a guest, who I would say was one. We met in San Francisco. He's one of my best friends in New York. But he and I we met at we met at a dinner in San Francisco in 2015 or 2016 when the AI community was very small and we were the only two East Coast people there. I was still in Boston at the time and we just ended up sitting next to each other talking about how much San Francisco sucked as a place to live, which I love San Francisco. I love to go visit and I have a lot of friends there, but like I wouldn't live there, man. That city's whacked. And it's just weird. Like it's kind of cool to see like all the hills in a city like that, you know, in an urban landscape is really unusual, but like it's also hard to navigate. The public transportation for a really liberal city is terrible.

Anna Kirk: It's hard to have trains on hills.

Rob May: I know. And so yeah, and I mean I know we talk a lot of crap about the New York City subway, but it's pretty good by US standards. US standards. Yeah. I was going to say global standards, but I was like, "No, it's not not [laughter] global standards." But, that'll be interesting. If anybody knows the head of the MTA or somebody at the MTA that does data science, like let's get them on here and talk about this. How do they use it to make the trains run more on time? That's always a question.

Joe Perrotta: It's very archaic from my understanding. They still have switches from like the 1910s from what from what I understand.

Rob May: Well, that's a good parallel, I think, into this agentic world, right? To talk about like, okay, if you wanted to optimize the MTA with AI, like you're like, oh, well, then you need a robot to go flip this switch out because it's, you know, there's still some some legacy constraints. And I think I think too many entrepreneurs have looked at AI workflows and thought they could do these things where you're like no there's actually it doesn't it's it doesn't work the way that you think and it's harder than you think to automate the whole thing. And I that's why I think this is going to for all the AI innovation I think it's going to diffuse through the through the you know entire you know ecosystem and economy a little bit slower than people think.

Joe Perrotta: Yeah. And in that instance I think founders like myself underestimate some of the bureaucracy. Like ask yourself if it's that simple, why hasn't it been done yet? So like money, cost, there's lots of reasons why we're still running those old wires and switches inside the subway. And disorganization of information like if if I have info in a file cabinet in my office that no one can look at when I'm not there. You know, you can't easily automate that same type of way. So you really have to meet people and meet companies where they're at.

Rob May: Yeah. Next time we have a VC on, we should talk about this idea that, you know, for all the So, I have an opinion that most of this startup advice is kind of And I the reason I believe that is because for all the advice we get about like, oh, do 100 customer interviews and do this thing. It's like we're not any better. Like the we have the same failure rate that we had 25 years ago. So, what does that mean? I think it means there's a lot more randomness in the process than people think. A lot more of what I would call path dependence. So yeah, that that'll be a good topic for another show. Not that there aren't there's things you shouldn't do and there are things that are wrong, but I'm not like a lot of people do all the right things and follow all the YC things and customer development things and they just

Joe Perrotta: you're talking about in the context of go to market like reaching out to customers or all of it.

Rob May: Yeah. Yeah. How do you how do you identify a problem space to work on and what do you know how do you get your first product and then how do you sell it initially and like I think a lot of the advice the majority of it's crap.

Anna Kirk: Yeah. I mean maybe a less pessimistic way of phrasing that would be like those are some things that you should do to set yourself up for success, right? Like if you can do those things and still fail, but if you don't do those things, there's a really good chance that you're probably going to fail anyway.

Rob May: Yeah. Seems people get pretty lucky doing stuff, but we'll see.

Anna Kirk: Good. Well, Joe, thanks for being on the on the show today. This has been fun. Hopefully if you're looking for a job, go check out SimpleApply. SimpleApply, sorry.

Joe Perrotta: SimpleApply.ai.

Anna Kirk: I put the sticker on my laptop.

Joe Perrotta: So, we rank on Google for both keywords, so you'll find us either way. So, [laughter]

Rob May: that's smart. Yeah. And do you rank in ChatGPT?

Joe Perrotta: To a degree. We were for a little bit. Not Not so much anymore, but

Rob May: Okay. Yeah. Peep. Everybody's got to work on that now. So, please like the podcast, leave a comment. Send us your swag. We'll wear it on the show. Yeah, you're a startup. Tell a friend. Send us somebody who should be be here. And please go watch the other episodes. We'll see you guys next week.