Skip to content
Silicon Valley GirlSilicon Valley Girl

Superhuman CEO: How to Position Yourself Now Before the Next AI Phase (2026–2027)

📌 Your AI agent needs memory and context. MongoDB Atlas is the data layer for both — works with Claude, ChatGPT, Codex and Cursor. Free cluster ↓ https://fandf.co/45YqQjF In this episode of the Silicon Valley Girl Podcast, Marina Mogilko interviews Shishir Mehrotra, CEO of Superhuman, founder of Coda, former Chief Product Officer of YouTube and a Spotify board member for over a decade. In this episode he explains why the best jobs never come through the recruiting funnel, the PSHE framework he built at Google for how promotions actually work, the point mid-career where the rules quietly change and everyone panics, the interview question he used for hundreds of hires until he talked about it publicly too much, and the three AI agents he actually runs every day. Plus what Bill Campbell told him in 2000 that changed how he measures his own success. Links: 📩 7 skills that get you promoted — https://siliconvalleygirl.beehiiv.com/p/7-skills-that-make-you-irreplaceable-adc6?utm_source=youtube&utm_medium=description&utm_campaign=futureproof-sub&utm_content=7-skills Timecodes: 00:00 Teaser 00:38 If you started your career from scratch today 01:04 Stay out of the recruiting folder 02:56 Why "learn every AI tool" is the wrong advice 04:26 We're hiring judgment 04:50 Every skill is best learned in low-stakes places 06:37 Is the career ladder getting erased? 06:55 PSHE: how promotions actually work 08:00 The trough of disillusionment 08:36 AI pushes the entire ladder up 10:08 Why he doesn't like the word "replaced" 13:18 What happens to entry-level careers 15:42 Eigenquestions: the skill he hires for 16:40 The teleportation interview question 20:24 How to practice this without risking your job 22:10 Inside Grammarly: 100 billion AI queries a week 24:54 Demo: building agents that work where you work 28:14 Chat, do, assist — three ways to think about AI 29:40 The three agents he uses every day 34:37 Bill Campbell's advice 37:20 The metric he still measures himself by 🔗 Links: Instagram: https://www.instagram.com/siliconvalleygirl/ Silicon Valley Girl Podcast Instagram - https://www.instagram.com/siliconvalleygirlpodcast/ Linkedin - https://www.linkedin.com/in/marinamogilko/ X - https://x.com/siliconvalleymm My Companies & Products: https://partnerships.marinamogilko.co #SiliconValleyGirl #podcast

Shishir MehrotraguestMarina Mogilkohost
Sep 8, 202638mWatch on YouTube ↗

EVERY SPOKEN WORD

  1. 0:00 – 0:38

    Teaser

    1. SM

      You learn to be a manager. I spent years giving people the opposite advice.

    2. MM

      What would you say to people whose job is already automated? This is Shishir. He spent 20 years deciding who gets promoted. He ran product for YouTube, he sits on the board of Spotify, and now he runs the company behind Grammarly. 40 million people use it every day.

    3. SM

      You think about a world with AI everywhere, what happens? It actually pushes that entire ladder up. The original skill didn't go away, it just shifted.

    4. MM

      The job you're doing today and the job you get promoted for aren't the same anymore.

    5. SM

      Since it's become my number one indicator of someone I wanna work with, someone I wanna hire, is how good are you at-

    6. MM

      You

  2. 0:38 – 1:04

    If you started your career from scratch today

    1. MM

      represent someone who's been in multiple careers. So you worked at YouTube, you worked at Spotify, you started your own company. Now you're the CEO of Superhuman, which also incorporates Grammarly and helps humans become super humans.

    2. SM

      Mm-hmm.

    3. MM

      So if you had to start your career from scratch today, what would you do in the next 12 months to stand out on the market?

  3. 1:04 – 2:56

    Stay out of the recruiting folder

    1. SM

      I'll give a couple pieces of advice. I think one thing that people often don't think about is how to stay out of the recruiting folder. What I mean by that is the most interesting jobs, in fact, all my transitions, for the most part, have happened as s- through an interaction where we weren't talking about recruiting. We weren't, we weren't going through a job conversation. And so I think a lot of people focus their energy on, "How do I interview well and how do I get through the screening," and so on. But actually, the most interesting candidates I've ever hired or places I've, I've jumped in usually happen through some interaction that didn't start with a, "I'd like a job," or, "Would you, would you take this job?" So for that, I think it's be interesting in the world. Start projects that people might see, write things that people might see. Not only is it easy to produce those things, the outlets for being able to spread them are, are, are so wide. I think that's really important. I mean, the way I ended up, um, at, uh, you know, I've been on the board of Spotify for over a decade now, and the, the, uh... You know, it started with a paper I wrote. You know, Daniel was i- interested in it and we started talking about it, and then it led to, "Hey, would you, would you do more at the company?"

    2. MM

      On social media?

    3. SM

      Actually, it wasn't even shared. In this case, it was passed around through different people. It was called Four Myths of Bundling. It's how, how to think about bundles. Somebody sent him the idea and said, "You should talk to Shishir about it," and we ended up chatting about it. So I think the main advice I give people, that in a world full of candidates trying to find spots, you have to find your way to stand out, and the best thing to do is actually avoid recruiting. That's not the funnel you wanna be in. You wanna ... Everybody has a f- uh, you know, in your email, you've got your, you know, we use Superhuman, so they, we have our splits and, you know, I have my one for recruiting, and it's a lot of people looking for jobs and, you know, those all go right to my, uh, recruiting team. But then there's other ones that are interesting people I wanna have a conversation with.

    4. MM

      Huh.

    5. SM

      And that's, that's the pile you wanna be in. Um-

    6. MM

      I like how you visually presented it to me right now.

    7. SM

      Yeah.

    8. MM

      Because once you're applying for a job, you're ending up in a folder that's full of other

  4. 2:56 – 4:26

    Why "learn every AI tool" is the wrong advice

    1. MM

      candidates.

    2. SM

      I mean, I have, I have an auto-split for it. Superhuman allows you to write AI-based prompts for auto-organizing your emails, and so the... I have one that explicitly takes everybody looking for a job and puts them in, in a, in a corner. And I go through them and I read them, but, you know, all I'm doing is directing where, where they, where they might go. I'm not really thinking, "Is this an interesting person that I wanna come spend time with and, and learn something?" And then unexpectedly, it might lead to, lead to an interesting opportunity. The second thing I'd say is, I think it's fa- fairly generic advice that everybody should learn every AI tool that, that you possibly can. I think the more interesting advice is you learn to be a manager. This is tricky 'cause I spent years giving people the opposite advice. Learn to be a proficient executor. The, the, the, the best managers we hire are ones that can do the job themselves and kind of work their way, work their way into management. And I think the world is reversing a little bit, and the, the skill of the how do you write code or how do you, uh, you know, how do you write the specific blog post and so on is very important still. But as we use AI tools, we're put in a position of feeling like, feeling like managers, and I think that's a, that's a, it's a different skill.

    3. MM

      How do you learn it when... In my opinion, becoming a manager, and I've learned this through practice, right? I started by myself. I did, I did a lot of manual things by myself. I learned what's good, what's bad, and then I hired.

    4. SM

      Mm-hmm.

    5. MM

      So this is how I transitioned into becoming a manager. What would you say to people whose job is already, like, manual job is already automated?

    6. SM

      Hmm.

    7. MM

      How do you even learn what's

  5. 4:26 – 4:50

    We're hiring judgment

    1. MM

      good or bad?

    2. SM

      To be honest, that's what we're hiring, right? So we're hiring judgment. I think it's really important that every skill that you learn takes practice, and you have to put yourself in a position to, to practice. I also think every skill you learn is best learned in low-stake scenarios. So one of the things I, I've a, a, a framework for, um, something we call Eigenquestions, which is the art of asking the right question. It's very related to, to

  6. 4:50 – 6:37

    Every skill is best learned in low-stakes places

    1. SM

      what being a manager feels like. I also, I, I wrote a post about it, and a huge section of the post was how do you learn to do this? And one of the things that I think people miss in learning is they, uh, try to apply it in the highest-stake situation they can. And if you think about other things you learn in life, you learn, um, an instrument, you learn to play a sport. Imagine if the only way to learn was to do it, you could only learn a sport and you could only play in, in, you know, broadcast games, or you're trying to learn an instrument and every single practice was actually a recital and everybody was watching and the stakes were super high. It wouldn't make any sense. You gotta, you know, you gotta come down to your basement and play your guitar. You gotta get out on the driveway and, and shoot some hoops. Like, that, that's how you, that's how you learn. And so I think what I advise people is if you're trying to learn the skill, if you're trying to be good at how to do AI-centered design, for example, do it in a side project. Do it in something that's very low stakes. Do it with a friend. Do it in a thing that you can repeat over and over again and get very quick feedback loops on, because then you'll get good at it before it's in that high-stake situation. And I think it's not easy to do 'cause you have to force yourself to... You know, the equivalent of get out on the driveway and shoot hoops is different in every job. But it's become easier now than ever before. As you're learning your skills, find your way to practice, find your partners to practice with, and do it, do it in a lower-stake situation.

    2. MM

      So basically, work on a project by yourself, and then when you're applying for a job, you have something that you can show.

    3. SM

      It might not be by yourself, but don't do it... Don't do it... You don't have to do it in the main work scenario. It's particularly true of skills that are creative, that are judgment, that are, you know, how do you, how do you learn- That form of judgment. Well, it's hard when you're in, when you're in the job, you're in the recital. Every, every meeting has lots of, lots of people in it, and you're the one that's trying to learn. And so for you to be vulnerable and say, "Well, I thought that was pretty good," is, you know, you're gonna get judged. And so you wanna find lower-stakes places

  7. 6:37 – 6:55

    Is the career ladder getting erased?

    1. SM

      to do it.

    2. MM

      So when this happens, when somebody who comes to your company has already some management skills, has some experience starting a side project-

    3. SM

      Mm

    4. MM

      ... what happens to career ladder? When you came in, no experience, did manual things, then you got promoted. Do you think it's getting erased?

    5. SM

      No.

  8. 6:55 – 8:00

    PSHE: how promotions actually work

    1. SM

      Actually, I think... So let me tell you a little about, uh, my view on career ladders. I have a, a framework I use for this. It's called PSHE, problem, solution, how, execution. So you start off as, in this case I'll say product manager, and I'll, I'll describe the others in a minute. You start at the bottom. We hand you a problem. We hand you a solution. We hand you the how. We tell you, "You should go talk to this team. You should write this document," so on. And all you have to do is execute. You just have to do that, do that task. At some point, we hand you a problem, we hand you a solution, you figure out the how. So you figure out how to organize the team. You figure out how to do the milestones. You figure out the, the cadence. At some point, we hand you a problem, and you come back with creative solutions, you know, and ideally solutions that nobody thought of, solutions that are really solve the problem beyond what people expected. And at the very top of this, we hand you a space, and you tell us the problems. "I know you told me to go work on activation, but actually our biggest problem is brand." Or, "I know you told me to go work on sales, but our biggest problem is marketing." That ladder, PSHE, is a very different axis than, than, than scope. And one of the, the leaders, um, is the woman who's running, uh, infrastructure for Google. She went and very mathematically took her whole team and put them on both axes. What came back was really interesting.

  9. 8:00 – 8:36

    The trough of disillusionment

    1. SM

      It came back as an S curve.

    2. MM

      Hmm.

    3. SM

      So what she said is early in people's careers, they mostly moved on the scope axis. You're mostly working at that E level, you know, you're the execution level, and we're just giving you bigger and bigger projects. Later in your career, that happens again. You're at that, that P level, um, and we're handing you spaces, and you're working on bigger and bigger projects. But in the middle, it reverses, and then, then so it looks like an S. And she put, drew a big circle around it, and she called it the trough of disillusionment, and she said, "This is the moment where everybody freaks out." Employees look and say, "I thought you told me the game was scope, and now it's not scope anymore." And managers

  10. 8:36 – 10:08

    AI pushes the entire ladder up

    1. SM

      and promotion committees look, and they say, "Well, these two people actually the same scope, but what matters now is how they do the job." So you can use the exact same technique for engineers, for designers, actually for salespeople, marketers, um, so on. It changes your value system and says, "What are we actually valuing with our more senior employees? What is the ladder guide about?" It's about identifying the right problem, about finding the best solutions, about figuring out how to take that and deliver it, and then how to execute. You think about a world with AI, with AI everywhere, what happens? It actually pushes that entire ladder up. Like it-

    2. MM

      Yeah. It seems like the problem and solution can now partially be handled by AI. But at least that's what I'm seeing-

    3. SM

      Well, I think-

    4. MM

      ... in my company

    5. SM

      ... actually, I would say executing, if you can define a problem, and if you're good enough to find the right frame for the solution, then sometimes... And if you can figure out how you're gonna get to market, AI can help you execute it.

    6. MM

      Hmm.

    7. SM

      Sometimes AI can help you one level up and say, "All right, I, I understand you can build this. Can you help me understand how to get everybody on board? Can you help me understand how to test the market?" So on. Sometimes it can invent solutions, but your a- your ability to judge that is, is in your head, and it can very rarely do something without a good prompt. And so the ability to identify the problem you should be going and working on is almost entirely in, in, in the, in the human's head. I mean, AI can help at every level, so it, it, it's just like having a great thought partner. But the idea of, you know, do the, does the ladder guide go away? If, if anything, what we've done is we've just given everybody this great set of executors.

    8. MM

      So basically, the execution part then, how much of it is getting

  11. 10:08 – 13:18

    Why he doesn't like the word "replaced"

    1. MM

      replaced by AI?

    2. SM

      I don't like the word replaced.

    3. MM

      Mm-hmm.

    4. SM

      Um, 'cause I, I feel like replaced signals a zero-sum game. It's interesting that this particular technology has drawn that word in a way that most, most past technologies have not. Maybe I'll give you an example. When power drills came out, there was a lot of questions about does this change the need for, uh, a construction worker, for a handyman, for so on. And of course, that's not at all what happened because what happened, we decided we're gonna build skyscrapers. Did it replace jobs is one in- is a very interesting way to think about it. The net number of people who worked in construction skyrocketed. I think change is hard. I do think we're going through ... change faster than most people are used to. And I think it's easy to paint a, a boogeyman narrative around any new technology. Amazon came as the end of bookstores, and, uh, you know, uh, we've, we've, we've watched these stories before. And I don't mean to say that there's not gonna be change, but I think the word replace in particular, I don't like that word 'cause I, I feel like it starts with a, uh, a zero-sum frame.

    5. MM

      So what's happening at that execution level then?

    6. SM

      I think you're getting a much broader workforce. What do you do with that workforce when, when all of a sudden you can deploy them? You, you go and build skyscrapers. You go bui- build bigger and bigger things.

    7. MM

      Quick pause here because what Shishir just said is such an important shift. AI is turning young professionals into managers much earlier. Not just managers of people, but managers of workflows, tools, and agents. And the real advantage is not just using AI, but knowing how to build systems around it. And that's why I want to mention today's sponsor, MongoDB. If you're building an AI product right now or even experimenting with agents, you run into one problem very quickly: Where does the agent keep memory? Where does it get context? How does it retrieve the right information without you rebuilding the whole stack? MongoDB is built to be the data layer for AI applications and agents. What I like is that it already fits into the tools builders use every day, Claude, Claude Code, ChatGPT, Codex, Groq Build, Devin, Cursor. So you can connect your data to AI without jumping into separate system, and that matters because a real agent needs current context, reliable retrieval, memory, and state. It needs to understand what's happening in your app right now, not answer from a stale copy of your data from three days ago. With MongoDB Atlas, you can keep your operational data, embeddings, retrieval, and context in one platform so you can build rack apps and AI agents without adding a bunch of extra infrastructure or locking yourself into a completely new stack. And if you want to learn it hands-on, MongoDB also has the Atlas Learning Hub with free courses and skill badges, including vector search fundamentals. You can spin up a free cluster and start testing today. If you want to be the person who doesn't just use AI, but actually knows how to build with it, this is a good place to start. The link is in the description. Big thanks to MongoDB for sponsoring this video. And now let's get back to our conversation with Shishir. Tell me about the progression from execution to problem-

    8. SM

      Yeah

    9. MM

      ... in terms of a career. This is something that's already happening, right? The, the

  12. 13:18 – 15:42

    What happens to entry-level careers

    1. MM

      careers have diminished for people from, I think, it's 18 to 25-

    2. SM

      Hmm

    3. MM

      ... beginners in their career. And sometimes, you know, I am hiring someone who is just starting, and they're a social media manager, and all they can do is can come up with stories, but I don't want them to just be able to draft a story. I want the strategy.

    4. SM

      Hmm.

    5. MM

      'Cause I already have Claude, right?

    6. SM

      Hmm.

    7. MM

      And it can do so many things. What would you say to those people then?

    8. SM

      I think you have to adapt. I mean, I think every industry change is, um, [lips smack] leads to a change in roles. I mean, if I go back to the construction a- analogy, being the person who can s- you know, run a screwdriver really fast stopped being important once power drills came out. Now, there's a different set of skills, and so you have to retrain. And I, and I do agree with you. It's happening so quickly that sometimes, sometimes people haven't adjusted. I think marketing, the example you gave, is a really interesting one. This change in marketing is large, but if you go back in time, uh, 20 years, this is... You know, marketing has dramatically changed in, uh, in, in the last, uh... You know, if, if you go to 50 years, you know, think about marketing in the Mad Men days. Uh, you know, marketing at the time was you, uh, you would hire an agency. They would work with a very small number of companies. They would build one ad-

    9. MM

      Mm-hmm

    10. SM

      ... would last an entire year with one jingle-

    11. MM

      Wow

    12. SM

      ... [laughs] one perfectly crafted thing.

    13. MM

      Yeah.

    14. SM

      And then digital marketing started coming out, and I watched this happen. While we were at Google, we'd watch these firms change over-

    15. MM

      Hmm

    16. SM

      ... uh, how they operated and said, "Hey, you're not building one ad. You're building millions." And now the type of people they hire completely changed, and it's people that were systems thinkers, people that had a different approach to, to, uh, to marketing. But honestly, like, the original skill didn't go away. It just shifted. And you said actually taste in coming up with that great message and that great jingle was actually elevated, but now you have to do it in a way that scales across a million ads instead of-

    17. MM

      Yeah

    18. SM

      ... across one.

    19. MM

      You don't have to be like this genius who comes up with just one good thing a year.

    20. SM

      Mm-hmm.

    21. MM

      You have to be able to produce [laughs] like multiple times a day.

    22. SM

      I actually, I think you need to... Like, if I go back to your, your example of I don't need someone that comes up with stories because I have Claude, actually, I don't think that's true. You need somebody who comes up with great stories because-

    23. MM

      Great stories.

    24. SM

      Because-

    25. MM

      It gives great ideas-

    26. SM

      Because they give the-

    27. MM

      ... not just editing

    28. SM

      ... that's right.

    29. MM

      Yeah.

    30. SM

      They give the great ideas, and Claude can go amplify them.

  13. 15:42 – 16:40

    Eigenquestions: the skill he hires for

    1. MM

      you learn that?

    2. SM

      It's back to where we started. I think you have to practice. I don't think it's an easy thing to learn, but it's a thing that... You know, how do you learn judgment is you practice over and over again. Um, maybe it's worth, uh... I mentioned earlier this, this, um, this technique I call Eigenquestions. So Eigenquestion, it's a made-up word. It comes from a mathematical term, eigenvalues and eigenvectors from linear algebra. Um, so an eigenvector is the most discriminating vector in a multidimensional space. The math doesn't really matter. An Eigenquestion is the most discriminating question in a set of questions. Or another way to say that, it's, it's the question when where- when answered, it will answer the most other questions.

    3. MM

      Hmm.

    4. SM

      The question that was number six on the list is actually the one that answers all the other ones.

    5. MM

      Yeah.

    6. SM

      So this idea became called Eigenquestions, or another way to say this is the hard part is not about, uh, finding the right answer. It's about asking the right question.

    7. MM

      Yeah.

    8. SM

      And so you go back to PSHE. That's what, you know, P versus S versus H versus E. P is the question, S is the answer, H is the, the delivery mechanism, and E is the actual execution.

  14. 16:40 – 20:24

    The teleportation interview question

    1. SM

      The... So it's g- a very connected idea. This technique, Eigenquestions, became a technique that we, uh, trained for, that we recruited for. Since, it's become my number one indicator of someone I wanna work with, someone I wanna hire, is how good are you at Eigenquestions? And so we spend a lot of energy on how do you learn this, and I'm coming back to your-

    2. MM

      Yeah

    3. SM

      ... I'm coming back to your question.

    4. MM

      No, it's fascinating.

    5. SM

      The, um... And it turns out that it's a very learnable skill, but the easiest way to learn Eigenquestions is to do it in safe environments. So I, I'll give you an example. One of my favorite interview questions, I can't use this question anymore because I've, I've talked about it publicly too much, so I can, I can give it away. Uh, but I've, I've probably done hundreds of interviews with this question. So I said, "Okay, a group of scientists have invented a teleportation device. How do you bring it to market?" I've asked this question to engineers, to salespeople, to marketers. Um, doesn't really matter. Usually what happens is people then start asking questions. "Teleportation, what do you mean?" I'll answer a little bit, but mostly I'll say, "Why don't you get all your questions out?" And so they'll start asking questions, and they'll say, "How big is it, and it, does this destroy the person or not? Is it two-sided? Is it, you know, is it fast? Is it slow? Is it expensive? Is it safe?" People ask all sorts of questions. "Is it black? Is it green? Is it..." [laughs] You know, they'll, they'll, they'll ask all sorts of different questions, and I'll just say, "Okay, keep asking your questions. We'll make a list." And then I say, "Okay, uh, hold on a second. These scientists, they are really annoyed with all your questions. You only get to ask two questions."

    6. MM

      Mm-hmm.

    7. SM

      "Which two questions do you, do you ask?" And it's really interesting to watch people go back to that list and say, "Okay, I only get to ask two questions. Which one?" There's lots of good answers. There's no, there's no right answer. But one, one answer that stood out to me was this person ended up drawing a, a two-axis chart and said, "One question I would ask is how safe is it?" And he said, "I think it can... Like, all I really care is is it safe enough for humans or not." That's a really important axis.

    8. MM

      That's the Eigenquestion, right? [laughs]

    9. SM

      That's one.

    10. MM

      If it's not safe, then [laughs]

    11. SM

      And if it's not... Well, actually, it's interesting. If it's not safe enough for humans, does it have a market? And I, I think that's actually a good question, and, and so I'll get to that in a second. And then the second axis was is it more expensive CapEx or OpEx? So CapEx, what, you know, is it more expensive to buy the teleportation units, uh, or is it more expensive to use the, the, the unit? So if you just think about that as a two by two, you get these interesting quadrants. So you say, well, if it's, uh, safe for humans and it's more expensive OpEx than CapEx, meaning that it's pretty cheap to buy but expensive to use, then you wanna put them everywhere. You wanna put them where you put telephones and fax machines. And in every house, in every corner, you never know when you're gonna need one, but we're gonna put a teleportation device there. Okay, let's take another corner. So it's safe enough for humans, but it's more expensive CapEx than OpEx. Well, then you wanna put them where you put airports. Like, we're pretty good-

    12. MM

      Mm-hmm

    13. SM

      ... at finding... And then your go-to-market motion is probably pretty close to how airports get, uh, get funded. You gotta work with-

    14. MM

      That's a brilliant answer

    15. SM

      ... you gotta work through the government.

    16. MM

      Yeah.

    17. SM

      What about the other axis? What if it's not safe enough for humans?

    18. MM

      Mm.

    19. SM

      Right? So what are the cases where it's not safe enough for humans? Um, like, what would you, what would you do with it? And I've, I mean, I've had them... Like, I often will prompt people into that, that category if they, if they pick this axis, and, you know, say, "Oh, actually, maybe what we should do is use it for, um, uh, hauling away trash. Um, I don't really care if it disappears. I don't care if it gets all mangled up."

    20. MM

      Because it's trash anyway.

    21. SM

      "It's gonna get distributed away." Or maybe it's actually things that are really high stakes. So, like, one, one of my favorite answers was, "We're gonna use it to ship organs to, to Africa." You know, people need, uh, need transplants. How do you get organs there? It's actually really hard to get them there. You know, this is the, the... this is maybe a way to get there. So this i- idea of, like, how do you come up with an Eigenquestion is really hard. That particular prompt, the teleportation device, what would you do with it, you can ask a five-year-old that question. You should try it with yours.

    22. MM

      Yeah.

    23. SM

      And you'll get, um, you'll get good answers. I mean, and, and interestingly, sometimes you'll actually get better answers than you get from adults 'cause they'll, like-

    24. MM

      Yeah

    25. SM

      ... they'll maybe even say it in a funny way, say, "Teleportation, do you blow up?"

    26. MM

      [laughs] Yeah.

    27. SM

      [laughs] Uh, but they'll quickly get-

    28. MM

      But that's a great que- 'cause this is the safety issue, blow up.

    29. SM

      Yeah.

    30. MM

      So they're, they're asking the right question [laughs]

  15. 20:24 – 22:10

    How to practice this without risking your job

    1. SM

      that's right. That's, uh, the question, right?

    2. MM

      Yeah.

    3. SM

      So they'll, they may not frame it in the, like-

    4. MM

      Yeah

    5. SM

      ... in the business-y terms-

    6. MM

      Yeah

    7. SM

      ... but they get to the heart of the idea very quickly. This idea of teleportation as an example, I have, I have a list of 100 of these, and you can come up with them. If you wanna get good at asking the right questions, find domains that maybe it's low stakes. Do it as a game. Do it with friends. I think it's important to practice in very low-stakes environments.

    8. MM

      I think you highlighted a very important mind problem that we're all having. We're focused on a problem we want to solve immediately, and we're not thinking about big picture.

    9. SM

      Yeah.

    10. MM

      And the bi- I... What you're talking about reminds me how we were discussing with my husband having a live-in nanny versus, like [laughs] just hiring help by hours.

    11. SM

      Mm-hmm.

    12. MM

      And, like, all my arguments, "But oh, my God, we need help in the morning," but his bigger picture is, like, completely different. Shishir has been hiring for the same thing for 20 years. He ran product for YouTube, founded Coda, and now runs Superhuman, the company behind Grammarly. 40 million people use it every day. Because Shishir is hiring a lot, I thought that pulling out the seven skills he's actually hiring for is an interesting exercise. What do these skills look like, and how to practice them without putting anything at work at risk? It is in this week's Future Proof newsletter. It's free. Link is in the description. Okay, we're working on taste. Can you actually show me some AI workflows-

    13. SM

      Uh-huh

    14. MM

      ... that you're using since you're running an AI company?

    15. SM

      Uh-huh.

    16. MM

      And what my audience is really interested in is not another morning briefing agent-

    17. SM

      [laughs]

    18. MM

      ... 'cause we've, we've heard of them. We know how to set them up.

    19. SM

      I, I get lots of morning briefings. Now I need-

    20. MM

      [laughs]

    21. SM

      ... a morning briefing of the morning briefings.

    22. MM

      [laughs] Exactly, right?

    23. SM

      [laughs]

    24. MM

      'Cause there are so many.

    25. SM

      Yeah.

    26. MM

      But is there something that a lot of people who are watching can set up in their job to make themselves stand out in their company or if they're a solopreneur in their business?

    27. SM

      Yeah. I'm happy

  16. 22:10 – 24:54

    Inside Grammarly: 100 billion AI queries a week

    1. SM

      to share. I think I'll take the excuse to talk my own book, and so I'll, well, I'll talk about a product we're building. I'll just give a little context before give you the demo. So I run a company called Superhuman. We build an AI-native productivity suite made of, uh, multiple different products. We make a really popular mail product, a really popular document product, but the, the product that we're probably best known for is a product called Grammarly. Grammarly is, uh, a very popular product, over 40 million daily active users, you know, does hundreds of millions of dollars in revenue. The interesting thing about, uh, Grammarly is that the core technology of Grammarly is actually not about grammar. So the, the core of Grammarly is, um- About bringing AI to work right where you work

    2. MM

      Mm-hmm

    3. SM

      So two fun stats about Grammarly. One, uh, Grammarly does, uh, over 100 billion LLM queries a week. Works out to per user per day, over 3,000 per user per day. So if you use Grammarly, we are likely your number one generator of LLM queries. It's mind-blowing to people to think about it

    4. MM

      Yeah

    5. SM

      And it's, and it's easy to understand why, because it works at the speed of typing. As you type, we're constantly trying to figure out what are the different suggestions that we can, we can give you right in this moment. And so we have to do it very quickly, and we do it at that, that huge scale. Second interesting stat is Grammarly works in a million unique surfaces a day. It works in every web app, desktop app, and mobile app that you can think of, uh, where we can observe what you're doing. We can annotate it in a way that's unobtrusive to you in the application, and we make... we can make changes on your behalf. We sometimes describe this technology, we call it the AI superhighway. It's we bring AI right to where you work. The interesting part of that analogy is today we-- or up till three weeks ago, we only ran one car on that highway, and that's the one that happened to be, uh, happened to have your high school grammar teacher in it, and it's an incredibly valuable car. Tens of millions of people find it, uh, immensely valuable every single day. Um, but it's a vast underutil-utilization of that infrastructure. So we decided to split the product into two, and so we, uh, took Grammarly and turned it into what we call an agent, and we took the bottom half, the AI superhighway, and we built a new product around it called Superhuman Go. And it's a very simple idea. It allows you to build AI agents that can do everything that the rest of the AI agents can do, including sending you morning briefings. But the unique thing it can do is it can work right where you work. It's like Grammarly, but with your own, uh, knowledge, your own connectors, your own tools, so on, and with your own prompts. If you download Grammarly, you'll get, uh, a, a big G in your extension bar, or you can do-do the same thing with the desktop app, and you can just go and say, "Use Superhuman Go," and it'll flip it over to, uh, to Go mode. And at that point, what used to be a G turns into, uh, the Superhuman logo, which we call Hero, and you get a set of, uh, of agents, and you can go add your own agents, and-

    6. MM

      Are these all yours or they're pre-built?

    7. SM

      These, these are, these are-- I, I have a demo account here, but I, I use a very similar set for, for myself, so-

  17. 24:54 – 28:14

    Demo: building agents that work where you work

    1. MM

      Mm-hmm

    2. SM

      ... uh, but I-- it, it's-- you end up with very personal data. So this is-- but it's very close to what I do. So I'll show you some examples. So one of my, one of my personal favorite agents is, uh, this one. Um, so this is called a, uh, knowledge checker. And so what this is doing-- So this is the agent builder. So you can build an agent that does anything, and, and I think a lot of agent builders are quite similar these days. Uh, you give it a set of, uh, tools that it has access to. You can connect any different tool to it. Uh, we have a, a, a wide set of connectors that'll-- that synchronize data, or you can connect any MCP. You have some instructions on what you want it to do, uh, but the most important part is the triggers. So these triggers are-- You can trigger on anything. Um, you can trigger on a schedule, so that's like you were saying, the morning briefing every, every day. Uh, you-- there's a set of events you can trigger on. Every time I get a Slack, uh, Slack message or so on. But the most important trigger for us is what we call the while writing trigger, um, that says, "While I'm writing, I want you to go and make these types of, uh, suggestions." K- you can pick a-

    3. MM

      Huh

    4. SM

      ... pick a, a color and, uh, and so on.

    5. MM

      Do I understand this correctly, where you're saying when I have this agent on, like, if I'm replying to someone with a summary of another email, like I'm planning a party and somebody send me information about their home-

    6. SM

      Mm-hmm

    7. MM

      ... and I'm replying, uh-

    8. SM

      Yeah

    9. MM

      ... to guests, I can just ask Superhuman to pull that information, like fact-check what I'm saying, like-

    10. SM

      Exactly

    11. MM

      ... putting-- Oh.

    12. SM

      Exactly. So I, I'll show you an example.

    13. MM

      So you don't have to have multiple windows, and like go back and forth.

    14. SM

      Not-- So it's not only just not multiple windows. So I think if you think about some of the use cases here, I'll show you, I'll show you a couple different cases. This is a professional use case. I'm writing a launch announcement. I'm a marketer. So these are-- these underlines look like Grammarly, and some of them are. Uh, this one is literally from Grammarly, but some of them are coming from other agents. So this, this one, for example, is coming from my-- I have a launch project manager agent that tells me that this is not actually, uh, pending legal review. I still need to go deal with the design, uh, design review.

    15. MM

      Oh.

    16. SM

      Uh, or this one might come and, uh, uh, this is my legal guardrail agent that says, "You're not supposed to say it that way. Here's a more correct way to say it."

    17. MM

      Oh, so it checks everything you wrote.

    18. SM

      It checks everything. So you could use it-- Maybe you use it in Gmail, uh, or maybe you use it in, in, in a document. Here's an example where maybe you wanna, uh, use it in an email.

    19. MM

      If I write, "So my revenue for this quarter is this," and if it's connected to whatever it's connected to-

    20. SM

      It'll go fact-check it.

    21. MM

      [gasps]

    22. SM

      Right? And-

    23. MM

      And I don't have to trigger it. It just fact-checks.

    24. SM

      It just fact-checks it. Now, this one, fact-checking is one of my favorite agents where-

    25. MM

      Where does the knowledge come from? So it's fully resolved, meaning, uh, it's a project somewhere?

    26. SM

      Each of the agents has different, uh, has different data sources.

    27. MM

      Mm-hmm.

    28. SM

      Um, [lip smack] and so, you know, for example, that agent I just gave is the source finder agent. It's this one I have here. Um, and I can go see. This one has access to-- In this case, I gave it access to my mail, my calendar, and my docs.

    29. MM

      Wow.

    30. SM

      Um, but you can give it access to anything. Um, and so that'll come through and give you suggestions based off of what's happening with, you know, each of these tickets, or you can-- it'll check a revenue number or, or, or so on. And I think in that way, you can design an agent that feels like-- I mean, one way to think about it is if Grammarly felt like it's magical to have your grammar teacher follow you everywhere. It's like sitting on your shoulder. Who else do you want sitting on your shoulder?

  18. 28:14 – 29:40

    Chat, do, assist — three ways to think about AI

    1. SM

      I think there's sort of three metaphors for how people think about AI. So one is, the most common one is Chat, and that's like this. It's like I, I have this, this, this agent that feels like a human, and it's, it's magical. And so it, uh, uh, obviously captured everybody's imagination. There's lots of products that have a Chat metaphor. Another metaphor is what we call Do. So that's a metaphor of, "I have a task list. I want you to take things off my task list." It's also a really powerful metaphor, and many tools that are, that are focused on that. Uh, our, our view is that there's a third metaphor we call Assist. So you can think about Assist, Chat, do, assist is, I need an agent that comes where I'm working before I ask anything-

    2. MM

      Mm-hmm

    3. SM

      ... and helps me in a way that I didn't expect. So for example, this, you know, would I go check a fact on, on, on a particular email? You might remember to do that, and if you-- like you're saying, you're writing an email to somebody about the, the party, and you put in the wrong address, or you put in the wrong time, or you put in the wrong name. You know, would you remember to take that and put it into one of your chatbots that's connected to the right data and actually get back that reply? Probably not. And that's the reason why we get thou- you know, our users give thousands of queries per day. If you are a really good ChatGPT or Claude user, maybe you would generate 10 queries a day. That would be a lot of, uh, uh, of interaction. But we get thousands. And the reason is because we're there working with you right where you work without you having to ask.

    4. MM

      Yeah.

    5. SM

      Um, so I think it's a very interesting way to rethink AI.

    6. MM

      So what are the top three life-changing agents? The fact checker?

    7. SM

      The fact checker is a, is a, is a really interesting one for all the reasons you

  19. 29:40 – 34:37

    The three agents he uses every day

    1. SM

      already [laughs] you already got to. Um-

    2. MM

      No, I'm over it 'cause this is what I'm doing. Like, I, I have two phones because I'm checking one email and replying to another.

    3. SM

      Yes. That one's amazing. I have another one I use personally called, uh, the placeholder filler, and I have a ... Th- th- these are very personal, by the way, 'cause it ... If you think about, like, you want a assistant that knows you. So mine is I have a writing pattern which is when I'm, uh, when I'm writing, I'll often leave, uh, empty things that I'm, I'll come back to, and I just put them in brackets.

    4. MM

      Mm-hmm.

    5. SM

      And so s- I'll say, you know, I'll be outlining something and I'll say, "We should ... I should start with a, a quote, right, you know, a customer quote about this," and then I'll keep writing and I'll ... I just put it in brackets, and it'll go and find it, and it'll go and try to fill in my placeholders with whatever the best suggestions it can.

    6. MM

      Mm-hmm.

    7. SM

      And I, I give the agent access to all my information, and so it can go-

    8. MM

      Yeah

    9. SM

      ... and, and fill those things in. Um, so that, that's a really productive one for me.

    10. MM

      Can it do online search as well?

    11. SM

      Yeah, anything. Yeah. So it's, yeah, anything you want it to connect it to.

    12. MM

      Mm-hmm.

    13. SM

      I mean, you have to decide what you want to give it access to. Um, you can have these agents be ... They can be collaborative, so you can-

    14. MM

      Yeah

    15. SM

      ... uh, you can share them with other people. You can put them in collaborative spaces. You can-

    16. MM

      Yeah

    17. SM

      ... you can put them in Slack or wherever else, uh, you like as well. Um, so, like, another common use case we see from customers is, uh, customers will use it, uh, for example, for compliance checks. You know, so for, for some companies that's, uh ... We're working right now with a, uh, a major magazine publisher. Actually, they have, they have, uh, a dozen or so magazines, and they have 16 review departments. So if you're a reporter and you're writing a story, you go to the fact checker, you go to the brand people, you go to the, the, um, uh, the citation checker and the source checker. And they, they gave all these different, uh, folks to do. I'm not even sure I know all the departments they go to. They took all of them and turned it into, uh, an assistant that they just deployed, and it works everywhere.

    18. MM

      Yeah.

    19. SM

      And so rather than remembering to go to every department, now your story's slowed down, it's just happening while you're going. And you also get it much earlier. So instead of waiting till the end when, oh, I thought my story was done, and, you know, I, I wrote the whole story based off of this fact that turned out to be wrong-

    20. MM

      Yeah

    21. SM

      ... like, you wanna know that much [laughs] earlier in your process.

    22. MM

      Yeah. I really like that, the, the proactivity and also being where you are-

    23. SM

      Yes

    24. MM

      ... in the moment.

    25. SM

      Yes.

    26. MM

      Anything else? Number three?

    27. SM

      Oh, boy. Um, let's see. Number three. I mean, I think there's a set of use cases. Uh, for me the calendar one is really interesting. So I have it connected to my calendar, and it'll do the, like the example you gave of you're suggesting a time, check that that's accurate. The more interesting ones is it'll come and see if I'm trying to schedule something, and it'll underline it with my availability and with an option to go and book the meeting.

    28. MM

      Mm.

    29. SM

      And so I'll come in and, you know, say, "We should meet tomorrow," and it'll come underline it and say, "Actually, I've checked both your calendars, and tomorrow at 2:00 PM works." Um-

    30. MM

      Well, would it highlight if ... Y- 'cause this is what happens to me all the time. I have two different emails threads where I'm discussing the same date.

  20. 34:37 – 37:20

    Bill Campbell's advice

    1. SM

      Perkins. I got to know Bill at a period where he was in his coach mode, but it was before all the big shots, and so I, you know, I got lucky [laughs] in that, uh, period he's, he's helping out all ... I had started a company that was funded by Kleiner Perkins, and he would hang out and, and help out whatever he saw as the most interesting companies. Immensely helpful person. I mean, he was ... I don't know. I can't overstate how, how much impact he had on me and on the team. Um, you know, almost ... I was a 21-year-old lear- C- first-time CEO learning how to do everything. He told, he taught me how to run my first staff meeting. He taught me, first time I had to fire someone, he walked me through this is how your talk track is gonna work, how to deal with my investors, how to hire people. Like, it was so many different pieces. The most memorable conversation with Bill was he ... I'd been working with him for about a year And I came to him and I, I said, "Hey Bill, I just realized that, uh, we haven't done an advisor agreement with you." And you know, I felt really bad because, you know, uh, we had other advisors at the company and, you know, we had given them some cases, uh, some compensation, some equity or, or, or so on. And you've been, you know, immensely valuable, and I don't wanna, I don't wanna take advantage of that, so I'm happy to put something in place. And he said, um, "Shishir, don't worry, I don't need it." I was offended, [laughs] to be honest. It was, it was the, you know, "What's wrong with my equity?"

    2. MM

      [laughs]

    3. SM

      [laughs] And, uh, and he said, "No, no, I didn't mean it that way." He says, "You know, I've been lucky through my life. At this point, everything I get goes right to charity. So if you wanna give it to me, it's fine, but you could also just give it to charity and that, that, that's okay, too. But honestly, you should hold onto it and use it to grow the company, 'cause it's not, not the, not the way you should use it right now." Um, and I said, "Well, that's really interesting, Bill. The- it feels like I'm surrounded by people that all want a little piece of the company, and you're like the only one that's put- adding so much value, and that- that somehow that's not b- that's not what you're- what's motivating you. So what motivates you?" And he gave me this really interesting answer, and he said, "Well, I just look at all the people that, um, uh, that have worked for me or that I've mentored, and I just make a list of how many of them are Fortune 500 CEOs, and that's my bar." This was back in 2000.

    4. MM

      Mm.

    5. SM

      So there's, you know, since then it's, you know, Larry Page and Steve Jobs and Jeff Bezos and so on. But in 2000, he already had a list of 20 people.

    6. MM

      Wow.

    7. SM

      And they ca- and he's just working through this list.

    8. MM

      Mm.

    9. SM

      And you could just see this sense of finding success in other people's success.

    10. MM

      Yeah.

    11. SM

      It was such a different way of thinking about-

    12. MM

      Mm

    13. SM

      ... uh, about the world. And it completely changed my worldview of, you know, zero-sum games, what are you... A- and I'm sure everybody's had that experience of somebody, somebody you love is working for you and decides to go on to do something else. And Bill taught me, if you're gonna be,

  21. 37:20 – 38:48

    The metric he still measures himself by

    1. SM

      uh, a good partner, a good manager, a good leader, you need to root for your people's success. You need to judge your success that way.

    2. MM

      Yeah.

    3. SM

      It's actually, it, it- that's how, that's how it carries on. And I, it was interesting, when Bill passed, they held his, uh, funeral at, uh, Sacred Heart, um, you know, up the road here, and, um, every single person who came up to the stage, uh, had almost the identical story. "He had infinite time for me, and he only cared about me being successful, and as far as I could tell, you know, that, that's, that's all that mattered to him." And every person looked around and said, "How do you have time for you?" And how did, how [laughs] ...

    4. MM

      [laughs]

    5. SM

      And how did he... H- you know, h- it could just seem like he was in your corner rooting for you in a way that was, that was very different, and so, you know, very much changed my way of thinking about interacting with people.

    6. MM

      This is such a great metric-

    7. SM

      Mm-hmm

    8. MM

      ... to optimize for.

    9. SM

      Yeah.

    10. MM

      'Cause we're always looking at numbers. Th- this could be a number-

    11. SM

      Yeah

    12. MM

      ... a number of people.

    13. SM

      It is a number. Oh, he was very numeric about it.

    14. MM

      Yeah.

    15. SM

      I mean, he was not a... You know, he was a sales guy. I mean-

    16. MM

      Yeah [laughs]

    17. SM

      ... he, he had, he had a c- he had a metric, as it... And he had, he had the list. He knew them by heart.

    18. MM

      The treasure. Yeah.

    19. SM

      Yeah.

    20. MM

      That's amazing.

    21. SM

      Yeah.

    22. MM

      I think, um, everyone who's watching this-

    23. SM

      Yeah

    24. MM

      ... try, if you don't have this metric yet, think about it.

    25. SM

      Yeah.

    26. MM

      Like, how do you measure people who become successful thanks to your work?

    27. SM

      Yeah.

    28. MM

      That's awesome. Thank you so much.

    29. SM

      Thank you.

    30. MM

      Thank you. If Shishir made you rethink what are you being promoted for, watch my conversation with Ryan Roslansky. He was the CEO of LinkedIn at the time of the interview, and Ryan will tell you exactly which jobs are appearing and what companies are looking for right now in the age of AI.

Episode duration: 38:48

Install uListen for AI-powered chat & search across the full episode — Get Full Transcript

Transcript of episode JEZVl3BkSLk

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

iOSAndroidClaudeChrome