Grammarly CEO Shishir Mehrotra on AI Agents, Productivity & the Future of Work | AI Rabbit Hole 2025

All right. Which which chairs do we take? Which chair should we take? Let me start with these two. Okay. Hi everybody. Thank you all so much for being here today so that we can talk about the future of AI and work with I think one of the sharpest speakers and thinkers in the entire world when it comes to how we apply AI in our daily life, how we use it to ex accentuate our efficiency and how do you manage an amazing team in this very fastchanging era. So please give me a give a warm welcome to the former director of product management for Microsoft, former CTO, chief product officer of YouTube, co-founder of Kota, and now the CEO of Grammarly, Shishir Mehrotra. Thank you. Now Grammarly excited to be in Alice in Wonderland. It's a deeper and deeper. So, Grammarly has 50,000 organizations using it, over 40 million people, and it actually it saved my life.

Yeah. Well, it didn't like pull me out of a car wreck, but you I was the editor at large for TechCrunch for 10 years, wrote 4,000 articles, and it routinely saved me from typos in my headlines that would then get tweeted to 10 million people. So, like for my own career, thank you for saving me. Yeah, that's good. But really today I want to break down some of your amazing frameworks about how to think about the massive change happening in the AI world. And so I want to throw it to you first. What is how is productivity changing and how is it evolved from you know our early era of just trying to digitize what we were doing to where we are now? Sure. hi everyone great to be here. so let's starting with productivity. I I think we're in what I call the third era of productivity.

So simply put, the first era was digitization. So we took typewriters, we turned them into word processors. the second era was about collaboration, which was basically about working with other humans. that was when tools like Google Docs and Slack came out. the the easiest way to visualize products from the second era of productivity is avatars showed up. So you could see who you were working with. And the third era of productivity, I think, is when we're going to move from working with humans to working with humans and agents. And a simple visual of that is it's going to feel like I have a bunch of new collaborators everywhere I go. And instead of sitting and working with my four co-workers, I'm also working with 20 agents as well. And I think that's a a helpful way to characterize where where we're headed now. Yeah, I was thinking that we're kind of changing the definition of IC or individual contributor to being intelligence coordinator.

you are can now finally scale your output even if you're not an amazing hirer or fundraiser or human manager you can still do more than you ever could before and like meanwhile it seems like if you're a manager you know the ability to do rapid fire brainstorming research is powerful I would love to hear just sort of how are you using this in your daily life are you doing a lot of deep research report brainstorming and sort of using that as a a sparring partner and then going to make decisions with your team or how do you use it differently than you use humans Yeah, maybe starting with your with the analogy which I think is a an important one and a sharp one to to think about. you know, my I have a I have a daughter. It's she's a freshman in college. she's a math and CS major and she called me and said, you know, maybe I should focus on math.

It kind of seems like all the engineers are getting replaced by AI. And so, we spent some time talking about your first observation, which is I I don't actually think that's what's happening. I think what we're doing is we're scaling people from IC's into leaders, IC's into managers. And I told her what you should expect is that when you show up at your job, it's going to feel like instead of starting at the bottom of an organization, you start as a manager and you got this team of a few hundred people. And there's some really interesting things about that team, right? The they're they're very capable. They've read every CS book that's ever been written. They've read every line of code that you that your company has. They're very productive. They're very fast. they're very they can be very collaborative with each other. they're also a little bit simple.

They do exactly what you say. they they they will collaborate but on the other hand and they'll never fight with each other but it also means that they'll never conflict. They'll never grow with each other. It means it's very important how you how you direct them. And I think that view of thinking about AI as scaling us through a much broader team is a helpful way to characterize what what we're going through. So I find a lot of what I'm doing is kind of similar. It's I can do things at scale that I used to not really be able to do. I can now, you know, I used to write just as an easy example. the the my sales teams or so on would ask me to write a handful of emails to customers every week. We'd go back and forth. You know, maybe maybe I'd get a few out a month.

I think in the last month I sent out a few hundred. and it just scaled in a totally different way. It's easier to write them. It's easier to find the opportunities. It's easier all of those pieces have turned into things that AI agents can do on my behalf. But it also means that I have to think and scale like a manager in things that used to feel very individual. So that's my my frame for how how to think about it. So I've always thought about AI being either a journey or a destination. And I felt like we've been trapped in this mode of seeing AI as a chat box that you go and paste things into and then copy things out of. And to me that misses the point of it being along for the journey wherever you are. And that's why I was always so excited about Grammarly because you're not just, you know, a doc somewhere.

You are a mobile keyboard, a desktop, you know, software, a Chrome extension. And so you follow me wherever I am. And so no matter where I try to make a mistake, you're there to save me. But I and when I joined SignalFire 5 years ago to be able to lead the consumer investor vertical, I was like, "Wow, amazing. Grammarly is already in my portfolio. I I look really smart." Even though the investment happened years before I joined. But it seems now with the acquisition of KOD and you coming on through leadership, you're not really choosing between being the journey or the destination. You're kind of trying to do both. Yeah. So may maybe I I can break both those parts apart. So the there's two things we're doing with with Kod and Grammarly. one we sometimes characterize as agents, the other we characterize as applications.

And and there's a sort of an analogy that goes with each one. So maybe I'll I'll start with the the first observation you gave. you know, I I think our view at Grammarly is that we're the OG agent. You know, in many definitions of the of the word, we are for the last 15 years or so, we help about 40 million people a day communicate. And the way we do it is somewhat unique. We do it where they already work. It's not a thing you go to. It's not a thing you have to remember to go to. It feels like a collaborator working right next to you. Now, if you look at the technology of what Grammarly is, there's a there's a meaningful part that's about being very good at grammar, but actually most of the technology we built is about how to run AI at the edge.

Run AI right where where users work. that allows us to work with about 500,000 applications that's websites, desktop applications, mobile applications, so on where we can reliably read what the user is doing, annotate it in an unobtrusive way and then edit, correct, and take action on their behalf. The one analogy we use for it is it's like we built an AI superighway to bring AI right to where right to the end user right to their existing applications. But today we only run one car on that highway. That's the car with your high school grammar teacher in it. And it's a really useful car. There's a reason why, you know, it's 40 million users and over $700 million in revenue. But the but that car is only one of many that could be helpful to you. And so one big piece of what we're doing is turning that superighway into a single car highway into a multi-car highway.

And we're going to open it up so many agents can run on that platform. And just just so you can picture it, imagine that as you're writing your memo for a customer, writing an email, you might be you want your grammar teacher to to help you with make sure you're communicating in as effective way as possible and you got the grammar correct and you're getting the right sentiments across, but you also want the agent that knows everything in your sales data to make sure you're giving correct facts to that customer. You want the the support agent to tell you, did you know this company had an outage recently and you should probably talk about that? or maybe your product marketing agent should help you with get the exact right messaging for how you're supposed to be positioning your your products. So this is the first part of what we're doing is turning you can think about as turning Grammarly from an agent into an agent platform or you can think about it as we're taking this AI superighway and opening it up so anybody can run cars on it and I think that's going to be one important piece.

Did you just launch a developer platform? It's coming. Yeah. Yeah. So I I think a big piece of that the the the core idea of that is why we combine the companies is is a technology we have in KOD called KOD packs and that's the basis of how we're building out this agent platform. The second half of what we're doing is is what you refer to as journey versus destination. One one way when I started talking to the Grammarly team about putting the companies together Max and Alex the founders of Grammarly they they summarize it as Grammarly is it's like a moat without a castle. It works everywhere which is really amazing but it actually lacks a meaningful core destination where people can do their their real work with a trusted firstparty experience. back to my my experience before starting KOD I used to run the YouTube group at at Google and to use an analogy it's imagine if YouTube only had embeds and there was no YouTube.com.

It's kind of what Grammarly feels like. It's like everywhere but that there's no home. And so the other half of what we're doing is we're going to turn the Koda doc surface into a firstparty native AI writing experience where which is the best place to work with the agents. So we'll we'll work everywhere pretty well, but we'll work the best in that in that spot. So those are the sort of two halves of what we're doing and I think it's I think they're really important to add more cars to the highway add the castle for that for that moat. I do love the idea of like being able to find a writing surface where the agents do the work that would otherwise distract me from my writing. Because as soon as you open another tab to go look up some information, you just get sucked into the joy of curiosity of exploring the internet and suddenly you've spent 15 minutes that you needed to be getting done, you know, what you were actually working on just learning about something new, which is great for self-development, but at the same time maybe not great for productivity.

And so, yeah, I would love to hear more about what does this writing surface feel like for for for people because I think everyone's been using chatbots, whether it's okay, let me oneshot an entire draft of something, then let me regenerate specific paragraphs, but I have been surprised that we haven't seen more of like generate a bunch of drafts in parallel and let me pick and choose the best pieces from them and combine them. Feel like we're still kind of in the the old framework of what does writing mean and it would just be way too much work to write the whole thing 10 different times and then combine them. you just do it more efficiently writing it once. But now with AI, you don't have to worry about that kind of constraint. Yeah, you mentioned a couple different things that are really important. And if you stick with my frame of I think of agents as as collaborators, think about the best collaborators you can work with.

If you're, you know, writing an article, if you're if you're writing a customer memo, if you're a student writing an essay, if you're writing a blog post, what whatever it might be, what do you want that collaborator to do with you? And, you know, certainly you want to take you want to take one output and make it better. But isn't it amazing if you can have that collaborator work with you at multiple phase of that journey? So help me with an outline and have have it change the draft as I go. Like that that modality is not an easy one. sometimes you want lots of variations. I mean back to my my daughter she's a freshman now. So she applied to college last year and I don't know if anybody here has have has kids going through that process. She had to write 150 essays to for her college applications.

And you know they're a little bit different. they're a little bit same and and so in each case you would write one and then she say well you know Colombia wants one that's like 300 words shorter and covers this particular thing that Colombia cares about but it's close that ability to have collaborators take and draft things in different ways is really interesting but it also changes the the the breadth of what you can do today I think when you're writing you try to come up with like one great outline and then you like commit to it and then you write the best possible draft of it and I think at writing with AI feels much more divergent. It feels much more like I got these kind of 10 different ideas. Maybe all I know about them is the headline and I don't really know exactly where they're going to head and I can't quite picture how it's going to turn out.

So maybe what I'd like is I'd like my agents to go write, you know, take these 10 versions and write 10 versions of each of those and and see what comes back and oh that one actually turned out really good. And maybe maybe I want something that that has a different analogy and I want to take the best part of this one and I want to merge it with that one. And I think that idea of feeling like you're in a writing surface where your agents can scale your work in an organized way. I think that's the heart of what we're trying to do there. So when you think of what you've learned from this, you know, creating this super highway but only having one car on it and now wanting to put more cars on that road, whether it's the, you know, the idea of the castle and the moat, you know, what would you say?

What what advice would you have for early stage founders out there trying to find their place in an extremely crowded and fastmoving environment? Should you say, "Hey, let me build race cars for everybody building highways." Is it, "No, I have to own the highway, otherwise eventually I just become, you know, I get I'm the dumb piece within a smart pipe." You know, what how do you navigate this if you're on the early side? You know, you guys are are giant and have the ability to, you know, leverage what you've already built to go out and build the highway. But for everybody else, what should they think about? Yeah, may maybe I'll give I'll give an example. I've had a chance to talk to a lot of different and learn a bit about how what what goals people want to accomplish. And no, I don't think the choice is just like build highways or build cars.

I mean, that's that's certainly how we look at the world for what we're building, but everybody has a slightly different frame for what they're trying to solve. maybe I'll give one example that that really stuck with me. I was talking with the team at Duolingo with Luis and his and his team, and I don't know exactly how this will play out, but it was a great thought thoughtprovoking exercise. And so I gave him this analogy, you know, I think that there's going to be this AI superighway and we can build cars for it. He said, great. So what's a Dolingo car? And we ended up coming up with a few different pieces of what what would that mean? You know, one, imagine if while you're learning a language, imagine if you could keep your streak alive anywhere you go. I'm learning Spanish and I don't have to do it in the app.

I can do it just by reading articles in Spanish, by working with a customer who speaks in Spanish or whatever that might be. Second, imagine if your coach, if your agent could help you wherever you are. And so I'm reading in English, it knows my Spanish reading level. It goes and replaces all the words I should know in Spanish with the Spanish equivalent. So I read that way. Or maybe the other way around. I'm reading something in Spanish. It's a little bit daunting. It knows my level. It goes and takes all the things I don't know, turns those into English, but it lets me go through it. Don't use Google Translate. Be my coach. That lets me slowly get there while I'm working. And then when I come back to that application, you imagine it knew what I was doing. So today's lesson is now today's lesson is how to how to give a talk at a at a conference or how to buy a house or how to hire a recruiter because I know the lesson can now be trained to what you're actually trying to get done.

So in that world I think I don't think the world is just you build agents or you build super highways and so on. I think it's you're trying to solve a problem and and from for every founders's perspective you still have your problem you're trying to solve. I mean Dolingo's job is to help people learn languages much better and faster than any other way to do it. But your opportunity is now different now. Now you can be with them everywhere they are. You can interact with them in every surface they're in. You can scale a personalized teacher that goes with them instead of having a generic teacher that sits with them everywhere they are. And you can and you can change based on what the person's doing. So I think that that way of thinking about is is more helpful for for the opportunities for each for each application.

So maybe it's just like allowing a sense of flexibility about what your front end looks like that you know especially in an era where operations and even you know engineering becomes more simplified and distribution and brand kind of becomes everything. Where can you find a new way to take the same value you've already built and distribute it through a new surface or in a new form factor which brings you into that journey rather than only thinking of yourself as a destination. Yeah. And I I would think of it as like instead of building destinations, I think we what we're really building is virtual people. And every application we have is is something that I interact with. And I know a lot of applications are focused on being agentic and they interpret that as I should be able to chat with my application. And that that is definitely a minimum bar.

I mean, you would not call someone a human if you couldn't talk to them. But the idea that that's the only way we define agents is is too too limiting. I mean, the fact that it it can follow me wherever I go, it can know what I'm doing. It can remember what I've done before. Like these are all parts of what it means for an application to feel agentic. The best in technology becomes invisible when so hopefully we can see more of the AI moving in that direction. So I've always thought of you as just like this walking set of frameworks like every time you open your mouth you're just like giving a fully cons you know architected essay and you know some of my favorites include you know your thoughts on Mark Twain's famous quote about eating your frogs your thoughts around the eigenquestion or the one question that really like is you know has the most potential impact.

maybe just run us through a few of your frameworks as quickly as you can and what has changed or been proven true in the new era of AI. okay. So, let's see. Frameworks that I think are interesting. you mentioned a couple. The the the frog one is definitely not mine. That's Mark Twain. I guess technically it's not even Mark Twain, but it's attributed to Mark Twain. that one's really simple. It says you should start every day by eating eating a frog first, which is really code for do your hardest task first thing in the day because if you do that first then the day is only going to get better. the iing question one I think is a very powerful one. I find I use that all the time. an eigenquestion is a is a madeup word for the most discerning question in a set of questions.

One way to think about it is if you I'm sure you've all had a a discussion where you've got 10 questions to answer and they're ranked in some way. They're ranked by you know which one is most urgent or which one is most has the biggest impact. But often there's a question in the middle of that list. Maybe it's question number six that if you answer that question, it answers all the other questions as well. So we call that eigenquestion. It's kind of similar to IGEN vectors as the most discerning the most discriminating vector in a multi-dimensional space. but that's a really powerful technique. maybe two others that I would if if you wanted to read other things. I think I find we spent a lot of time talking about decision-m and I think the idea of how to build great rituals for decision-m is really interesting that I gave a talk at Lenny Summit last year on that framework.

How do you ask the right question, get the right people in the room, structure discussion the right way, and create the right forums for I think that's particularly relevant in a world where we're about to see this explosion of products. this came out of my time at YouTube where I think for a long time my view of how to do bundling of products just felt felt evil. Didn't feel didn't feel like it was a very productive way to think about pricing and packaging. And I ended up writing this paper. It's called four myths of bundling. that was basically an argument with myself and all the dumb things I used to say about bundling that actually turned out to be the reverse of true. And it's really important in a world where our products are all interrelating with each other much more. and to just give an example, we're going to be hiring agents to work on things.

You know, is that a world of like lots of choices where I'm going to have to like choose everywhere I go. I'm gonna have to pay each one as I go and so on. Or is it going to be a world where it feels like I I kind of buy from a set of people that pull together things you'll use the see the terminology in the paper that I may be a super fan or a casual fan of and that the the best products mix those things together with easy easy pricing that allows customers to experience value for multiple products that they may not have thought out with individual choices. So it sounds like, you know, in this age of AI where suddenly we have so much agency, we can do so much. Knowing what to do and making the right decision about that is what really matters.

And so to quickly recap, you know, we talked about the eras of productivity. We started with digitization, typewriters to word processors. Then the collaboration era was era 2 with Google Docs and Slack. Now we're in the era of agents where suddenly you are the manager. And so rather than maybe thinking of us all as individual contributors who just have this these other agents, we all kind of need to learn how to be great managers in this age because that's what gives us the scaled potential for output. And you talked about the information you know the the superighway that you guys have built and that you know with these big platforms rather than just thinking about hey we built all of this distribution let's just run one car on it how can you run more cars build you know an even a road that no one can live without.

And I think the then you talked about the castle and moat strategy. Similarly, it's like if you can take your product and make it be part of the journey wherever people are going. That's really useful, but you still need some deep experience, some deeper surface, which is going to actually help them convert eventually. And then, you know, we talked a little bit about your top frameworks with eat your frog. It's like when you can do anything, if you do the most important thing first, you're just going to get a lot more done, you know. And I think finally the thing that excites me most is this concept of the eigenquestion in the modern age where like if you can do anything if you can figure out if you can apply this technology to do so much so much faster. It's not a northstar metric that you're measuring.

It's having the right question to ask. So thank you for always helping us ask the right question. What a wonderful summary. Sure. Thank you so much. You're welcome. Thank you.