[Highlight] Grammarly's Evolution into an AI Agent Platform with Shishir Mehrotra

0:00Hey, Ben Kesnoga here, co-founder and general partner at Village Global, a network-native venture firm. What you're about to hear is a clip from a longer conversation. If you'd like to listen to the full-length version, find the link in the description. And of course, subscribe so you don't miss the next one. Enjoy.

0:21Sasheer, great to have you. You are a longtime legendary product leader in the Valley, epic run at YouTube and Google, obviously founder of Coda and now running Grammarly, which is one of the best kept secrets in Silicon Valley in terms of scale and impact and revenue trajectory. And so we're really excited to have your insights today. Just to ground us, tell us Grammarly and Coda recently merged, came together. Tell us what the business currently is doing, its current level of scale and what the next couple of years look like. and then we'll get into some questions for everyone's interest in leadership and management and hiring, product, pricing, and of course, AI and how you're thinking about it.

1:00But first, ground us. Yeah, I'll tell a little bit of the story of what happened, and they'll explain a little bit how to contextualize some of the stats and so on. So as Ben mentioned, I started Coda about 10 years ago. Before that, I was at Google for a long time. I used to run the YouTube group there. Coda, we grew really well, got to a few million users, about 50 ,000 teams, a bunch of big customers. New York Times, Amazon, Snap, a bunch of actually one of my favorite Coda customers is The Sphere in Las Vegas runs on Coda. Who's all the Backstreet Boys? Who wanted to see the Backstreet Boys?

1:40A lot of FOMO. They're doing Wizard of Oz now. They're doing Wizard of Oz? That's apparently remastered and done very well. Planned that whole thing in Coda. Anyways, Coda was on a good journey, and one day I got a call from one of my investors, a guy named Hemantaneja from General Catalyst, and he says, I'd like to introduce you to the founders of Grammarly, and you should talk to them before we were about to raise a round of financing. You should talk to them first. It's one of those calls where it wasn't really an ask. It was like, you should go talk to them. But I said, I don't quite get it.

2:17Grammarly and Coda didn't feel like they had that much to do with each other. And he said, no, no, take the call. You'll enjoy it. So I flew up to Vancouver, spent a couple days with Max and Alex, the co-com. The company was based in Canada? No, the founders are. The company's based here. Well, the company was founded in Ukraine 16 years ago. So I have 150 people in Kyiv. We're actually, Ukraine is to Kyiv what Spotify is to Stockholm. It's like the most desirable tech employer in the city. It's like anything we do makes countrywide news in Ukraine. It also means I have to be very up on what's happening with the war and what the latest things that politicians are doing and so on.

2:58Anyway, but the founders now live in Vancouver. So I went up there and I learned a bunch about Grammarly. And as Ben mentioned, I think Grammarly is one of the best kept secrets in the Valley because it's a business that's very easy to underestimate. So this is about a year ago now, but at the time said Grammarly was serving about 40 million daily active users, a lot of people. We were at the time doing about 700 million a year or so in revenue. We've crossed 800 cents. It's a much bigger scale that way than people think. And maybe one other fun stat is Grammarly does about 100 billion LLM calls a week.

3:36So it's actually a much bigger AI platform than people think. But they said there's one big misconception about Grammarly. The misconception is people think it's about grammar. And that seems like an easy mistake to make. It's in the name, although not for much longer. We can talk about naming as well. We'll talk about naming and branding soon. But they described to me what the technology of Grammarly is, and it was a little bit different than how I thought about it. So the actual thing that Grammarly has built is a technical architecture for bringing AI right to where people work. Now, we didn't call it AI when that happened.

4:15And the team has sort of thought about the whole thing differently. But the view is that Grammarly built the first agent, the original agent. And if you think about it that way, what Grammarly does is it works in about 500 ,000 or so different applications. So every web app, desktop app, mobile app that you use, everything from Slack to Gmail to Salesforce to your performance review app to anything, Grammarly can read your screen, annotate it in a way that is unobtrusive to anything else you're doing or the application is doing, and it can make changes on your behalf anywhere all day long. And so the team inside likes to call this the AI superhighway.

5:00Most of the hard technical work has been how do you do that level of mostly front-end integration with all those different products, each one done a little bit differently. Happy to tell stories about that as well. So they call this the AI superhighway, but there's one big problem. We only run one car on that highway. So that car is the one with your high school grammar teacher in it. Turns out to be a really valuable car. and probably not the agent. I'm guessing half the companies here are doing something with agents. Probably not the agent you started with of how do I get the high school grammar teacher to go do something.

5:31It turns out to be really, really valuable, especially when your grammar teacher can sit right next to you and help make sure you spell your company's name properly or do hyphens and dashes correctly. And he said, what would it look like if we opened this up as a platform for anybody to build agents? And so just to give you an example, imagine you were sitting down to write an email to a customer and you had a set of agents working with you and one of them came in and said this is what's wrong with your grammar or your hyphens or whatever it might be another one said uh hey you've misquoted you know another one knows your sales system and says you misquoted the price on this product another one knows your knows your support system and says this customer actually had an outage last week you should probably say something about that Another one is your virtual EA and says, you said you can meet Tuesday at 7 p.m., but you actually can't do that.

6:21You have your daughter's reception at that time. Are these agents that Grammarly is building for customers? You're saying customers are building on top of your platform. So the idea is to open up the platform to anybody. And, you know, from that perspective, Grammarly starts to feel, and part of the reason they wanted to talk to me about it was I'll talk about the Coda piece in a minute, but it starts to feel a little bit closer to what we built at YouTube. It's an open platform where a big piece of what we do is get distribution for your agents. And if you think about most companies, they build or buy a bunch of AI software that is very hard to get people to use.

6:56I'll give you one fun contextual stat on this. I mentioned earlier we do 100 billion LLM calls a week across about 40 million daily active users. That works out to a few thousand calls a day. If you just sort of contextualize that and look at other scaled AI players, in my mind, the horizontal AI players roughly break into three buckets, what we call Assist, Chat, and Do. So there's a group that maybe I'll start with Chat. The most iconic chat company out there is, of course, ChatGPT. That's one of the fastest growing companies in history. If you use ChatGPT a dozen times in a day, you'd be a pretty heavy user.

7:35There's another bucket I call Do. Those are all the people building agents and AI that goes off and does things in a headless way. I think the Anthropic team mentioned that something like 39 % of their query load now is for long-running tasks, much of that to cloud code. That's really exciting, too. Again, if you can get to a dozen things in a day, it'd be a lot. Grammarly gets a couple thousand chances of bringing AI to you every single day. We do it every time you press a key. We do it every time you load a document, every time you load a web page, every time you open an app on your phone. we're trying to figure out what is the best way to bring AI to you.

8:09And so that's sort of the heart of what we're doing. Along the way, we are also picking off a few. So we're building this platform for agents. We're building some of those agents ourselves. So Grammarly becomes one of those agents. But we're opening that up broadly so anybody can build them. That'll happen later this year. And what does that look like? Anyone can build on top of Grammarly. What does that look like at a product level? Yeah, I mean, I think there's lots of different examples. Maybe I'll give a couple that I think are interesting. One, I'll pick a consumer application. So I had a really fun meeting with Luis, who's the CEO of Duolingo.

8:48People know Duolingo. It's a really popular translation service and application. And we were chatting about Grammarly and AI Superhighway and what the future of agents looks like. And he's kind of intrigued, but he kind of ended that part of the conversation. said, that's really cool. Probably doesn't apply to us. We build apps. Somebody else will build agents. And I said, no, no, no, I think it actually applies to you as well. And so we ended up brainstorming for the next hour on what would a Duolingo agent look like? And we ended up coming up with three ideas. The first one was imagine that one of Duolingo's most remarkable features is streaks.

9:28Imagine if you could keep your streak alive just by doing something in that language. And so today I read three articles in Spanish, and I wrote an email in Spanish. Check off today. Anyone using Duolingo? Yeah. And what's your current streak? It's just ended like 83 days. What happened on day 84? I don't know. But maybe you weren't getting credit. Maybe you weren't getting credit. You and I just do the math. Yeah. What language are you trying to learn? I'm learning Spanish. Learning Spanish. Okay. So on day 84, you might have read an article. You didn't get credit. So that was idea number one.

10:05Idea number two. Now imagine that that agent, the Duolingo agent, can not only watch what you're doing, but can actually interact with it. So can come in and say, hey, I see you're trying to read this thing in Spanish. I know you're reading level. You're not going to be able to get through this. I'm going to pre-translate the words you don't know so that you don't have to hit that Google Translate button. And now you should be able to get the rest of this. Or go the other way around. I see you're doing this thing in English. This is a perfect time to give you a little inline lesson. And so I'm going to take the words you do know, and I'm going to translate them into Spanish.

10:35And now you should try to make it through this in that way. And then the third thing was now you come back to Duolingo. You come back to the office, and you're ready to meet your tutor. But this time the tutor actually knows something about you. And so it says, today I was planning a lesson on how to order food at a restaurant, but it looks like what I should really help you do is hire engineers. So today's lesson is going to be how to hire an engineer. So in that way, I think that right now we think about agents lots of different ways, but in this world, we're taking an application and saying, what would it feel like if every application felt like a human that followed you everywhere and could actually interact with you?

11:10I think that view of agents, I think, is really interesting. So that's maybe one example. Yeah, fascinating. Thanks so much for listening to the Village Global Podcast. You can check us out online at villageglobal.vc. We'd love to hear from you, your feedback, your ideas, your inspirations. You can email us at hello at villageglobal.vc.