Seth Rosenberg:
Keith, Henry, thanks for taking some time today. You know, it's been a crazy few years for you guys, you know, building one of the top AI applications in the market with Tome and then now going after, you know what I think we all believe is an even bigger opportunity with an AI CRM. Obviously all three of us have known each other for over 10 years. Tell us a bit about your background, like how you got here, maybe starting with you, Henry.
Henri Liriani:
We met when we were at Facebook. We worked on Facebook Messenger for, I guess for three years together. We were building a pretty interesting platform there, right? We were building a business chat. I think it was some parts, bots, some parts talking to customer service agents on the other end. And then eventually like the sort of early RNN AI stuff with M, right? Where you could have M call a flower shop for you or like –
Seth Rosenberg:
It was actually overpaid people in the Bay Area right behind the scenes.
Henri Liriani:
Yeah. Uh, we were really faking it to make it there. I was a designer there, um, for a few years and then switched to product and then sort of worked on, like core messaging for a while.
Seth Rosenberg:
I remember we had a Zoom call where you were kind of a little bit disenfranchised with whatever you were working on at Meta at the time and looking for something new. Maybe talk about that for, for a second.
Henri Liriani:
And I think you had, you had wanted me to chat with, uh, with Keith.
Seth Rosenberg:
That's right.
Henri Liriani:
Um, and we were talking about visual storytelling and there was like a loose thesis around like what if there was a third to sort of like Notion and Airtable in the new wave of SaaS productivity. I think I had also just come off of tinkering with a bunch of like ideas and wanting to get back into prototyping and building apps.
So we sort of started sketching. I remember drawing these two tiles on this sketch pad. One with an image in it and one with text in it. And was like, what if there was this really simple primitive where we could arrange them and tell any kind of story with them? We left that zoom call and then, uh, felt like, you know, every time we hung out, it felt like we could print years of roadmap in terms of things we wanted to build. So, it just felt very productive.
Seth Rosenberg:
And then Keith, obviously you and I worked together for Facebook Home, the Assistant, and then you were head of product at Citizen. And maybe tell us a little bit about how you got here.
Keith Peiris:
For the 10 or so years before this, I was just spending time trying to build productivity tools for regular people, you know? And the first one, especially very relevant to Lightfield, I worked on this product called Graph Search at Facebook. And you know, it doesn't exist anymore, but it was like technology so that we could find Seth's friends who live in New York who studied CS and speak French.
Uh, and it turned out that you actually needed quite a bit of sophisticated ranking and retrieval technology to get that done there. There, you know, we sort of had the right technology, but maybe the product wasn't quite right. Like humans don't need to run these sorts of searches every day outside of a recruiting context.
You and I worked on Home where we wanted to build this sort of proactive personal assistant. And I remember, the Chief Product Officer at Facebook saying, you can't build a personal assistant for everyone. You know, the personal assistant is the iPhone. It's not an app, you know, and then the iPhone uses Spotify or calendar or mail. The only way to do this is to make it for a specific person, which is I think maybe another foreshadowing of where we are.
And then I worked at Instagram for a bit, learned a lot about the handshaking of strangers speaking to each other on Instagram, and that also feels like related to doing business with people and tracking to CRM.
And then it was building this, I mean, I would call it a personal utility, which was Citizen – it was a safety app. I think that led me to working with you at Greylock.
Seth Rosenberg:
Mm-hmm. And I remember you knocked on my door, we're gonna do a co-working session. And you're like, “Hey, I actually just quit my job.” Like great. Like let's get a – I think we literally bought a whiteboard. It was during the pandemic, and then we kind of both had a captive audience because we were the only people in our bubble.
Keith Peiris:
Yeah, exactly. It was pretty great. And I think, you know, reflecting on that time, as Henry joined us and as Reed was showing us the early makings of GPT three, I think we all had this, you know, almost this thesis that the way that people communicate at work is going to change.
And you know, we weren't sure if that meant like new Word, new PowerPoint, or something completely different, and I think our first thesis was maybe it's new PowerPoint. And then throughout the journey, I think we realized it was actually going to be something completely different.
Henri Liriani:
Yeah. Just to build on that, I think we spent so much time at the beginning on the format of this like visual presentation layer. But as we got better and better at using LLMs, we realized that the main thing that we needed to do was calculate the right content with the available context.
And then you also have the like context calculation problem, which really is super high leverage because if you can know what people are trying to say and suggest ways to say it, I think you ultimately saved them more time.
Seth Rosenberg:
You guys built Tome, which was an AI presentation tool that had its own unique format that was tile based, that was very easy to create, looked beautiful, cloud-based, mobile, mobile optimized, and it was also deeply integrated with AI. And Tome scaled from zero to 20 million users in the first 12 months. It was one of the number one AI applications. And then you decided to pivot. So why?
Henri Liriani:
Going back to that sort of mission, we really wanted to make great business communication tools and uh, I think we found that we were sort of looking for whom we could do the best job for, right?
And we ended up with a few types of personas, right? We had founders, we had salespeople, that were kind of the most retained in the product. And we had a long tail of other types of users, a lot of students and that kind of thing. With founders, we found that we could help most founders do a pretty good job with the pitch deck, but they would use it once a quarter maybe. And that was for founders like raising all the time.
And I think for salespeople, we could do, I think a pretty good job, but we had to ingest a ton of context and then we sort of went on this quest right, where we just tried to ingest context to make great sales decks. And that was pretty challenging.
Keith Peiris:
Yeah, I think it was this weird realization of, “Wait a minute. Maybe a horizontal productivity tool isn't the right thing to build for this technology.” Right. Because to Henry's point, we had all of these different users. We could help you make your first presentation, but we couldn't help you make your second, third, fourth hundredth, because we just needed to know so much about you. We needed to know, like, who are you speaking to? What do they care about? What company do you represent? What are your goals? And sort of this never ending context problem.
And I think we came to this realization that the next generation of great companies will be everything for someone. So what should we focus on? You know, we sort of had a lot of sales and marketing folks using the product because of slides. And, um, I also think as consumer people, I think that the idea of building the everything product for this really large group of people that's really important to a company is pretty exciting.
Yeah. I think, what is it, something like 30 or 40 million people use Salesforce every day. Um, you know, with varying degrees of proficiency, they have very complex things that they need to do, that need to get out of it. And I think like that, that sort of attracted me, you know, personally.
And then I think we also realized that you sort of need – I think to do great work with LLMs, you actually need a couple of things. One, you need all of this unstructured context, which we'll talk about. Plenty of like what did people say, what did they do? You know, where are they going?
But you also need some structure for the model to navigate it. Or otherwise it'll just take 25 minutes, 30 minutes to like, do anything. Um, so we were like, “Oh, this is actually like the perfectly shaped problem to reimagine. Like what should a CRM do?” You know? Because a CRM is just the structured relational database, but if you have that plus the lake of interactions, that would be an incredibly powerful product. It would be an incredibly powerful piece of technology that I think every company would need, you know?
Seth Rosenberg:
Right. So now you're building Lightfield, which is an AI CRM. So what is an AI CRM?
Keith Peiris:
I think the like defining characteristic of an AI CRM is that it has the context to do the work, and do the work for you. And it's sort of a – it's an interesting definition because like if you asked a hundred people, you know, “If I opened up Salesforce, would I have the context to work on this account, or this deal, or this or that, this opportunity?” I think 95 people would say like, no, this, someone filled out these fields. These fields are sort of cryptic to me because I don't know this person.
So, um, I actually think in the pursuit of building an AI CRM so that LLMs can do work, you actually end up creating a much better product database system so that humans have enough context to do the work too.
Seth Rosenberg:
So the first wave of AI go-to-market tools, we are seeing a lot of noise, a lot of billboards around San Francisco of like, “I'm your AI SDR.” So how do you compare what you're building at Lightfield to those kind of initial companies?
Keith Peiris:
It's um, it's sort of funny. I think a lot of those companies were trying to do the work without the data. And then what you end up doing is you end up with like great demos, right? Where it sends an outbound email. Where it can like, you know, pass your, uh, or like fill out your classical CRM for you.
But I think when you push those things a little further, you realize, wait a minute. I need to understand this company. I need to understand, I need to understand the company that we sell, right? And, uh, the products we sell. Then I need to understand the relationship with the customer. Then I need to understand how do we communicate, where are we?
So I think those, a lot of those products they just like miss, we're missing the data and the data structure to like go the whole way. And they were trying to sell the use cases. I think, you know, we all know the story – a lot of those companies got pilots at large enterprises and you know, the majority of those pilots didn't work.
So we sort of took this like data structure first approach of we're going to suck in everything, we're going to index it, we're gonna structure it. And then when you have that, you will be able to get it to write emails on top. You'll be able to get it to generate proposals and slides and read statements of work and, you know, and understand your transcript better.
So I think we sort of took this database first right approach, and now we're starting to see the applications layer on top. And our customers sort of love the results.
Seth Rosenberg:
So let's say I'm an AI B2B company going through YC right now, and right now I'm tracking my customer list on Google Sheets. So why would I use Lightfield and, and what is the product?
Keith Peiris:
The first reason to use Lightfield is your Google sheet doesn't capture everything, right? You probably have a Google sheet and a call recorder, and they probably don't talk to each other. And one of the problems with your Google Sheet and your call recorder is that your call recorder doesn't actually know anything about you, and it doesn't know anything about your customers. So that's why the output of it, the action items, the tasks, the transcript, it's always like a little naive compared to what you would know about your company.
I would say the first reason is, um, you just get high fidelity capture, high value recall. So let's talk about what it would look like. You sign up for Lightfield. You connect your email and calendar. As soon as you connect your email and calendar, and then more If you want, like your Slack and your customer tickets, we'll start storing and indexing all of those customer interactions automatically. You just press a button, it happens. And on top of that, we'll sort of generate the objects that Henri spoke about on top. And then we'll start sending a video recorder to all of your customer meetings.
Um, and so right off the bat, before we even get into automation of work, you've just got an automatic CRM. You know, one that listens to all of your customer interactions. That updates the CRM for you. That, um, gives you a data lake that you can query for, like “What did Greenlight ask for yesterday?” You know, or “What did RAMP say that was important to them?” So you've just got this great sort of working data lake that you can query to with records that update.
So that's sort of the base of the product. And then, the cool thing is now that you've got this base, this foundation of data, now you can get it to do more stuff. Right. So now you can get it to create tasks, assign tasks, write emails, find you customers that meet criteria, read proposals, generate proposals. So suddenly it becomes like a real partner to you in sort of sorting your go to market.
Henri Liriani:
Something that's pretty magical in this sort of customer segment we see a lot is that people can query like, “What did my last hundred calls really tell me,” you know? And, “What does this mean for my ICP or some stand, uh patterns that might be standing out.” And I think we actually find that the nature of that CRM and data model in this case lets you pivot around, um, which is pretty useful as an early stage company and really hard to do if you think about like having to bake your data model before you talk to people, right?
Which is kind of the status quo of traditional CRM, right? It turns out that if you've been talking to, let's say, a hundred customers over the course of a few months, and then you realize that you actually care about how many go to market people they have, but you haven't been capturing that field. You can just kind of ask that question and retroactively fill it, right? Because you're always talking to the source of truth.
So I think that's kind of a useful thing. And all of our customers, even though it's a bit of a squishy value prop, all of our customers I think really appreciate Lightfield for that. They sort of ask it every Friday, what did we learn this week? Right. And it sort of creates a bit of a drumbeat of like what they share with their teams about, you know, what they're learning from their market and stuff like that.
Seth Rosenberg:
So it seems like you're, you're focused on this next generation of AI B2B companies. And why did you decide to kind of target this segment?
Keith Peiris:
So there was this period between sort of tone and Lightfield when we were just working with go to market teams, you know, 500 person companies, 1,000 person companies. And the first thing that we noticed is that the data was just everywhere, sort of in the wrong forum and the wrong structure. And it was a high latency operation to check Salesforce and then Gong, and then Zendesk, and then Snowflake, and then pull all of that stuff together. And we thought that there was just a lot of benefit to putting it all in one place.
I think if you had built your go-to-market stack 10 years ago, you probably use this tool for sending out email sequences, this tool for reading proposals, this tool for like forecasting this other tool for filling out the CRM and this other tool for reading your transcript and pulling out insights.
So you have workflows that are just fractured all over the place. And to be honest, most of these things are like simple LLM calls now. You just need to have an agent and teach the agent about your tool and have evals, and you could build all of that stuff really quickly.
Henri Liriani:
I think, um, for one, we see a lot of technical founders right now starting and doing sales on their own. And then kind of approaching building a go to market org from first principles, which is like, what exactly do I need? I need to find these customers that look like these three, like design partners that I had.
And then I need to like, you know, figure out where they are, send them emails, book meetings with them and carry them through a really efficient process. And I think that there's a lot of interesting engineering thinking being put towards building a go-to market machine that isn't so sales craft focused and is much more like, you know, how do I just plainly express the value of my company to the right people that need it.
And actually I think that's like most of the people that we work with right now. They're sort of in that sort of style of go to market. And I think that that colors how they build their team too. I think they're leaner teams. I think that they're more technical, and they know how to kind of wire up the minimum tools to have kind of the right information flow and have it land in their source of truth. Which in this case would be the CRM.
And I think the way that they work is,you know, they'll try to get you very quickly to a proof of value, or a proof of concept, and then convert you as a customer and build a really like active conversation over Slack or something like that, and just like keep it going, almost like you're an extension of their team. And it's just sort of a different feeling.
Keith Peiris:
Can I get into the industrialization of knowledge work?
Seth Rosenberg:
Let's hear it.
Keith Peiris:
Over the past 20 years, a lot changed in tech. I think that the thing that we're seeing now – I'm just texturing what Henri was saying is that it's not clear that every software company is going to be like a hundred billion dollar software company where you need to have this hyper-industrialization of go to market.
I think what's happening now is that you want fewer generalists who have, who understand everything. And using a tool like Lightfield, you're efficient enough and you have enough visibility where you can spend time looking for customers, talking to customers, serving them, take care of them. I think we're sort of building for that world, that world where you sort of have like superhuman sellers that do it all.
Seth Rosenberg:
And so do you view long term Lightfield actually expanding beyond the go-to-market org?
Henri Liriani:
I think we really want Lightfield to just be really accessible to anybody at the company. And long term, I think the exciting potential for a CRM is for it to not just be focused on being a cockpit of managing a relationship with a specific customer, but just sort of be this like hub of customer truth for a company, and like what they, what that's power anything.
You can imagine using that to inform, like a roadmap with a ton of detail. I mean, we do this actually now. Like we were wondering how many people have asked for CSV export. And where to rank that, and Lightfield’s pretty good at querying stuff like that. And really you just wanna be talking to like your memory of customer chats.
Seth Rosenberg:
Yeah. So it's a super interesting vision of kind of like the counterbalance to the hyper industrialization, hyper kind of, um siloing of the go to market market org that impacts both the go go to market org, but also the company as a whole in terms of just being more disconnected from the customer. And Lightfield is basically a way of bringing it back. I'm curious, are there gonna be more human account executives in the future or less?
Keith Peiris:
That's a good question. I think that there will be probably fewer go-to-market people. And the people there will do more.They'll just be able to serve more accounts.
They'll be able to, you know, help with sales engineering, help with customer success. They'll be able to communicate better. Um, so I think it'll be a story of just slightly fewer people, but doing way more and honestly way more autonomous.
Henri Liriani:
I think there will just be a lot more companies, like small companies, and I think there might be a lot of – there might even be more AEs as a result of that. But certainly each team won't need as big of an army of AEs, I feel.
If you think about like your ability to address a narrow market with a ton of care and consideration, you know, and that ability changing with AI and like how much more carefully we can do that, I think sales actually stays really important because sales is just this sort of ma – I mean, I think it's best form, it's this matchmaking process between needs and solutions for those needs. I think more people will just be able to do that effectively.
Seth Rosenberg:
Yeah. And we've talked around this a little bit, but why is Salesforce not the AI CRM of the future?
Keith Peiris:
It's a good question. I, I think with infinite resources, I'm sure they could build something great. But I think you're sort of mired in what do your existing, how do your existing customers want to work and how do they work?
Seth Rosenberg:
It's like basically a hub and spoke model.
Keith Peiris:
And then how do you think about the ramifications to your ecosystem? Right? And I think you just end up in this interesting spot where you, you like probably can't change the way that Salesforce works anymore, right? You just have too many giant enterprises who rely on exactly the way that it's worked.
Seth Rosenberg:
And what's the challenge of how it works in the architecture of how it works in terms of accomplishing what Lightfield accomplishes?
Keith Peiris:
That's a good question. I mean, I would think about the relational database almost like a, like a strainer, right? Which is to say that you've got all these fields and then you'd need to either fill out these fields either through a, you know, a sales professional, like manually filling them out, or AI filling them out. But you know, you actually miss quite a bit outside of the field.
Seth Rosenberg:
It's extreme compression, basically.
Keith Peiris:
Yeah. Um, so you can augment it a little, maybe by having like a customer data lake on the side. But I think what you really want is you want this like, marriage of your fields and sort of the true voice and events that a customer together. And that just feels to me architecturally different than the way that the traditional CRM works.
Seth Rosenberg:
So how did you get started? Like how did you actually, you know, get some initial design partners?
Henri Liriani:
When we were thinking about kind of the pivot and the commitment to this new thing, once we sort of realized what the opportunity was, we realized that we were pretty late to the opportunity. Like there are already a bunch of companies kind of swimming around in this pool looking for the thing that we were realizing here. Because you know, you're right that there's only so many tectonic shifts that could create the room for a new CRM, which is such a central thing.
If you can be the CRM for a company and do your job, meet expectations at your job, you will retain. And probably expand as the company grows. So it's a very coveted role to play. And I think to have the chance or the privilege to play that role is like something that everyone has started fighting for in the past few months. I think that's kind of where we're at, and I think we've been working on it for a little under a year.
Seth Rosenberg:
Well, I think it's a classic Silicon Valley situation where from our perspective, it might feel crowded or there's a few other players, but if you just talk to any random customer, it turns out that no one's using the AI CRM of the future. And so I'm very excited that you guys are available for everyone now, and I'm excited for people to give it a try. Yeah. Um, yeah, so thanks for spending some time and this is fun as always.