AI Hasn’t Crossed the Chasm to Teams: Inside Superhuman’s Bet on Collaborative Agents

0:00You don't really want users to have to think it's time to use AI, right? Like that's kind of, in my opinion, a bit of a failure mode. So instead of having to go to a place, type into a chat box, we can bring it directly to you wherever you are.

0:16Matt Paige:Welcome to the Talking AI Podcast, where we talk AI with both experts in the field and early adopters. I'm your host, Matt Paige, and we're here to demystify AI for you so you can get some value from it. Let's talk some AI.

0:31Matt Paige:AI is still stuck in single player mode. Every individual I know is faster with AI and ask them how much faster their team got. And I hear stories about silos and fragmentation. And all of that productivity is pooling into personal chat windows and it hasn't crossed the chasm to the team. My guest today has a thesis about why. He's in a better seat than almost anyone to fix it. Lang Shackleton spent 10 years as the chief product officer at Coda, which is an amazing product, by the way. Loved using it. which is now Superhuman Docs. It's getting even better, part of the Superhuman suite. And he was a product manager at Google and YouTube before that.

1:06Matt Paige:And today his whole job is what he calls the last mile of AI, bringing agents to your team in context instead of waiting for you to open a chat window. And we're getting into why team productivity stalls while individuals fly, the story of rebuilding iconic products in an AI native way, how Superhuman runs its own AI transformation internally, and whether the chat window survives the next few years. Lane, welcome to Talking AI. Thanks for having me. Glad to be here. So many threads I want to pull on, but almost every guest that I talk to now tells me the same two things. They say, I'm personally way faster with AI, but my team's not experiencing the same level of productivity.

1:43Matt Paige:And you call this AI failing to cross the chasm, the iconic book. But when did you first begin to notice this pain point or this problem? and why, how would you diagnose why it's happening? Yeah, I think a lot of sort of tools are rapidly changing, but at the same time, there are certain sort of stable collaboration tools. Everyone's used to using docs, spreadsheets, presentations, all of that. But at the same time, I think starting in 2023-ish, we all learned about this magical chat window. And it started with this constraint, which simultaneously was really simple. Like you just popped whatever you wanted into a chat window, but it was primarily a soloist thing.

2:30It was you typing into that chat box and then getting something out the other side and usually copying and pasting that result somewhere else. So in some ways, the two kind of classes of tools have grown up differently with different design constraints. And I think that's sort of led to where we are today in some ways, where you have incredibly powerful tools in a chat space. And then you also have collaboration tools that are built for collaboration and teams. And you've yet to see those things really unify in a meaningful way until some of the more modern tools.

3:08Matt Paige:I'm curious, back to the early days, like you mentioned 2023, go back to Coda in those days, chatGPT comes out, all these things. How did you think about it then? Did you think it would be this impactful? Were you already thinking about, oh, how can I leverage this within Coda? Do you recall back at that moment what you were thinking?

3:30Yeah, actually, it was probably one really influential board meeting. And Reid Hoffman has been a board member for a long time of Coda. And I think he was making at the time a pretty impassioned case that we should spend more time on AI. and he got us access to the sort of very first precursor to an API that OpenAI was releasing. It wasn't actually a public beta yet, but he helped call in a favor and we got access to that. And I think probably over one weekend, we all became fairly convinced fairly quickly. And it was really the idea that this was going to be far more capable than I think previous iterations that we had all been playing with.

4:20And when you sort of applied that to our surface, you know, it didn't take much to create some really, really impressive demos of writing formulas, filling tables of data, doing all kinds of things that people are doing at the time manually. So I think it didn't take long for us to go all in after that. It was, you know, as with a lot of software development, it starts with a great demo. And, you know, we had probably two or three engineers working over the weekend. And then the rest is, you know, history. I think the second thing that happened around that time, we had a PM named David Kosnick.

4:56And David and Shashir and I worked on kind of a vision video thinking about, like, how is this going to change? How is this going to change productivity tools? I think that video is actually still up on YouTube. But the kind of core idea was how do all these surfaces change and what's possible now? And then we kind of went through the entire product thinking about, okay, now workflows change, the way that teams run change. So it was kind of a process of thinking through, given this unlock, how to, how to team workflows change in the kind of a collaborative setting versus I think at the time we were all being impressed in kind of an individual capacity or solo.

5:37You have a great mind in product and product rituals and just everything around product

5:42Matt Paige:teams and how they function. You mentioned vision video. What is a vision video and it's, was this public when you created it? Was this more for internal? Was it external? Was it both? I think most companies don't have an AI problem. They have a prioritization problem. There are hundreds of different places you could be using AI, but the hard part is figuring out which ones are actually worth investing in, which ones you're ready for, and what should actually come first. That's why we built Gen ROI at Hatchworks AI. Instead of basing your AI strategy on a few workshops or whatever happens to be in the room, Gen ROI uses AI guided discovery to gather insights from across your entire organization.

6:14Matt Paige:It helps us surface the themes, patterns, and connections that are easy to miss. Then our strategists work with your leadership team to turn those insights into a prioritized AI roadmap, weighted business value, feasibility, readiness, and what actually makes sense for your organization. So if you're investing in AI, but still trying to figure out where it can actually create the most value, I highly recommend checking out GenROI. Go to hatchworks.com backslash GenROI or just hit the link in the show notes. So the background on that vision video is a lot of times, at least I have found one of the best ways to set a vision is to write like a tweet thread, blog posts, vision videos, something that kind of sets forth, here's how good it can be.

6:54And then importantly, to draw a contrast between today. And instead of with AI, instead of writing a blog post, which is kind of boring, we built a set of prototypes at the time that kind of stitched together into a narrative. And so that was meant to be external, because we were in the process of building each one of those prototypes. Okay, we got to see if we can find that video

7:16Matt Paige:on YouTube to put in the show notes for everybody. But I was doing my research and you referred to what you're building with superhuman docs and everything as a new surface for interacting with AI. And I thought that was such an interesting word to use, like the surface for using it, because right now, you know, much of our working mode with AI is through the chat, synchronous conversations. And I'd say that's starting to evolve, right? I feel like I'm doing 10 different things at one time across multiple tools, you know, in multiple things. But how do you think about that evolution of what you're calling the surface?

7:58Matt Paige:And do you see this being a broader thing outside of even what you're building, but just a new kind of modality in terms of how we interact and work with AI? Yes, but really it's just about how do we do things? Yeah. And I think that's in terms of you mentioned crossing the chasm earlier. I think one of the keys here is that you don't really want users to have to think it's time to use AI. Right. Like that's kind of, in my opinion, a bit of a failure mode. Now, all of us that are AI pilled and in these tools all day, every day. Yeah, sure. We're thinking that way. But I think the mass market part of our goal is to use a little bit of the Grammarly DNA, which is always brought AI.

8:42At that time, it was called ML in front of people where they were actually working. So the Grammarly sort of example is I'm going to help you with your writing, write with confidence, check your spelling, check your punctuation. And I'm going to do it importantly in your Google Doc, in your Coda Doc, in your mail application, in your Slack, sort of everywhere. And so I think in some ways what we're trying to do is build a set of AI capabilities that really solve that last mile. So instead of having to go to a place, type into a chat box, we can bring it directly to you wherever you are. now sometimes you're not working in a mail application or a slack application in kind of that uh that mode of collaboration you're working in a surface like you want to actually show your work simple example is writing a document or collaborating on a bunch of data where you need other people to see the kanban board or the calendar or whatever it is and so i think that in many ways the kind of value prop of now superhuman in the suite is pairing those two things together.

9:49So if I want to build something in a collaborative setting, in a collaborative surface, I can build it. You can see it. We can collaborate on it. An agent can be a part of that collaboration. But importantly, it doesn't end there, right? So if we build an agent that helps us do project management or some workflow, it should travel into my mail application and into your mail application. It should travel into your calendar and into my calendar. And so I think the kind of pairing of those two things together is, is fairly unique. It's funny.

10:19Matt Paige:I was actually having this pain point the other day, I can't remember if I was in quad or chat GBTs, whatever they're calling now, codecs at work, but I had that same problem. I wanted to get somebody into the conversation and I believe they have the functionality to do this for the life of me. I couldn't find it or figure it out, but I had this mental thought for a second because context is so important in the conversation. And I'd be curious your take, because this is literally how the product works. When there's multiple people collaborating in a doc, anybody can add context, which can be an amazing thing, but could be a negative thing.

10:55Matt Paige:You could think of in meetings where you have 10, 20 people in there and then they go off the rails. I'm just curious, how do you think about that in shared workspaces and context coming from other angles? And maybe that's the nuance of AI and its ability to parse things out. But curious, any thoughts on that, the cross-collaboration and context? Yeah, I mean, if you think about the way that a lot of people are using AI tools today, I think there's a real context problem. And the context problem is that they're continuously stuffing via copy paste context into this, into this LLM and trying to use that sort of thin slice of context to get an output.

11:38and then usually the memory of that and everything else is gone in the next session. And so I think that part of what having a kind of collaborative surface does or like a place that you're actually managing that knowledge does is that the AI is operating on shared context most of the time. So in Docs AI as an example, if you have a team hub where all of your PRDs or all of your project management or all of the marketing briefs for your launch are, AI already has that context. You don't need to go paste a bunch of these things into your Solus chat window. And I think that shared context across a team is kind of underappreciated, but when people use it, they immediately see, oh, this understands the rest of everything that's in this team hub, and I don't have to do this kind of like continuous copy paste in and out of these tools.

12:36And so I think that's sort of like one piece. I think the other piece of context that's really important is continuously updating that. Historically, before AI, teams may have had something like a decision log where they would kind of say like, here are all the decisions. These days, you mentioned Claude, you know, we have a lot of customers who are using our MCP to continuously update the memories for a team. And so in that way, they might hook up a granola or a fathom or like some note taker and then say, hey, every decision that we make, just stuff it into SuperDocs as a memory into our team hub.

13:16And then that way, you know, you have a place where you can kind of go in and say, actually, the way that you recorded that memory wasn't correct. That wasn't the decision. So it's both consolidating that context, but it's doing it in a collaborative place where we as a team can edit it, right? Versus a lot of these AI tools have memory that is individual. And I think that in a team context, it's most important that the context you're using for LLM is collaborative.

13:48Matt Paige:Yeah, it's almost like having a hive mind type of concept. Like I've done the whole second brain where you have the LLM, uh, Andres Carpathi's LLM wiki. And it's amazing like what it can do when it has context outside of just your single chat window and the, the thread there. I want to go back and not everybody probably knows this, but if we go back, so Grammarly acquired Coda around what, early 2025. And then they acquired Superhuman around mid 2025, I believe if my timing's correct. And then the rebrand around superhuman is like the parent brand. What was that experience like? Because you were, you built and ran Coda for what, 10 plus years.

14:30Matt Paige:Like what is, what was that transition like throwing on the fact that you have AI and just everything changes in two seconds in today's age? Yeah, for sure. It's, it's a really interesting sort of journey. We built Coda for 10 plus years and then it was acquired by Grammarly. like you said, Grammarly, very different product, large, really large kind of consumer base, super well-known brand. I think Coda at the time, very teams-based product, obviously have both consumer and enterprise businesses there in both. And then we bought Superhuman Mail as a part of that, rebranded the entire company to Superhuman.

15:08We've since bought other companies or at least announced some of those acquisitions. So we bought a company called Rose in Portugal, which is an AI native spreadsheet. And so part of this is the broader arc. You might ask why and how does all this fit together? The broader arc is what we're really trying to do is build kind of an AI native productivity suite where, like I said, we have great first party surfaces, mail, docs. We just announced databases as well. I think that would be really central for what we just talked about in context. And then importantly, Go, which is the AI assistant, It should work everywhere.

15:46So not just in our first party surfaces, not just in our suite, but it should really work in wherever you want to work. If you want to choose us for docs, that's great. But if your team is working in Google Docs, that's great too. The agents that you build in our suite should work across all of these tools. And so that's the pairing of AI native productivity suite with first party surfaces that are great, but at the same time, able to, you know, able to choose any tool. It's interesting to think of Grammarly.

16:12Matt Paige:It's like one of those most recognizable software brands out there. Its superpower was that it was just always present, right? It was like present no matter what tool or platform you were in, which I keep thinking that's the same strategy in a sense of what you're talking about in a lot of ways with the MCP connection and being able to work with whether it's directly in your platform or in Cloud or whatever it may be. It's super interesting kind of progression there. We talked about the couple of things that were launching and I was digging through this July 8th, I think it was a few weeks ago.

16:45Matt Paige:And it wasn't just, Hey, we rebranded or we launched this one thing. It was like docs, AI, AI views, MCP databases, desktop and mobile apps, visual redesign. What was that moment like? And then what was the decision being on the product side of leading AI in like determining, okay, we're going to do this big bang approach versus phasing this thing out. What was that like? Yeah. So one of the things that we feel pretty passionate about is that we're constantly delivering value to our customers. And so in some ways we didn't want to make it just, hey, we changed the visual design of this tool. We want to say, there's all these problems that we want to be able to solve for you.

17:30Databases has been a great example of where our customers have been asking for versions of what we launched with databases for a long time. Scale up to a million rows, super fast performance, ability to share data selectively with different audiences. And so a lot of that is really built on the learnings from Coda, but now in the superhuman suite. Yeah, I've used a really fun one. The team is just on fire. Tell us about what that is. They are building basically...

17:59Matt Paige:It's probably one of the more interesting ones, the stuff that was launched. Yeah. AI views, basically the kind of core insight here is you have data that you're building around, collaborating around. And oftentimes, historically, Coda, now SuperDocs' building blocks were you could build buttons and you could do layouts and you could do a lot of fun stuff. But at the same time, we wanted to unlock people's full creativity. And AI views is really meant to do that. So you can basically take a table, a data set, and build really anything on top of it. So you can prompt your way, been like a lovable fashion if you've used lovable on top of a data set.

18:42So say you're running some process, say you've got a wedding that you're planning, you can build a seating chart that looks like an actual seats in a wedding, or you can do$100 voting where you can actually see money flying around. Like there's just so many fun examples where you take a data set and that data set becomes live. And it's really up to the maker's creativity for how they want to use it. And so I really, it's been a lot of fun watching that team. They're iterating super fast and yeah, I can't wait to see it.

19:16Matt Paige:I think the other, I'd be curious just from a product standpoint or roadmap, because you have a, you have an existing base of users that were on Coda. Then you have all of these other brands. How do you think about serving all of those different customers? Cause you're launching all of these new updates. How do you think about that while also trying to acquire new customers as well? I'm assuming it's this balancing act. between the different personas and groups that you have? Yeah, I mean, so I think Coda and now SuperDocs has always been focused on teams. And so I think in some way, that's kind of like a unifying thread.

19:54And there's certainly capabilities that we build for small teams that a large team may never use and vice versa. But one of the fun parts of building a very horizontal platform is that you kind of don't know how people are going to use it until it's out in the wild. And now that it's out in the wild, you know, we have teams that are running entire project management orgs on it. We have teams that are running their goals and OKRs on it. We have teams that are running that look more like kind of a knowledge management set of use cases. But I think the through line is really teams. You know, it can be small teams.

20:32It can be medium teams. It can be some of the largest companies in the world. So that has been the through line. And I think the cool part about AI and the thing that we're observing now is that, you know, it really makes it way more accessible to build for anyone. You know, the kind of principle of the tool has always been low floor, high ceiling. You know, we should be accessible to everyone. But at the same time, if you want to build something that's like really solves a critical workflow that might require three other pieces of software, that should be possible. and historically you know a smaller number of people could actually accomplish that manually without ai now that's like just describe the workflow go through a little plan mode you know set of steps and that whole thing is built for you and so in some ways people i think are starting to understand and you see this unlock in customers where previously they may have struggled with a formula that would have automated something and now they're just like it just worked I just described what I needed and now I have the perfect tool for me.

21:36And that's, I think that's the other kind of through line is that software should feel personal. And when you build it yourself, when you sort of describe exactly what you want, you're not describing that to some software team and then asking them to build it for you. You're seeing it happen and iterating with it and evolving it with your team.

21:55Matt Paige:When you realize that you literally just can do anything nowadays. days, but you talked a lot about teams and you're a great, a just product mind, but just in general of how you organize product teams, get things done. But the traditional product team, you have product manager, designer, engineer. How have you personally seen that tripod dichotomy evolving internally? Have you noticed any nuance in just in terms of how product teams work and function now that they have AI and agents becoming part of the team? Oh, absolutely. I feel like it's transforming every aspect of how we work. The most obvious one, which I'm sure you hear from all your guests, is that those roles are really collapsing in interesting ways.

22:40So I think that even before AI, I think one passion point of mine was that the best PMs were often had some design chops and had some engineering chops. And same way with a great engineer who was a kind of product thinker. And the kind of Venn diagram of those roles overlapping was a good thing. Now it's like, not only is it a good thing, it's happening all day, every day. So when a designer feels passionate about an idea, they're not writing, they're not building a prototype that takes them a long time. They're writing production code to demonstrate how that would actually feel in production.

23:13similarly i think having people who really deeply understand the technical constraints on the engineering team being able to build product ideas that i think some pms or designers would never think of because they don't understand the sort of like deepest underpinnings of a system has led to what you're seeing in databases in some ways so i think a lot of that is extremely exciting. In my mind, you can take it, you can take the entire EPD cycle and go step by step and just watch the transformation. How does feedback come in? Okay, now it's like much easier to summarize, aggregate, and then go look at individual pieces of feedback.

23:56How does EPD, the actual process, the making of the software, that's obviously changing with anyone able to the right code. And then on the other side, steady. Our goal is really to have like a steady stream of launches where the customer is always informed of a continuous set of changes as opposed to, I think, always having a bigger chance of launches.