You're Using AI Like Google Search. Here's What Silicon Valley Does Instead

Publicado em

Vídeo em Inglês, publicado pelo canal Vaibhav Sisinty.

About this content: The summary, key points and topics below are generated automatically from the video transcript. They reflect the content, opinions, claims and predictions presented in the video and do not represent a position, validation or fact-check by IA Dots.

Video summary

Akshai Kotari, co-founder of Notion, demonstrates how Notion operates internally using AI agents, custom connectors, and a central context layer. He discusses how Notion evolved from a note-taking app and document tool into an operating system where humans and AI agents collaborate.

Key points

  • Notion uses internal AI agents and custom workers to automate tasks like syncing data from tools like Salesforce, Gong, and Zendesk.
  • The platform functions as a central context layer, enabling engineering, sales, and people operations teams to share information and collaborate with AI agents.
  • Custom agents and custom blocks allow employees without coding backgrounds to build functional tools, such as office seating maps and feedback routers.
  • Notion started as a no-code software creation platform in 2013 before launching its first product focused on notes and documentation in 2017.
  • Kotari emphasizes the concept of malleable software, where AI generates structures and applications based on user input, while humans retain agency to review and edit.

Topics

AI AgentsNotionContext InfrastructureCollaborationSoftware DevelopmentEnterprise SearchAutomationProductivity

What you will learn

  • AI agents can maintain two-way memory sync with Notion, enabling consistent context across different AI tools and platforms.
  • Connecting external data layers via workers allows teams to bridge silos, such as linking Jira engineering tasks with Salesforce customer requests.
  • Non-technical teams can build customized internal applications and micro-apps using custom blocks powered by Notion databases.
  • Centralizing company knowledge and meeting transcriptions provides AI agents with the necessary context to answer complex organizational questions instantly.

Questions and answers

What is Notion Workers?

Notion Workers is a developer ecosystem product that writes a small amount of code to sync a Notion database with external data layers like Salesforce, Gong, and Zendesk every 15 minutes.

How long did Notion spend without launching a product?

Notion spent the first four to five years of its history (from 2013 to 2017) developing its platform before launching its first product.

What is the purpose of Notion's data scout agent?

Data scout is a custom agent built by a Notion data scientist that allows employees to query snowflake data, review customer usage trends, and analyze spend details without needing direct data science assistance.

Video transcript

Transcript from available captions; it may contain errors.

Man, I I have to say thank you for showing all this good stuff. I'm seeing a $10 billion company showing me how they use AI. >> Oh, this is so bloody cool. I'm just going to show how we work. >> Yeah, >> internally. >> That is exactly what I want. >> And uh >> I love it. >> Akshai Kotari, co-founder of Notion, the app millions use as their second brain. And this isn't a demo. He's opening up Notion's internal system and showing us everything inside. >> How people inside of SF use AI. Man, I stepped away for 3 weeks and I was like, "Wait, what do you mean? You like you're not using agents? You're still answering Slack responses like a human?" People say like, "Oh yeah, I'm using Claude. I'm using Tragy." And then you look what they're doing. They're literally using it as like Google search. But the agent stuff is real. I just showed you a bunch of different things where it's not like a demo. This is how we work. This is how a lot of companies in SF work. >> Uh there is no place for an average anymore. So either you're the top one person who has taste, who has design, who has flare, who has original ideas or you don't. People in the world today are builders who can sell or sellers. No matter how AI changes, those two roles will will stay. There are some customers that use notion even better than notion uses notion. >> Yeah. Think you know what? Today I want to learn how to use notion like you. >> Let's do it. >> This is your prompt. >> This is my prompt, >> dude. Even the prompt is formatted like a notion doc. >> Yeah. Interesting how people think of notion as like oh you started in note takingaking because >> So you started as a notetaker as well right? >> People forget the first five years of our history which was actually painful cuz we didn't launch the product. >> You guys didn't launch a product for 5 years. >> Yep. >> One agent is moving Zenes tickets to here. Another agent is like writing the plan. Claude is writing the plan. Codex is actually coding the thing. >> My team on the other end is listening to everything and they know what's coming once I go back to Bangalore. >> This is syncing with the seating software that you have. >> No. So, so this is the cool part. >> Problem. >> This is so cool. >> This is not spam. This is actually good, good good. >> We have another. >> Okay. Up to you. So, Coleman's the >> I would not want to stop till you say we have to stop. >> I think you >> Second Brain I think got created because of notion because everybody's second brain was on notion before AI came in. >> Yeah. because I would, you know, >> I wonder what the roots are. But I think it was around the like very early in our innings, we saw that. >> I think somebody had a >> people would sell templates, right? A second brain template. And that is the first time I've ever heard of a second brain >> and I I bought these templates, man. >> I bought these templates. Not that I've ever used it. Honestly, I was never disorganized. >> That's that's the other thing that's been an unblock. >> Yeah. Now with AI, >> like my wife, I I've tried to convince her to use Notion for seven, eight years, and I failed. because she would look at these things and she'd be like, "Oh, yeah, this looks very organized. I don't want to do that work." >> Yeah. >> And then six, nine months ago, I said, "Okay, forget what notion is. Just talk to this agent." >> And now she's like a paid subscriber cuz like she doesn't even touch the the system of record part of notion. She's just talking to the agent, do this, do that, and it's just going and doing it, right? So the second brain idea has always been interesting, but it's always been a little bit like, wait, that's a little bit of work. But imagine now you have the second brain set up and now you have agents and you just tell them instructions like, "Hey, look at all my email. Look at all my customers. Look at all these different things and then keep filling this thing by yourself." Like I'm not filling it out. >> You keep building this out. So now you get to use your second brain, but you didn't know work. >> Yeah. >> Yourself to having to like maintain it and make sense of it all because agents can do that, right? And so all this it's like it's kind of fun for me to think about this because I really think of notion today as like critical infrastructure. >> This was infrastructure we built so that humans can collaborate >> and interest increasingly now we can extend that to agents. >> No I I I read your homepage. >> Yeah. >> It says a platform where humans and AIs can collaborate together. Something like that. >> Exactly. >> Right. I also saw billboards on the way. >> Mhm. >> And says the same. So, you know, that got me more intrigued saying, you know what? Today, I want to learn >> how I could redo how I work. >> Mhm. >> And how my team can work with notion as my OS. >> Yeah. >> And that's exactly why I pinged you saying, "Can you teach me?" This video is sponsored by Whisper Flow, the most accurate speechtoext app I've used. It lets me write just by talking. I hold one key, I speak, and it types for me in Notion, Slack, Gmail, pretty much anywhere. And they've just launched their own speech model, Kanto, which takes it to another level. And with AI, that really helps because the more you tell it, the better it works. And we say a lot more when we talk than when we type. And this video is also sponsored by Fulcrum. A lot of companies know they want to use AI, but have no idea what to build or who can actually build it. With Fulcrum, you just describe what you want to get done. It gives you an AI assessment of exactly what it would take. And then their team helps you turn it into a working solution. And this episode is also supported by Open Art. Every AI model is good at something different. One for images, one for video, one for keeping the same character in every scene. Open Art puts a lot of the best ones in one app, so you just pick the right one for the job. And I'll show you all three of these in action later in the episode. You'll find the links to all three in the description. Should we do some demos now? >> Let's do And as we doing demos because AI does take a couple of seconds because we don't have cbras on for everyone yet uh we could get chatting. So what I want to show is can I get to pick brains of someone like you >> to get a sense of how you actually use AI every day >> and how you use your product >> to do your work better. >> Sure. >> So that people can learn to replicate your process. >> Okay. >> Right. So that's the idea. >> Okay. So, so here's notion um notion employees are called notinos. Notion started as like a elegant document >> knowledge base, right? And so here's where we have our strategy docs, our values, our road map and so forth >> uh which is like pages inside pages. That was notion 1.0. Notion 2.0 really brought the concept of databases and we built it as like a general purpose database, right? So instead of us saying we're your project management tool or we're your application tracking system or we're your CRM, we thought what if we just build a general purpose building block of databases that is easy to use something as easy as like my mom can use it and find value with it. And so those databases is really what um powers like these are the documents I've written or the EPD documents you know we have inside the company. Same database also has all our meetings, right? So all the sales meetings, all the EPD meetings are here. >> The same database also powers all our tasks. >> So are you saying this is as scalable as a I don't know like a Reddis or Aurora would be >> getting there. I think our our database today can handle maybe like of the order of 100,000 rows and then we're trying to make it handle 100 million rows >> from 100,000 to 100 million. So if you get to 100 million that's pretty much enterprise grade. >> Yeah. >> Yeah. >> I think it's important because I think human scale is probably good enough at 100,000 >> but with AI scale >> but AI agent scale is like three orders of magnitude bigger right. So he agent can keep filing a lot things faster >> and so we're basically building it so that >> this is like >> has anybody has anybody hit the limit? >> Yeah. Yeah. Yeah. I think the 100 there's there's people who have a million rows today but it just becomes slow and not as u you know so there's some paper cuts there we're trying to remove. So anyways, tasks, projects, you know, projects allows us to do things like um you know, let's say uh like I guess campaigns and launches are a good way to think about it. Like these are all the things we're shipping. >> And the thing is like you can take a look at it like I have this bird's eye view of just like everything that's happening and I can go into the project into the task into the people and I can fly back up right and because everything is right here I don't have to use any other tool for the EPD site. Now the go to market side is interesting. On the go to market side, we actually use a lot of other tools, right? And so what we can do then, >> what tools do you use here? >> So for example, we use Salesforce for our CRM. We use Gong, we use Zenesk, >> right? >> We use, you know, a bunch of other tools right? >> And so what notion can do is we released a product called Notion Workers. >> Notion workers. >> Yeah. This is like a development developer ecosystem product we released 4 months ago. And what workers is like workers writes just a little bit of code uh and it will sync a notion database with any other data layer. So like you can see your Salesforce data, your Gong data, your Zendes data inside notion to the notion database. >> So it's basically running crons to keep pulling data. >> Yeah. Every 15 minutes it does it keeps it sync. >> I'm sure this is pushing your 100k limit again. >> Yes. Exactly. Yeah. because I think some people's you know Salesforce are like 5 million records we don't we can't support it right now but we're trying to meet them there uh what's cool about this is that you can now join data tables that you never you were never able to join before >> so for example let's say you're using Jira >> your engineering team is using Jira >> oh >> and your sales people is using CRM and now you can create a worker to Jira which will then be connected to the feature request that you're seeing in Salesforce and so For the first time, your engineers can see the customers and what they're requesting and your sales people can see the projects and how they're connecting. >> So you're pretty much becoming like the glue. >> We're the glue. >> We're the central context layer. So anyways, so we we connected all of this stuff so that for every account. >> So these are all workers effectively. >> These are all workers. Got it. >> And then we built a customer hub where I can see a specific account and I can see a dashboard. I will not show you that because I think it will open. These are all live accounts. These are all sales accounts. But but you get the idea and I can essentially have a dashboard. So for I can see growth school and I can see all the Zenes tickets I have all the Gong conversations I have all the open Salesforce opportunities I have in one place right and so this has been like again you can build this in other tools but the you're sort of like limited by the capability or the API and notion becomes that loop. >> So are all notion dashboards on notion? >> Not all of them but I think a lot of them are and then we obviously connect right. So for example, if you use uh Tableau or Hex or something, you can embed those here. >> Okay, >> you can use the MCP to create a lot of that stuff, right? So >> So anyways, all of this then powers I would say our enterprise search, right? Enterprise search here for example, if I if I go here, I can now ask anything about the company on any tool and it's incredibly powerful, right? So I can even say like I am you know meeting webhub uh sis >> s i s i n t y >> t today uh can you help me prep? Right. And now it's going to look at notion of course but it's also going to look at um uh Slack where maybe a lot of the stuff that's happened in Slack. As you can see it looked at a skill. Meeting prep is a skill that uh my coworker Jeffrey wrote. It's a really good skill for it helps prepare the right kind of prep doc. >> But because it exists in the workspace, I it's able to leverage a skill that my coworker wrote and then it's going to take a look at all the context it has and it will write up a good prep doc based on that skill. Right. >> Interesting. There's a good chance that you can end up you might end up finding the customer record of me as a notion customer as well. >> Definitely because I think it is synced with all the other tools we just talked about, right? So, as you can see, it it obviously picked up the public profile, but then it's going to like look at all these other things. >> I'm curious to see if it picks up the cold email that you never replied. >> Uh, yeah, I should probably tell the agent that this is not spam. This is actually good good good good. >> I think it still gets it. I think the agent still gets it wrong cuz in some ways it's like following whatever sense of >> No, but I think cold emails are by default spamming back. >> Yeah. >> Right. So they call it cold for a reason. >> Mhm. >> Right. So I'm not surprised. >> Yeah. So it also has the doc that I think you shared, right? I think in terms of the questions we have. >> I guess so. And so this is actually really valuable context because again it's it's this this agent experience uh default has everything that I have access to. >> So I don't have to think about permissions I don't have to think about data access whatever uh if if if my coworker asks the same question again it will only show things that >> based on your coworker's context and access. This is actually the thing that people underestimate sort of the time it takes to get it right. It's like it has to adhere to the permissions, has to show you the stuff, right? And so, okay, so here we go. Let's see how it did, right? >> And it's built into notion. >> It's built into notion. >> Interesting. >> Yeah, I think that's actually a good point cuz I think like a lot of the enterprise search tools are just search and then it stops. Whereas here, it can search and then you can continue the work. Can I tell you laugh at it when I say this? For a very long time, I would get an output uh from let's say a claw, chat, jeopardy, whatever. >> I would be like format this in a way so I can paste it on notion. >> There you go. >> Then I wanted to read it >> because just annoying. I hate docs. >> I hate it too. >> I hate docs. uh so and I think there is a kind of formatting that you guys have which makes it way more readable >> and you can express a lot more of right >> so so anyways there's you know what are you meeting known context some folks were lined up for demos um possible directions and the references and so forth right so so this is like this is like a personal agent you can actually do a lot of things like you know >> I'm curious curious to see what it can find about me as a consumer of notion. >> Okay. So that's actually a good one for me to take you to data scout. >> Okay. >> Okay. So >> everything out. >> So data scout is actually super interesting. So I'm going to start with the settings, right? So data scout is a a custom agent that a data scientist at notion built >> for notion. It's available for everyone >> for notion the company. >> So this is one of the custo custom agents. >> Yes. Custom agents that a data scientist built for us. And >> but do you have to be a data scientist to build a custom agent? >> You can actually build several custom agents. This one I'll tell you why uh what >> I also want you to show me one custom agent by building one custom agent. >> Yeah. >> Yeah. >> Um so I'll tell you a little bit about data scout. We'll start the query for you as a consumer and then we'll come back to other other custom agents in general has three parts. It's like when should I work? what's my job description and what's my context that captures like triggers, right? So triggers could be like something happened in Discord, something happened in notion, something happened in my email is when it's wakes up to do the work. Then it has instructions. Instructions are basically like >> pretty much a prompt agent, right? >> Um and in this case, like if I if we go to the whole page, um >> it just says get instructions >> in this case. Yeah, like some of the instructions are put somewhere else, but you can use it as a document >> and then it has all the tools, right? So, in this case, we've obviously given all the pages in notion. >> We have Slack. >> Oh, so when you're building an agent, I can specifically give it access to a few pages inside of notion. >> Correct. >> Not just the whole notion. >> Correct. So, as you can see, like, you know, you have all these different tools, all the different MCPs, >> and so it starts with default no access. >> Um, so it's almost like you're spinning up a new employee. realizing okay well what access do I want to give you okay you can read all the public channel whatever I can in this case you can see it it has been given a the data scientist set up our snowflake data directly right >> uh then in this case it it has a specific model you can pick and it has a computer that it's that it's been given >> is this a virtual machine or >> yeah so essentially it has all of that so now I can go and like Every company I've worked with is you're always complaining about not having enough data science resources. He was like, "Ah, like why can't we have more data science?" So, data scout actually is the first time I we have basically seen people not complain about data science at notion. >> That is so cool. >> Because everybody has access to this, right? So, one person built it, a bunch of people collaborated with it, but everyone at notion can view and interact. >> So, multiple people can also build an agent inside of notion. >> Yes. >> Oh, that's a first. I've not seen that on any other platform. >> Correct. Okay. >> So, collaboratively you can build an agent and then once you get it right, you can the whole company can use it. >> That is very good. >> So, okay. So, now we're going to put it to work. So, I'm going to be like I am meeting with web hub. What's your email? Um can you look at their personal usage and also their company usage? Um give me trends and spend details. also look at our enablement docs and suggest ideas for how they can use notion better. >> What is enablement docs? It's >> like our sales enablement docs, right? >> You're going to be so upset. >> We've been bad customers. >> Okay. So, we're going to let let this run. >> Okay. >> Uh and then I'll tell you a little bit about So, so data Scott is one of our more powerful agents. like it actually runs a lot of queries. Probably each query probably costs a few dollars >> or a few dollars. >> But but for what it does, it's totally worth it, right? You could do like a 100,000 queries that would be the cost of one data scientist at notion in San Francisco, right? So totally worth it. Uh but but you can also build like simpler agents. So like a simple agent would be like my mail triage, right? when when people are uh are coming, I just literally have like you know some simple things like I have filters built for this. I have you know never send notifications and so forth. So like >> this is your prompt. >> This is my prompt. >> Dude, even the prompt is formatted like a notion doc. >> Yeah. And it is actually a notion doc. Like you can literally put your tasks database here. It would have all of that context. So screw markdowns, screw XML, just do notion for prompts. >> I think people who understand markdowns can, but most of the world >> this is markdown in a way anyways. >> Exactly. >> Yeah. >> And so then every email that comes in gets gets run through it. This is probably something everybody can build on their own. It's literally you you go in and you say like, "Hey, build me a mail triage." It will >> And what does this what does this do for you? It uh uh removes spam, archives all those things. It drafts responses to the email >> inside of your email inbox. >> Yeah. >> Uh and um and then you can tie it to the tasks too. So if there's like to-dos, it will pick it up and automatically formulate that. >> Nice. >> So this is like a very personal use case. Smilers is an interesting um team use case. So this is something that a people team made. No engineering background. The people team is always sick of answering the same questions over and over again like where's my health insurance? Where's the stapler in the office? Where's the printer? Where's blah blah blah? Right? So, they were sort of like man like we've done we've talked about this over and over again. So, they just wrote a simple instructions doc. Instructions doc is you are monitoring this Slack channel. >> Mhm. >> And you're essentially answering all these different things. here's all the people who work in these offices and just go answer the question. Just basically just, you know, it has like voice and tone. >> Wait, so you're saying I could use a notion agent for my Slack? >> Mhm. >> Oh, okay. I didn't know that. Okay. >> So, I'm going to now show you the Slack part right? >> On the Slack part, this is the this is the agent. >> Just to clarify, it's simply a notion connector. >> Yep. To Slack >> on Slack. I can connect my question. It's the same connector that you guys have. >> So like I can invoke an agent. >> Yes. So you basically say like hey read my knowledge use my knowledge to answer the question. When you don't know the answer create a task when I when I answer it put that back into the knowledge. So what this is what we call the self-building knowledge base right so every time someone is here okay 10001 is quite hot. Can we recruit it? Okay what temperature is this? The SF target range is 6876. I'll use the reading to route the request. This is this is the agent doing that work. Ticket created, right? Um, hi, can I please add a visitor? If that's your best SMS, then you are literally the people team only has to answer when when when uh >> I love it. Every every 30 minutes, you have a message from your team. >> Exactly. And as you can see, the you know, the the agent answers within like a half a minute of that. Man, I I have to say thank you for showing all this good stuff. This is so good. >> Okay, so I have another one for you. Um, >> so another one that I really like as a product person is that we have a central place for all the feedback. >> Mhm. >> And when we were 20 30 people that was fine cuz like we were all looking at the feedback, we were all working on the feedback. Now we're,200 people. So the amount of stuff that gets into this channel is a lot. like I can't digest everything myself, right? >> And this is usually employees dropping feedback or it could be like a tweet or >> Yeah. I mean, in this case, as you can see, this is like a tweet added here, but a lot of them are just employees reporting bugs or whatever. >> And so in the middle, I was like, okay, this is getting out of handle cuz there's like a 100 messages here every day. Who's looking at it? Are we doing anything about it? And so the first step was like telling the employees, hey, can you report the Android bug in the Android channel? And then can you tag the Android manager? Nobody's going to do that, right? >> So now we built a a a agent that is our feedback router. >> So this is like a triage for your feedback inside of your Slack. >> So yeah, you just centrally dump your feedback in there. The router will pick it up and post it to the right person, create a task for the right person, and everything is being looked at, right? So in this case, it says when AI tries to update a custom agent instructions, I see this thing. It's like, okay, thanks for flagging. I've cross-osted this. I've created a bug and then people can still interact with them. Right? In some cases where it doesn't know, you can also tell this guy like hey um this is the wrong thing. It should be this other channel, right? >> To the agent to the router >> and you'll learn. >> So now I'm going to show you the Now I'm going to show you the router, right? This is actually like one of the coolest things is like if you see the instructions, right? routing rules database. This has like I don't know thousands of instructions. Not a not a single one was written by a human. >> Oh, this is very interesting. So people are giving it feedback >> in Slack. >> In Slack and it's building its rules >> here. >> What if what if someone tries to collude it? Why would they? It's within the team. >> Yeah. And >> this is so bloody cool. Okay. >> And people have fun with the router too. Like I remember Ivan was like, "Hey, every time there's a question about AI, can you make sure you add an emoji?" Wink wink. And then also thank the person who and then it's like, "Okay, remember that rule." So every time someone talks about AI, it remembers to do that. >> So this is like memory, right? >> Memory. This is this is your primary memory for the agent which is being >> it's like basically also completely like um being updated day-to-day right it's like the context is getting updated and the rules are getting updated and humans don't even have to like no employee has to deal with this router this router was built by the product ops team in a way that can learn and I have to interact with the router here in slack Right. And it will just go change the rules for me. What is so cool about what I'm seeing right now is basically getting a glimpse of how a 1500 people company could use AI in the best possible way. But what I really like about what you guys are doing here about the demo is that it's a single agent being used by multiple people. >> Correct. and multiple people are simultaneously improving it for everybody else. >> For everybody else, >> I think that's also like, you know, the other interesting part of what you said is like one person built it, 1500 people are actually experiencing it, are getting >> and making it better, everything better. >> Yeah. >> It's sort of what we sort of like, you know, it's like true multiplayer collaborative AI. You're you're collaboratively building the agent. You're collaboratively using the agent. And um I think there's probably I think there's probably more fun ones here. Let's see. Um all right, maybe I can go back to notion and show you. My team on the other end is listening to everything and they know what's coming once I go back to Bangalore. >> As they should, right? >> Okay, here's another interesting one. >> So this was actually built by one of the logo, >> right? So as you can see canban board is actually an amazing way to think about like the flow of software engineering work right and so imagine a work one agent is taking all the bugs we see on the field it creates the new tasks here >> then an agent picks it up and moves it to review so then if I come into a specific one let's say like here right and then you can see like it will create it will pick up the context it will maybe even like create a plan. So, let's see if there's a plan here. There's a plan here. Okay, here's a plan that the agent wrote, >> okay, >> about how to solve this specific bug. Right >> now, as a human, you can read the plan and you can just comment on it. >> Yeah. >> And when and then the the >> You don't have to though. >> You don't have to though. >> But the agent can read your comments and then move it to the next space. Then it can read and write the code. In which case like if if I'm here you can see that it's it's it has this part but then it also has like you know codeex >> multiple agents finished. >> Yeah. Multiple agents did their work right in this case cuz it's like one agent is moving Zenes tickets to here. Another agent is like writing the plan. Claude is writing the plan. Codex is actually coding the thing. And then maybe another agent is actually shipping the thing. Right? So you literally have a factory and again this is doesn't have to be a only engineering factory right what is performance marketing it's like >> creative >> idea creative >> you're buying >> uh agent creates the creative >> you approve it agent puts it into the thing then it even pulls the MCP finds the data like okay did it do well and then back to it right or or what is the performance reviews thing it's like oh so everything every person can become a factory owner in a in a good way Right. And and in a very collaborative way. One engineer built this setup and everybody now gets to use this setup. >> And you don't have to be an engineer to build the setup any which ways. >> Yeah. >> So so this is like the power of custom agents and it's I feel like in some way >> so just just to give context to people and tell me if I'm wrong. >> Yeah. >> This is effectively the multiple agents picking up bugs from your Slack from everywhere that there is whatever you use linear Jira whatever or you don't use any of them. you use this correct >> of course >> but you don't have to like you know you could build like a synced database to linear or J you can still run the same thing >> you don't really have to change your stack per se to make this work it to make it work like this and then >> AI is basically doing the documentation work understanding what's broken telling what needs to be done probably shipping the code and human your engineers are just approved rejected here are the comments they're orchestrating them >> yeah so like a lot of the doing has taken away. So now it's like now's more time for me to think of new ideas. I can vision >> or I can I'm still reviewing it. Correct. Still reviewing it. >> Uh but but we're reviewing it together. We're building ideas together. It's like it's almost like >> it's like we we sort of release this campaign of thinking together cuz we really believe that. It's like this is thinking together. Uh you can improve the factory and I get to experience it right after that. >> So much fun. I'm seeing a 10 billion company showing me how they use AI. >> Well, it's it's so I mean it's as you said it's just like so fun to build this together, right? Okay. So, let's see how we did here. >> Okay. >> Um Okay. Growth school has a meaningful but fragmented footprint. So, you can see you have a bunch of different workspaces. >> Mhm. >> You can see your usage. You can see the spend. Then then again it looks at like the we have to convert you into a paying customer. Yes, >> but we have to like we can we can if I'm meeting with you it pulls from this enablement what I could do right so you could be like hey why is you have 100 workspaces why is it not canonical why is it not >> I have 100 workspaces >> I guess each team member probably has is that >> and then what questions I can ask right so if you think about like I guess for for smaller companies the product just sells itself so we don't engage but for for mid-market enterprise companies I had to first ask the data data scientist if I'm meeting a customer CEO or CXO how is the usage what the trends are then I have to ask the revops person who's the owner of this account then I have to ask uh the rep like what conversations have you have then I have to ask the enablement person like what should I talk about now I make a query literally on my phone is usually how I use it and 5 minutes later I have un I'm prepped >> the fun thing is you don't have to be a notion to be able to pull this off about your customer, >> correct? >> You could plug in your I mean everybody's data is somewhere, >> correct? >> It could be their Salesforce, it could be their whatever CRN they're using, HubSpot, whatever they're using, >> and they could plug that in >> and orchestrate them to get these answers. >> Yeah. >> Yeah. It's super cool >> cuz it's like in some ways Yeah. As you said, it's like previously I'd have to convince you like, hey, you should move all your stuff to notion because I can connect it for you and make sense of it for you. Today I'm like, don't move anything. Just connect it, build the context infrastructure here and then take this context wherever you want. Right? So if you, as I showed you, like you can use the cloud agent, codeex agent, the cursor agents. So if that's your agent preference, you can still use that. you still pull from the same central context. >> I'll be honest, the reason why we are not a paid customer and I'll be very honest with you, right? I can be. You're the founder. You should I should be. Sorry. >> That's how I refer to. >> Yeah. So, the problem was I really liked notion. Uh we used it quite a lot in the first year and as employees kept getting added uh it got expensive. This is the old notion, right? It was just documentation. And the question that we had to answer, we are an early startup. Every $2,000 is valuable for us in the company. And the question that we had to ask ourselves was, hey, if I could do the same thing, okay, slightly in a disorganized way on a Google doc. >> Maybe I'll do that on Google Doc. And that is when we said, okay, you know what? We paid for a while. I it didn't capture that. We paid for an year. Uh probably $30,000 would be our overall spend with notion. And then we like okay you know what it's getting too expensive the team is getting bigger >> and then there was fragmentation that happened few teams were on notion few teams were not and whatever right >> I also think in this case like probably like picked up the email we gave it right so like if there was any spend or usage in your personal email or another email it probably like didn't capture that >> I I don't the company is also 7 years old I don't think it pulled all the seven years of data that's its recency uh is what it has done but what I'm trying to tell you is >> now Now that I'm seeing what notion can do today. >> Yeah. >> Now I feel bad why I moved away from it. >> We can still fix it. >> Yeah, you get my point, right? Like I think this is so good. So >> I think you're you're putting Yeah, it's I resonate with that cuz I think like you know like for a long time like wiki got a bad rep. It's like cuz I feel like whatever the existing set of wikis were not that sort of innovative or not that inspiring >> and also like like knowledge like who cares about knowledge >> but now >> probably engineers and PMs cares >> but now like agents in the world of agents where everybody can leverage the power of agents your people team your marketing team your finance team can okay if you believe you can leverage the power of agents like what's going to make your agent sharper the right context So now the power of knowledge like I don't know if you saw Andre Karpathy's tweet about like the LLM wiki right it's like all this is just like writing to everything that they're learning about that person >> and Gstack by Gary >> Tan and so now if you just take that for a 400% company like your core advantage is the right context to the a right agent so that they can go do work because then you you can free up so much of the time. Um, I actually show you one more thing that hasn't launched, but I'd love to sort of um let me see if I can pull that up. >> I just saw this today. >> Um, actually, this is this is super cool. I I'm not going to say too much, but um there's there's one thing here which is uh maybe I should just open the Okay, so this is this is so crazy, right? So talking about the like the power of um uh power of AI diffusing not just to engineers but through like everyone right so this person Austin is like a workplace works in our people team right is is responsible for the office >> think of him as like our office sort of environment lead SF environment lead over 7800 people working here each of them has a has a seat right >> and uh we have a database of all the people who are here and then the seat position they're in. And you know, everybody buys this $50,000 like office software which just literally tells you how to move a seat, how to plan a move or whatever. >> Crazy. There is a software like that. >> He built this on notion, right? So let him I'll let him speak. >> I record like a quick video of this custom block as like an admin, how we would interact with it and like all the functionality and options that we have to use. So right now admins 13 I have set. You would hit this rearrange seats button to turn this on. And then right now I'm in San Francisco level eight, but I can go to any office, any floor that those offices have. Like Dublin has a couple floors here. And then say, let's say I want to move this person to this seat. I could do that. Or if I wanted to put this person in the bank because maybe they're going to move up to level five. I could do that. Or say actually I want to move these three people. You could shift click here and then drag all of three of these people to the bank and say like, okay, they're actually going to go to level three and we're going to place them like in this seat, this seat, and this seat. So, you can do all of that. And then you can also do like share seats. And then this is the people bank. This is what you're looking at. And then down here is like the movies that we have planned. Right now, it's flagging to us that we have two people, Jack and Kevin, in the same seat. Do we want to share that seat or do we want to put Kevin in the bank or Jack in the bank? We're okay with sharing for this example. Cool. And then whenever you're done, you'll hit complete, save. And then this will save and write to the database. So yeah, feel free play around. There's a lot you can do with this. One other quick thing to show I guess we would just be like we can turn this off and then you'll be able to search for anybody. Joseph Duncan this is already sitting in New York. Where am I sitting? Austin Cameron here I am in SF meeting rooms. You can toggle on or off if you people want to see the IPs. Yeah, a lot of options with this a lot of functionality. So feel free play around. Just know whenever you hit complete save three moves this will update the database in the back end. So that will be like live on the map. >> Cool. Thank you. One other thing I forgot to mention is once you have the rearrange seat option turned on and you know you want to place a certain person in a seat, you can just select the seat and then over here on the side panel you can go ahead and just type in the name of the person that you want and then it will move them there as well and then you can click this and it will show the plan moves. >> So this is syncing with the seating. >> Sorry. >> Yeah, >> this is syncing with the seating software that you have. >> No. So so this is the cool part. So like you know notion spend a lot of time building the views right like a table view a canban view whatever this stuff we have a new thing called custom blocks as a this >> oh this is custom block >> custom block >> so that's like a >> somebody platform inside of notion >> yeah so essentially like in this case Austin just gave the maps and he's like okay here's my database or all the people and it essentially vcoded uh that view with database as the data layer. So when >> and the database is in notion >> is in notion, right? So when he moves uh one person from this side of the map to the other side of the map, the database has been updated, right? So it's not like a oneoff web app that doesn't connect to the data. It's actually like connecting to it. So you can imagine this is like thing I'm really excited about is like people will build useful work apps, >> micro apps to solve their own problems. >> Yes. And it basically like it you can create all sorts of views with this right. So you do not limit it to our chart view or our table view or our canban view. You could just be like oh I I you know you literally say to AI sort of what you want and then it will create that. But most importantly with the data in you already have it can tap into all of that. And so when you change the visualization the data changes. When you change the data, the map changes. >> And the most and the most interesting part is like this person's like in an environment team like it's never coded before. >> Yeah. >> And they were able to build it themselves and launch it to the company so that now people can find where people sit. They can change their own seats. They can move to a different seat. Like it's completely democratized, right? This was someone's like full-time job like pain in the ass to like deal with it. Now Austin's like, "Okay, now I'm going to go build the next thing." >> Incredible. Incredible. Everybody has to be a builder. >> Every uh and as you can see, I'm sort of like every time I see one of these demos, I'm like, "All right, time to do your Twitter debut." You know? >> So, is this is this like a channel that you run inside of your Slack where people post the cool stuff they're building or this is like >> I mean, in this case, it's custom blocks, but Okay. But we do run like a storytelling channel where um uh we have this guy ish who's whose entire job is to just get people uh comfortable with posting online. We give them help. We we give them ideas but we don't try to overmanage it. It's it's essentially like just giving people the ability and it's way more authentic. Austin tweeting about this about this this is his job. This is something he's built is way more authentic than notion tweeting about this. >> Yeah. Absolutely. And so like our affiliates have like you know five to 10 times more impressions than our brand account. And our brand account does pretty well but like you know affiliates are way more interest >> and your affiliates are your employees. >> Our employees. >> Sweet. Love it. >> Okay. What else? Um we talked a little bit about this. One of the things maybe one last thing I can talk about is like the personal side, right? So, one of the things on the personal side that I'm excited about um uh is this idea of notion pelling your agent. I'm using a lot of personal agents. I'm using Muse and Instinct, Grockbot, all these different things. >> What's your favorite? >> Um >> you're a Would you prefer Instinct or do you prefer Muse? >> I actually like all of them and they were all interesting in different ways. >> Yeah, >> I like Grockbot a lot cuz I feel like it's so deeply connected to Twitter and I'm on Twitter a lot. So like if I want to do anything on Twitter like >> that's the edge >> like the it's it's fast it's it's very well designed. A lot of people who used to work at notion are there. So I'm also like you know I have a soft >> um sort of corner for Grogbot. >> Instinct I like because it's on iMessage. There's I think they've unlocked this thing where I don't have to get to another app. I don't have to remember it. I go to I open my iMessage way more than I open any specific app. Have you felt instinct steps oversteps sometimes? >> Not yet. In fact, I would say like um I like that we're all as an industry trying to figure out the more proactive ways. Like AI progress will not happen if I have to remember to ask something and then it will do it. >> No, no, I'm not talking about proactiveness. It's actually the good thing about it. >> Yeah. >> Overstepping is what I'm talking about. >> I think like so there's there's a fine line between those two things. They're figuring it out. I guess >> they're figuring it out and there's a fine line because I feel like unless you take some risk on the proactive side, you're not going to be like, >> you know, >> if you're a startup, you have to play it play it. Yeah. So, I I think that's good. And then Muse I like cuz I feel like um it's just the first one that I found where um all the integrations they have specifically around like you know being able to like buy things and the computers browsing browser use is like incredibly fast. So >> Muse >> Yeah. >> Okay. >> Even fast you see >> the VM on VM. Uh yesterday they released computer use as well. >> Correct. Correct. Yeah. >> Or or at least like my my behavior is like I just like it's like oh I'm looking for contemporary Indian art you know and here are the few art uh auctions happening can you just do research and then I'm just like left leave it for 15 minutes and then I come back and it's like a you know >> it's it's browse background agents multiple background agents. Yeah. >> Um >> it's very interesting of the approach right if you noticed Grogbot is multi-agent approach instinct is single agent approach with multi-threads in the background. Yeah, >> Muse is very similar. Multi threads in the background, but single window agents. >> Yeah. >> What do you like more? Multi- aents or >> I think I'm still figuring it out. There's something interesting. I I was talking to um uh some folks behind Muse about this, right? And I think they were telling me that there's an interesting debate of like should they have done side chats at all, right? Because I think when you're talking to your EA, it's all one chat. >> Correct. And if Muse is your EA for your personal life, for example, like can it just be one chat? Uh, personally, I've actually enjoyed the side chats because I'm like, oh, like I don't want to mess up the main thread. I just want to like create a side thread. >> Me too. >> Um, but at the same time, I I thought of feel like default. I like this is more approachable than having to think about like which Grockbot agent am I producing? Like that's heavy compared to music. >> That's work. >> That's work, right? And it makes sense because I think there's a more work use case, but I think there's that's the trade-off. >> I think I think they started I mean they still say that this is your personal intelligence. >> Yeah. >> But biggest of the use case are going to come with work. Now they ship Mac and Windows apps with computer use. >> Yeah. So that actually is kind of like this is actually one of the things that I a coworker Nathan just sent me right and I I was telling him like man like why do all my agents have different memories, different skills, different artifacts, different souls? like can I just like point them to my notion art uh notion central base and can it just read from that right and uh and and so I did it like by just prompting it and then he came up with this document which actually is kind of cool it's like all you need to do is you just copy this instructions and give it to all your agents >> okay >> and what it will do is >> what you what you meant by all your agents is notion agents Or I could go to like chat, GPD, cloud, >> all of them. You can just send it to all of them. >> And then what that did for me was that it created a agents doc in my private uh notion >> and it created these docs that that each of them is writing to and reading off of. >> So you pretty much can use all the four, five, 10, 20 agents that you want. They all have the same context, same memory. >> Oh, that is very cool. So like if I go to let me let me open muse here, right? Um if I go to muse and if I go to uh >> you would open a side chat maybe. >> Okay. So if I go to memory then you can see like memory writes two-way sync with notion memory. >> Nice. >> Right. >> And the moment you send that prompt it just builds it across. >> It just builds it across. >> Oh wow. That's very cool. And so essentially you can do that and then I think this is sort of like I think what we talk about is like you know you can it also has these instructions right. So u what I like about this is essentially you get to this setup gives you the ability to start a new agent product without having to teach that agent again. >> Yeah. >> And I think that's really powerful. Uh >> you know what you did a >> you basically made the cost of movement free. >> Yeah. >> Cost of moving from one agent to another free. >> Yeah. And I think that portability is very important for customers to have. And again like you can actually build this specific thing in a different product. Doesn't have to be notion. Uh but I I I work here so I'm biased but I feel like notion is an amazing place for because we've built we've put a lot of care in building our MCP right. So every agent already has a connector and can already write to it, read from it. Uh and actually also think every agent product gets better. like if I'm using a let's say a new agent product that OpenAI releases next week right now it can now read from memory uh and so I have a instant on >> uh in that product but then everything that I'm doing with OpenAI is now being written so that if I come back to Muse I still can still learn from it right yes >> and so if you think about >> kind of like this is kind of the the heart of what notion's become now right for companies and for people it's become this portable context thing that you can carry with it. It's literally like a backpack that you can literally, you know, give to another agent and they can run the next part for you. And and and and in this case, it's like, you know, maybe you prefer Muse for something, maybe you prefer Grock for something >> or you don't or you or there's a new agent that's out and you want to use it >> and then you have your memory, your skills, all of that stuff like central. >> Uh this becomes even more interesting in a multiplayer company context, right? cuz it's like well what is not just my skills but what is the company's skills that >> that is oh I could I didn't think of that >> so like if >> that effectively makes >> oh >> you play the multiplayer for every agent that you built right now >> yeah so if I go to libraries for example right um I now have here skills right and skills some of the skills are mine but I think you now have a Um see why this is not loading up. >> Notion uses lipron. >> We do a little bit although like the mobile stuff is definitely moving towards there you go right so then you have all these different skills from all these different people all these different users and you can enable it for me. Right. So for example like self-re writer it's performance season >> just enable it for you now now this skill is now available to my muse and to my gro. >> Oh nice >> right >> this is because it's on notion. >> Yes. >> So instead of building a skill on claude you have if I build a skill on notion I could pretty much use that skill on any agent that I'm going to use going forward. >> Yeah. So you can point to it so that they can do it. But also here we give you options to deploy it in other agents too. >> Yeah. But I could do this on GitHub as well. >> Yeah. Yeah. >> Yeah. But you're using the base as notion. So you can use it inside of notion as well. >> And I think the GitHub point is interesting because like we're not this is not something new. But I think it's like for the normies like this is easier to use than >> Yeah. Like when we try to teach GitHub to >> I mean non- tech people. In fact, you would be surprised how many tech folks don't understand GitHub. >> Exactly. Exactly right. So it does get very complex. It makes a lot of sense >> and I think we will soon have it so that you build a skill in cloud and then it just instantly deploys it on all your agents. You build something in codeex just deploys everywhere. Right. So >> and so we're >> yeah I think our role has become very clear that we can be this context infrastructure that helps individuals and companies get the most out of it. Um, and it's, you know, the portability. Well, portability is also true for for like notion's given you all the tools to get your stuff out and Google another another tool too, right? Like it's essentially the customer's data. >> I remember reading a tweet of yours where you said there's a world of how people inside of SF use AI >> and then the folks who are in SF go out to realize how people outside of SF are using AI, which is pretty much a search engine. And that's what opens up new markets for both the worlds. >> Mhm. >> Is this what you meant? Do you believe that everyone in SF is using AI like the way you guys are or do you believe that you guys are a little ahead? >> I think if you take the um AI companies like AI native companies, I think we're probably middle of the curve. So when you say AI native, you're talking about open AI or philanthropic >> companies that have started in the last 5 years. >> Okay. >> That have literally grown with AI. Um I think we are like maybe like 50th maybe 60th percentile. Um I think there's actually much more crazier setups I've seen. Um there are crazier setups in >> What is the craziest setup that you've seen? Um I think >> that you would want to get to someday or >> I would say like okay there are some customers that use us that use notion even better than notion uses notion. >> Yeah. >> Um I think um ramp is one of the companies. >> Um >> how do they use it differently? >> I think it's a lot of the similar ideas we talked about. Uh but then they are able to uh leverage the context graph and even more um like agents that they've produced. They have like a really good router. They've built they've managed a lot of the spend better. They've used it to sort of build a lot of products. They've spun up a lot of different products. Um I've been really impressed with the way they have run. >> I also remember you speaking somewhere that there's a three people team who run like 300,000 agent queries a month or something like that. >> Yeah. Yeah, I mean even for us like I think there's there's um there's definitely pockets inside the company that are probably using it as the 99th percentile but as a company we are probably like you know middle of the curve. I I think if you think about the San Francisco city overall, I think we're probably 95th percentile and then the world we're definitely 99th percentile. But but I think a lot of the stuff I've showed you would not would be interesting but not crazy impressive for AI native people cuz they're like, "Oh yeah, of course you do it that way." Like what else would you do it, right? And this is this is this was the reason I tweeted about that cuz I feel like man I stepped away for 3 weeks and I was like wait like what do you mean you like you're not using agents like you're not you're still answering Slack responses like a human like why the hell are you doing that right? Uh so it just feels a bit of like wait like such a foreign world like it's like oh yeah people say like oh yeah I'm using claude I'm using chigy and then you look what they're doing they're literally using it as like Google search right >> you know what we were making a assessment for people uh to understand where they stand in terms of AI any type of question and answer didn't work you know finally what we did we said if you want your real assessment >> you have to export your chat and dump it here >> then I'll give you a real assessment >> that's a good way to do it. Yeah. >> Yeah. I would I would guess if you do that like 99% of the people in the world today are probably just asking >> just replace Google search with it. >> Uh but the agent stuff is real. I mean it's like I just showed you a bunch of different things where it's not like a demo. This is how we work. >> This is how a lot of companies in SF work, right? So if someone visits SF, they get to experience it. They're like, "Oh my god, like the world is so different here in this bubble. It is going to diffuse." And so like what is the opportunity for me to help with that diffusion? And then the same thing if you're sitting here all day long and you just step away and you're like, "Oh my god, like I'm worrying about agents taking over, but like the rest of the world hasn't even heard about agents, right?" And so it's like it's an interesting sort of the gap between the capability and diffusion has never been greater. >> Yeah, that's true. >> Uh in my lifetime. >> That's true. And and and I think that's where the alpha is. >> You meant adoption, right? Like the capability versus adoption. >> Yeah. Well, even if you think even if you think even within SF, if you think about how far the technology has gotten and how much have we understood how to use it, it's still like such a big gap. Like uh there's this great meme. I don't know if you've seen, but like on one side they have these humanoid robots literally cutting corn fields one by one, right? That's how we're using AI right now. It's like I used to do it that way so my agents have to do it that way. >> Exactly. Yeah. >> And then there's the other video which is like there's a big combine that's literally swooping like 100 carrots out of the ground in one go. That's how knowledge work will go to but like nobody has seen what that combined version of farming looks like right so right now what we're doing is like we're basically mirroring how I used it so agents will use it that way but at some point we will make the jump to like okay well what's the agent native way of working right that doesn't involve a lot of this stuff like assume if everybody had all the contacts assume if every connection was already made how will knowledge work >> evolve And what role does a human play or how does how do we make sure that sort of humans intelligence can be augmented by all of this power? This stuff takes time. History has shown us that it takes time. Uh like the two examples I think about all the time is like when the printing press was released sort of invented we still printed handwritten things from the printing press right only after typography we realized like oh there's actually a different way to print it way more efficient >> or when the film was released for years we filmed things on the stage. >> Yeah. >> Uh until we realized oh my god like we can just move >> you can do it anywhere now. So, so I think that's kind of like the phase we're in. >> I think everybody is like all all large companies who want to who want to play on large champs like yourself as well. >> Yeah. >> You're all converging towards two or three ideas. >> Mhm. >> Right. Uh I don't know if it's a fair assumption to make. You can tell me if I'm wrong. Uh what I understand of notion as of today, of course, you guys have moved from uh initially a noteaker. You started as a noteaker as well, right? But non AI note taker to a productivity/documentation/ I don't know how to put it like >> a platform that all of us have used. >> Yeah. >> Uh to right now pretty much it looks like a operating system. >> Mhm. >> And uh which where humans and an agent can work together on >> Yeah. And I could add things like Glean on top of it, Zapier on top of it, documentation on top of it, database on top of it, >> right? Uh and noteaker on top of it as well, and a bunch of other things I guess. >> Y >> right. Uh but the quirk that I saw here or I understand is you've been very vocal about I'm not building my own model. I'm going to be orchestrating on top of great models that are already being cooked. >> Mhm. So is perplexity trying to right >> they have their own small models but they're trying to be pretty much an orchestrator. >> Yeah. >> While on the other side if I look at all the model companies >> Yeah. >> they're also trying to fight the same. >> Mhm. >> While building their own models. >> Yeah. >> Uh yeah. It's quite fascinating what's happening. >> Yeah. I guess in some ways it's like, you know, you back in the day, maybe not back in the day, but like 5 years ago, you could live in your niches and like that was good business, right? So you could be like, "Oh, I'm just building a to-do app." >> Yeah. >> And you'd be like, "Oh, I build the best to-do app." And like that's a business. Now, if you're building a to-do app, what's the next question? You're like, "Well, I'm going to help the person do the things or I'm going to help the person execute better." And very quickly, you'll be like, "Well, but I sort of need the context. unless I know the context, how do I pick up what should be on the to-do list? How do I execute on things, you know, and so you're just one or two steps away from building the whole thing, right? Because you're like, oh, now I need to build the connectors. Now I need to build the you know the vector database of all the knowledge and I need to then take that and like you know so I feel like we're everyone's trying to just essentially help the people or the companies do work and so you could start in the very granular part and then you take two steps and now you're building the whole thing right >> and so it's interesting how people think of notion as like oh you started in notetaking because I think people forget the first five years of our history which was which was actually painful because we didn't launch the product for four or five years. >> You guys didn't launch a product for 5 years. >> Yep. I think we we started the company in 2013. We launched first product in 2017. >> Is that because one of the co-founder designer and they wanted to be wanted everything to be perfect? >> No. Because the original idea of notion was essentially trying to democratize software making. We were trying to build a platform where people can build software without having to learn how to code. >> Oh, so like a no code, low code platform. Yeah, that was the original idea and we built a platform where you could literally build end to end apps. Uh, and it was awesome but >> like an adelo or web flow or >> yeah some version of that >> WordPress or something like that >> and uh cuz it was the same vintage of air table and you know Figma and Kod like we were all started 2012 2013. Now, what what we saw was that it was a platform. It could do a bunch of different things, but at least at that time, nobody woke up wanting to build software. Unless you were a programmer, you didn't really do that. >> And so then we said, okay, well, that's still a dream. And the dream basically, we sort of broke it down to like, oh, we're sort of Lego of software, right? It's like, we'll give you all the building blocks, you can put it together, but we had to narrow it down to something that you would use. we sort of had to like take a wedge into the market and the wedge we picked was docs and knowledge and that's how people sort of think of notion as a noteaker because I think >> a lot of people moved from Evernote to notion back in the day >> but I don't think that we didn't really >> think about the company as like a productivity suite or like you know I think the the romantic idea of notion was actually more broader of like why is it that there's a billion people in the world who are knowledge workers and only 20 million people are developers like why do 20 million developers decide how the billion people have to work >> so you can build your own doc >> and this was a preAI era right >> today it makes a lot of sense it's like everybody's actually thinking about building their own software >> but but in the in that decade we we spent a lot of times building the building blocks and we specifically had this idea what we called malleable software >> uh this is something that one of our engineers here um Jeffrey Lit talks which is not just any software. It's software that you can kind of like >> adapt it to yourself. >> So like even if you say something to AI, it creates something for you in notion, but then you can go in and actually change many things about it. >> Uh you don't have to then keep talking, right? You can you you get a piece of software that you can actually kind of >> maybe change the visualization of it or maybe you move some button. >> You basically get the Lego blocks AI and then you build your own castle. >> Yeah, that's actually a great way to put it. Yeah. >> And in the in the preAI era, we only captured the people who were interested in building the castle or who were interested in using the castle. In the AI era, it's actually like the opportunity has multiplied every day because now I can speak in my language and the castle just emerges. >> Yeah. >> And then I can tweak the castle. >> You're wipe castling it. >> Yeah. Exactly. That's a great product. And so so yeah, so I think I think that's that's sort of the direction we've sort of found ourselves fortunate to have landed. Like I think you could have imagined a world where AI was not happening and then it would have been a hard ramp trying to convince more people to build their own castle. >> But AI has been a huge unlock cuz like all the hard work of building the building blocks we had already done. the hard work of convincing people that you can actually put these things together was freaking hard. AI made that so much easier. And so people are now like now starting to realize like oh now I get what notion can do cuz like uh they're not starting with you know remember like at least when I was growing up Lego or sort of the the bucket you used to get a bucket of small blocks and you' be like okay well build whatever you can. >> Yeah. way harder than when you get a Harry Potter castle and you're like, I can build that by following the instructions right? >> And so now what you end up with is you actually end up with a Harry Potter castle and then you can make it some changes and it could become a Cinderella castle like who knows, right? >> Yeah. >> And and then that malleability is what we care about. That's sort of the romantic vision for us is we think the software should be malleable. >> It should not be as rigid. >> Uh and and this is not actually not our idea. It's this is the idea from the 70s like if you look at uh what Alan K and Douglas Anglebar talked about they always thought of computers as a creation device >> they always thought of software as something you can make your own and so we're essentially just bringing back those ideas to life. >> Oh that's such a good analogy now that I put it together with AI coming in now AI is the builder and the thing that you build is yours. >> It's a great intern. >> Yeah. for you to build whatever you >> I don't think it's anymore an intern anymore. I think it's well beyond that at this point of time. But >> I I say intern as a way where I still feel like humans are in charge. >> Okay, quick pause because this has changed how I work. After every podcast, I open notion, hold Fn key so it goes hands-free and just ramble through the best moments and it turns all of it into clean notes. Then I tell notion AI, make a content tracker with every real idea and an owner. actually know an owner and a deadline in a board view. And it kept only what I meant. When I talk instead of type, I give AI way more context. So, my prompts actually get better. Unlike my laptop's built-in dictation, Whisper learns how I talk and keeps getting better. I've spoken almost 180,000 words with it. That's basically a whole book. And if one of those ideas needs something built with AI, that's exactly what Fulcrum is for. You describe what you want in plain language. So, let's say an AI agent that answers every course inquiry on our website and books a call with the serious ones. It analyzes the problem and gives you an AI assessment report of exactly what it would take to build it. And if you want it done, you can book a call with the Fulcrrim team right here. And they take it from there to a working solution. And whatever you're launching, it's going to need visuals, right? So, let's open Art and create one character for a brand launch just by describing it. Now, let's put that same character in three different scenes. and it stays exactly the same person every time. And with one click, that image becomes a short video. All three links are in the description and each of them has something special waiting there for you. Now, back to the conversation. >> Uh, >> so you're not one of those people that's like, "Oh, AI is going to kill everyone." You're not in that. >> I sort of feel like Yeah. I mean, I'm sure they're seeing something that I am not close to. >> That's what I feel too. Yeah. But I think there's this concept of AGI which is artificial general intelligence which is this idea which is like one branch of computer science where people feel like we can build machines better than humans and I actually come from a different branch which is Engelbart which is Alan K which is Steve Jobs where really think of computers as which is augmenting human intelligence >> which is very much about like bicycle for the mind >> right >> and and and so in in our view like computers AI today is probably the electric car for your mind. Uh but but it's for your mind. It's like you know we still believe in human agency and human sort of you know design and taste and I think AI is actually helping us uh get better. Um, >> also worse in a few ways, but >> yeah, and I think we were talking about this before, right? It's like it's a little bit of like it's like the average, right? And so like the the person who didn't have taste is now better because now they get the 50th percentile >> waits, right? It's like like, oh, I can make really good design. But I think there's still value in the 98th percentile designer who >> who actually has some ingenuity about like you know where it's headed and some taste and like some edge in terms of like the >> I think um I mean I usually say this right like uh there is no place for an average anymore. >> You either have to be very good at it or you're everybody else. >> Yeah. >> Because anybody can be an average or more than an average right now because of AI. >> Yeah. So either you're the top one person who has taste, who has design, who has flare, who has original ideas, or you don't. >> Yeah. And at the same time, I don't want to come off as like snobbish about the top like the 99th percentile because I feel like I think my friend Dylan at Figma says this well. It's like AI has like it's like lowered the floor and increased the ceiling, right? >> It's it's actually both are important. >> That's true. >> More approachable. computers are a much more approachable uh now like I remember when I showed uh Chachibi voice to my grandmom she was just like couldn't believe that she can like have a conversation in Hindi and switch to Marvi and like it's just like it just keeps up with it right uh and and so I think it's definitely making it more human in some ways uh but at the same time I feel like if you're like the top 1 percentile like you now >> can do even more right it's like the possibilities are like um if you were a designer who just like peak design but didn't know how to build now you can build like isn't that >> my product designers ship today. >> Yeah. >> Yeah. And my engineers design today. So like it's it's a very interesting world that we are going to live in or we're living in already. I guess >> that's actually like super interesting to me because I feel like the whole world of engineering product design has gotten so blurry. >> Yes. like at this point um you know you might as well just call them a builder right like um you don't really have those boundaries anymore which is also now I finally understand why member of technical staff or member of product staff or some version of that makes sense because like how do you distinguish like wait you're a designer and I'm an engineer but I'm doing your job and you're doing mine >> so as of today the way it's working for me I I would love to know how it works here as well because the scales are very The way we are doing it internally is every single person in the company. It could be customer success, it could be sales, it could be marketing, it could be product, it could be design, all of them ship their own requirements. >> Mhm. >> Engineering is not going to touch your MVP. You build your own MVP. >> Uh we have a basic uh skill built out to make sure that you know it integrates with the O and uh it pulls in right data and it's secure, it's not breaking. We have a workflow built out. Yeah, >> you ship it, push it to the learners, push it to the users. >> Mhm. >> Get it tested. If there is adoption, you know the metrics. >> If there is adoption, you're seeing good signals. If you don't understand it, that's fine as well. If you think it's working, then let's have a conversation if they should be part of the main product or not. >> So, what we have seen happening >> is everybody are solving their own problems without actually getting to engineering because we actually have a small while we have a 400 people team. >> Yeah. >> My engineering, product and design is only 20 people. It's a business function which is massive. >> Yeah. >> So they ship they take this uh product saying hey I ship this I ship this portal this is working look how the engagement is everybody knows how to how to put post hog into it >> right. So they have the data they present it to my product lead saying >> we should bake it into the product >> and that is when engineering starts picking it up. Of course, they take the code base from there >> and they make it part of the main product. And this has been so fun to see everybody shipping their own stuff. >> Yeah. I posted on Twitter maybe a couple months ago. I said there's only two roles left. You know, you're either a builder or you're a seller >> or a seller. Oh, yeah. >> And I think the best people in the world today are builders who can sell or sellers. You know, >> I agree. And so I think those are the two roles that feels like no matter how AI changes those two roles will will stay right because I think even in a I don't know post AI world there will be a premium for humanto human interaction and so I think that feels like you should maximize that with your customers as much as possible like that's the metric I push our sales team it's like what percentage of your time >> is going speaking with the customer. Yeah, >> cuz that seems like >> Yeah, >> no matter what tools we get, >> that's the human part of it. >> That's the thing that feels high alpha. >> Yeah, that's true. Have you have you come across the anthropic report, the economic uh >> the graph where they talk about which ones? >> No, that's the old one. This is uh you're talking about the spiderweb, right? >> Mhm. >> No. So, they dropped a report I think a couple of weeks back. >> Okay. >> Talking about impact of economy because of AI. M >> uh where they kind of say that you have to start looking at job as a bunch of tasks. >> Yeah, >> you need to stop looking at a job as a job. Every single job is a bunch of tasks and there are four kind of tasks. One a human can only do, second which can be augumented by an AI. Third which AI can do end to end and the fourth is a new one that got created because of the AI space. >> Yeah. >> And the one which only human can do is probably the conversation part of it. Mhm. >> I think that's pretty much it because I mean 80% of my sales calls actually is done by an AI today. But that is it's it's a fun conversation to have because those are conversations a human could not have had any which ways. >> Yeah. >> Right. So I feel my humans my team is having 10 times better conversations than they did two years back. >> Yeah. >> Which is great. >> I think that's Yeah. In some ways it's um there's so much easier to um train and coach your team too, right? Because like the I remember, you know, 10 years ago when I first read about what Bridgewwater was doing, it felt so crazy. Yeah. >> I have no idea what that is. >> Bridgewwater is like this hedge fund that 10 15 years ago um you know the founder talked about how every meeting is recorded and every document is public. Oh, that's it. >> Yeah. And it seemed such a foreign idea at that time because it felt so invasive to everything, right? >> Uh and and now if you look if you just look at well how are we operating? It's like everything is recorded >> like everything is >> you want everything to be recorded. >> Yeah. Because everything is very legible >> and so you can make sense of it all. like business has never been more legible than today and it will only continue to be even more legible because if you live in a world where your systems can just connect to all the different APIs you have every conversation transcribed um including like meetings like in-person meeting is also transcribed so now you have >> notion can do that right I could just yeah >> now you have AI that has infinite context >> context that humans cannot put in their brain but AI can and >> why would you say that? >> Why can't >> I cannot read every Slack channel, every project, every doc being created in notion today. >> Fair >> as a company like >> before taking a decision >> impossible for me to do that. >> But AI can do that. AI has infinite context that it can c can track and so I can say something like well why did the release date move from Monday to Friday and because every meeting is recorded because conversations are there because the documents are there it can be like hey actually the team met yesterday and they decided that they should move to Friday so that this stuff was ready right I would have never gotten that few years ago I would have had to ask like five people about this problem this is so Cool. But for something like this to happen, like you said, every single thing has to be documented. Like every single meeting has to have a noteaker without fail. >> Yeah. And I think what we've realized is like that had the highest product market fit for us. >> Like at this point Yeah. At this point, like if there's like a meeting of six people, like six people are about all ready to turn on the button because then we can all be present in the meeting. And so 95% plus of the meetings unless you're having a one-on-one or having some personal conversation is in the is recorded, transcribed, summarized with task items in the meetings database that is available to everyone. So I can read every sales meeting that's ever happened instead of the go back 10 years or whatever 5 years since we have this. Uh and then that's crazy context. >> You were a noteaker for 5 years on notion. Maybe like humans were taking it before, but I think like >> they were using notion as a Yes, you're right. >> I think we were sort of like on the edge of like, oh, we always took notes. We were sort of >> and now I think the AI does that for you. >> And so anyways, if you're a new employee to notion, imagine starting to do your work and having all of that context. How do I do this? Who sold to this person? When was the last time we talked to this person? Like why do why does Notion do it this way? like it just gives you the answer instantly questions you would not have you would have been afraid of asking your manager you can and then there's a question of like um effectiveness for the company for the company it's not less about convenience it's like yeah you're letting your employees use whatever you are but how will your agents get sharper if they have the right context where does the context come from like codeex cannot pick the context from claude claude cannot pick the context from grock or whatever right and so so you sort of have to whether it's notion or something else it doesn't have to be notion but I think you have to create a central repository where your context is >> uh and actually in some ways like for for engineers it's GitHub right GitHub is the place where you're committing you're merging your version controlling >> and so notion is basically building that but for rest of knowledge work right if you're a marketer >> how do you make sure your performance marketer and your demand gen whatever they all have the same context uh to run from and and this is an interesting one because like we did not think we would play years ago. The market just pulled us into it. Like about a year ago, we we like hesitatingly like built MCP. >> But notion is like the first five connectors any hardness ships >> which is like kind of awesome that this happened. But we sort of hesitantly like released MCP because we were also like other companies worried like what happens with all this data >> and then we were like you know what >> customer owns their data customer wants it let's just build it. So we built it knowing that it may affect the flows of the product flow of the business >> and this thing is literally like a vertical line the way it's growing right >> is that when you said okay we have to transition completely to AI first >> AI first happened earlier but I've seen look I've seen your earlier AI integrations right like that was hey could you rewrite this or could you set this into a table it was block building instead of notion using AI >> right that is how much brand new. >> Yeah. >> And then in like an year the whole product changed to something else. >> Yeah. So we built we we actually the alpha was released like a week before chat GBD. So we were early uh to build the writer product. >> Um we have time stamps. So like you know so essentially like you can complete the doc. It was like a GPD4 wrapper. >> Then we >> yeah then we build enterprise search which is essentially you can connect not just notion but other tools and it will just tell you the answer >> similar to glean. Um then we built the meeting notes product roughly a year ago. Um then we built the agent so you can basically anything I can do as a human agent can do it for you. Yeah. >> So you can build a database it can connect to things like all of that agent can do >> and then earlier this year we launched custom agents which is like you can define an employee give it the right permissions the right tasks that can go do that. So we did all of the work so that you can use notion very much like another agentic platform. I think the shift to the context stuff is probably in the last 3 6 months what we've seen is that people are saying like oh I love using notion but I'm also using these other products >> and increasingly I want a connective tissue across all of these things. Uh and and the context part is interesting for us because it's like that's the that's like we spent 10 years building that part. Yeah. Like we built the best editor, we built the best database that people can use and play with. We built all these connections. So like all that stuff we already had >> and so we were like sure we can we will happily be your contacts there. And uh and and um we we're now spending a lot more time enabling those use cases. >> Yeah. No man, I think when I when I saw the tweet, uh I thought uh honestly like a lot of buzzwords thrown and uh you see some level of execution being done and we're getting there. >> Yeah. >> But I think what I saw today, I know it's a glimpse, but it gives me a very good picture of what you guys are building, how people should be using AI. >> Mhm. and uh how you can leverage the power of this incredible thing to the fullest. So, this was incredible. I didn't expect this. Thank you so much. >> Super fun. I'm glad you found it. I It is also good feedback for us that we need to do a better job telling the story cuz I think a lot of >> agree a lot of the stuff exists. We just have to help people understand how to set up. >> I want to teach it. >> Yeah. >> I think we should figure out a way uh at least because we teach millions of people every single month. Okay, let's do a quick rapid fire. Let's do it. >> Yeah. I have a bunch of questions and I want you to answer them uh as short as possible. >> Okay. >> Yeah. Okay. The first one is, what's something you do with AI that most people would find strange? >> H strange. >> I feel like is there anything strange now? I feel like everything everything's fair game. >> It could be strange for you. >> Yeah. So, do you have that feeling of h this is strange? >> I mean, it's definitely I find myself definitely when I'm driving alone, I will just turn it on the voice mode on that's my strange answer as well. So, clearly >> and sometimes I don't even know like what I'm going to ask, but I just love that I can I can have someone to talk to. >> Hey, what's up? >> Yeah, exactly. >> Okay. >> I think that Tesla makes it even easier cuz you literally hard press it and Grockpot turns on, right? So, >> I think that makes two of us strange. Okay. If you had to delete every single AI tool, including notion, and you're left with one, what would that be? >> Uh, only left with one. >> You could only use one AI tool, not motion. That'll be biased. >> Uh, I love Riverside. I I have a podcast with my with my daughter. >> Okay. We do weekly podcast and I am blown away how quickly we can go from like a Apple voice message to a you know within 30 minutes it like makes it makes us sound amazing and like cuts out >> published if you want to >> fluff and it just like you know few tweaks here and there and it's like published. It's really cool. >> Sweet. And we use Riverside today. You guys should be paying us now. >> Okay. Uh, what's your favorite AI tool outside of Notion? Again, >> you can't your answer can't be notion for any of this. >> Yeah, of course. Definitely. >> Uh, there's so many. >> The current one. One. >> Yeah, let me see. Let me check my phone. Let me see what I've been spending time on. Um, oh yeah, I think um, uh, I'm I think there's an explosion of hardware tools. >> Do what do you use >> that I there's a lot of them, but like I think maybe the if I had to pick one, I would say the Matic um, >> it's a >> it's like a room like a vacuum cleaner. >> Oh, the Yeah, I've seen them. It's vacuum. I think we use it maybe two times a day. The kids love it. It's like a pet in your house and >> it's taken off, right? Like >> so good. So good. >> So I I couldn't recommend that more. >> Uh there's also a new uh Pebble device uh >> the watch >> the the ring where >> Oh yeah. Yeah. The founder of Pebble actually shipped it, right? >> Wait. I I've been using that. >> Oh, is it? Yeah. >> Oh, cool. Cuz what you can do is you can record quickly right on the go. You don't have to take a phone. You can MCP into notion and the agent >> pretty much can take action. Yesterday uh charm is being shipped by Meta as well. Right. >> That's right. Yeah. >> I'm actually excited like all that Tamagotchi devices will come back in into uh into you know I guess into the pop culture again. >> What's your take on future of jobs? >> I think it's bright. I think it's uh there will be displacement but I think AI is going to be really net good for the world. I think there's going to be many more jobs created than we lose. I think people have to just be >> and they have to just you know I don't think AI will take their job but someone using AI will so they have to lean into it. >> I think it's still relevant. >> Yeah. U what is one thing about AI that blew your mind? something you never thought was possible but just happened. >> I feel like that even the four years into it, I feel like that every day cuz I think the the how conversational it is feels like somewhat like >> it's just getting better though. >> Yeah. It's like kind of incredible, right? It's uh it's like my wife's like, "Oh, this person listens to me more than you listen to me." It's like, "Yeah, I I agree, >> dude. That's exactly what my girlfriend says. You talk to your AI more than you talk to me." My god. Okay, that makes two of us again. Uh, one thing that you would never let your AI do for you. >> I love that AI gives me more time with my kids. >> I kind of guessed it. You're so much you you've mentioned it couple of times and like kind of guessed it, >> but okay, this is good. >> So, I would like to like, you know, like think of it as like giving me more time with my kids rather than less. >> Yeah. >> If you were not building today, what would you be building? something with my hands that doesn't involve screens. So, I think my current plan is if I were to get some more time off is to join a culinary school and make food. >> Be a plumber. >> Be a plumber. >> They said >> I like making food more. >> Yeah. I mean, that's what I meant, but yeah. >> Uh, what is one of the most underrated AI product that is not yours that everyone should probably give it a spin or something that you just came across and like why aren't more people talking about it? Tesla full self-driving. Uh, >> everybody talks about it, bro. Positive or negative, but everybody talks about it. >> I think it's still vastly underrated. >> Yeah, that I feel. So, >> I think 18 months ago, it felt like, oh, this thing is going to kill me. 12 months ago, it felt like >> maybe I can try. >> This is pretty good. 6 months ago, I became a subscriber. I have not driven a car in two months now. And it's so good. It's maybe the most underrated AI product right now. This thing is driving me from here, taking me 150 mi through the city, through the highways, through the forest, and parking the car for me, and I'm literally like sneakily looking at my phone so that it doesn't look at my camera and tell me that I shouldn't do that. >> Did you see that viral video where they where the guy tricked the camera? >> Yeah. Yeah. Yeah. And it's like, yeah, I I I can't wait for them to remove the steering. >> Steering. >> Uh because like what am I doing? >> You don't need me there. >> You don't need me there. >> That is true. I think Elon finally made it with just the cameras. >> Yeah. >> So good. So good. >> Awesome, man. This is incredible. Thank you so much for your time. >> Awesome. Great to hang out. >> Thank you so much. Can we take a quick picture? >> Yeah. Great to hang out with you. Yeah.

Assistir ao vídeo original no YouTube

Assuntos: Agentes de IA, AGI. Publicado no YouTube em 2026-11-08.

Seleção automática por assunto e idioma. O vídeo pertence ao canal de origem.

Comentários

Postagens mais visitadas deste blog

How to build a JARVIS AI Voice Agent for FREE | Full Livekit Tutorial (2026)

How Does AI Actually Work? 🤖 | AI, Machine Learning & Generative AI Explained