Loading summary
Transcript40 lines
- [00:01]
Kieran
Okay. This is one of my favorite episodes of the year because it's an entire episode about my favorite product in 2024, that is Claude AI. This is an episode with Scott White. He is the product leader for Claude. And we get into my daily usage of Claude. What are some of the things that I should be doing with Claude to integrate it into my work to be way more impactful? What are some of the ways that you can start to use AI if you haven't touched it this entire year? Let's get into today's episode of Marketing against the Grain. Okay. We're here with Scott White, the product leader for Claude AI, probably my most used product of all time.
- [00:52]
Scott White
It's your favorite AI product of all time. Let's get real.
- [00:55]
Kieran
It's my favorite product of all time. I use Claude more than any other product that I think I've ever used outside of HubSpot, of course.
- [01:01]
Scott White
How many hours a day you think you're using Claude or like, what measure do you want to use?
- [01:05]
Kieran
It's just my daily co person. Every single thing I do is mostly checking in with Claude. You might actually tell us, scott, don't do some of these things. That's fine. But last night I decided to start to invest in a new fund and I used Claude to do a bunch of the analysis and figure out my dollar cost average and things like that. I got a bunch of tests I had to do recently for my health. And one of the things I really think Claude is great at is if you just continue a evergreen chat. I just continue to give it all of my results and then I can ask it. Okay. Actually, can you remind me last year, what was the thing I did? And I obviously use it for writing. I've done a ton of content on LinkedIn about how I use Claude. It's just my go to thing. I'm like, oh, I have to do something. I'll go and talk to Claude and then figure out what I want to do. So good job, Scott. You are all building a great product over there.
- [01:55]
Scott White
Yeah, thank you. I think about this a little bit. Similarly, Claude to me is a virtual collaborator. That's how I've sort of talked about it internally. And it is spiky at times. Like, it is something that I use every day and feels indispensable for me for small things like even at home trying to decide movies that I want to watch and getting movie recommendations based on things that I previously liked. Similarly, I do my personal finances and that's like a spiky thing that happens every once in a while and doesn't. And then there are also really important crucible moments where I need to do a ton of work. And historically, it would have been something that would take me weeks of figuring out my strategy or my pitch for something. But now, Claude having access to all of my content, I can sit down and actually get something that would have taken me a week to do before done in a day. And it's so flexible in that. But I think this concept of a virtual collaborator sort of spans the entire thing from something that I just want to have in the background that I can reach out to really easily to answer questions or make progress on something and get unblocked. But also the thing that I typically would have had to get a group of people into a conference room and, like, whiteboard something for a week, it's also that depth of collaboration. So, yeah, it's similar for me.
- [03:13]
Kieran
Yeah. What's interesting about Claude and AI assistants in general is what people fail to realize is the more context you give it over a longer period of time, the way more impressive it is. And so maybe the first time you ask Claude for something and give it, like one document, it's fine. But then if you continue to use that chat window for a very specific task, like, I basically have many different Claude instances that are singular tasks, and every time I go back to that task, I just go back to that singular chat and give it more context. And give it more context. And so I think your opinion of AI changes drastically based upon how long you've used that kind of chat window to actually do that singular task and giving it more context, because the results are like, pretty huge in range. And I don't know if that's something you kind of agree with and see as well, in terms of how people use AI.
- [04:07]
Scott White
Yeah, absolutely. I think, to me, if I break down this concept of a virtual collaborator, it kind of goes into three segments, I think. One is what Claude understands about you, about the world, about the situation around what you're trying to do, how Claude communicates with you, and how it understands your preferences, understands how you like to receive information and your outputs, like actually what can it display to you and in what novel ways can it communicate around information to you. And the third is, what is it actually capable of? When I want to do something, can it write a document for me? Can it go and do something in the world for me? Not yet, but who knows in the future? What is it actually capable of in terms of taking something that I would have historically had to do off my plate and doing it itself. And I think right now, to me, there's so much opportunity in Claude, just understanding more. Obviously, it's a really smart model that has been trained on a lot of information, but like you said, giving it access to more information, increasing its knowledge base about things that are important to you, makes it move from just something that can answer questions to being an indispensable sort of collaborator in your world. And I think, like right now you have to go back to these chats. I like use a project, but I also sometimes use chats. I'd actually argue it's still too hard to do that. Like you have to have this mental overhead of remembering that you have various tasks where you go to various places to do these various things. In the future, I'd love Claude to just understand that stuff globally. You don't have to think about the cognitive load of navigating our product and our architecture to figure out where and how you interact with Claude and and what it should understand based on what you're trying to do. I would love to move to a world where no matter where and how you're interacting with Claude, it sort of understands the task specific knowledge that it needs to invoke to do what you want it to do.
- [06:05]
One of the things we're actually telling everybody, listening and watching today is either through chats or through a feature like projects in Claude. Just outline a handful of use cases that you want to just use that assistant and AI for, label them like that from the start and come back to them anytime you have that use case. It sounds very simple, but people don't really talk about doing that. And that way it's kind of just like a forcing function for you. Like, oh, you can easily say, oh, I use Claude for these five things. Anytime I'm doing one of these five things, I go back and use Claude and it builds that habit in a very powerful way.
- [06:37]
Kieran
Yeah, I guess the thing you're kind of moving towards, Scott, is like there would be an orchestration layer. So maybe Claude is dividing all of those tasks into subtasks, but you actually don't see it call in those subtasks because there's like a orchestration layer. So you're always interfacing with Claude and it's doing all of the thing that I do in the background, which is I'm, oh, I click on this historical chat, I click on this historical chat for that task. But eventually Claude will be able to do that part myself. But the thing I wanted to kind of maybe pivot into is because I create so much content in Claude. Like, there's two questions I continue to get asked all the time, which is, you should get an affiliate commission for Claude. So I just want to, like, put it out there, Scott, that people do say that about me.
- [07:14]
Scott White
Always be making money. Kieran, nice work, my friend.
- [07:16]
Kieran
Yeah. Oh, I love to always try to make money. The second one is actually, why is Claude so good? Why do you like it? And that one is a hard one for me to answer because I can look at the benchmarks and I can see how well it performs and how great it is, all these different tasks. But, you know, I don't think about the benchmarks. I just think about my experience with Claude and it's hard for me to articulate why. I just gravitate towards Claude all of the time. What I would love to hear, and maybe start with is you're in a pretty competitive market like Claude. Google, OpenAI, Grok, when Elon gets his 100k cluster trained and has Grok 3 coming out. And I think everyone is kind of moving in somewhat similar direction where there's going to be a personal assistant for your life and like a knowledge coworker within the enterprise. And so I would just love to hear maybe your perspective on building the product. How do you think about differentiation or how Claude separates itself from what all of the other companies are doing? And I would like to flag that. I think you have separated yourself enough that OpenAI have copied you a bunch of times. So I think that's a testament to how great at work that you're doing.
- [08:15]
Scott White
Awesome. Well, as we spin up our influencer marketing motion, I'll make sure to contact you for affiliates.
- [08:21]
Kieran
Always be asking. There you go.
- [08:23]
Scott White
That's amazing. It's a great question. And I think I would go first to the highest level mission of what we're trying to do, which is make safe AI systems the standard and accessible to as many people on the planet as possible. And that's supported in actually a number of different ways within our business. Like, we have an API that people can use to implement our models within the software that they're delivering directly to their customers. And that is one way that we are distributing these safe AI models that make it the standard within the world. But it's really important that Claude AI can also be that, because it is the billboard for safe AI. Right? It is the way that people think about what this system looks like, how they interact with it, what anthropic is what CLAUDE is. And it gives sort of a face and a name to what AI systems that are safe look like in the world to people that can adopt them themselves without having to be a developer. So it's a really important lever for us getting these systems out to as many people as possible and making them a meaningful part of their lives. So I'll talk about those two things I think are important when I think about Claude being distributed widely and meaningful. One of the key areas that I think about is how it can advance professional development for people. And that's an area where I think we really, really want to focus is how can Claude be that capable collaborator that I was talking about, especially in a workplace context and along the dimensions of what I talked about earlier. What does it need to understand? How can it communicate with you, and what can it actually do for you in the context of your work and what you're trying to do? We want to be at the frontier and excellent at those things. So if I think about advancing on the dimensions of Claude understanding for you and helping you in your job or your role in your professional advancements, you can look at things like the integrations that we've shipped with Google Drive and GitHub, and accessing organizational context, allowing for multimodal PDFs and analysis of that content. The model context protocol, which allows you to build flexible ingestion of knowledge into your repositories to then work with. These are all ways that we're very excited for CLAUDE to understand more of the context in your world, especially in the professional environment, to advance and be a better collaborator. And then you can also look at previous things that we've done, like artifacts, helping people create the most meaningful work product outputs, not just answering questions, but actually getting things done that exist in the context of what your role is. And so that is an area that we're very excited about, is advancing professional development through being a better virtual collaborator for those use cases. Something that you said, like, you don't always know why you want to go back to claude. And it can be amazing for a use case like content generation, because you've put a lot of your content in there historically, and we're uniquely good at content generation with our model. But there is something more mercurial about Claude the model in the sense that it is collaborative, it is a virtual collaborator, and it feels that way. And it's like saying, why do I want to go and get a drink with a particular friend? Like, is it because they have, you know, a specific job Or I've known them for a specific number of years or, you know, they are someone I look up to because of their professional advancement. No, it's because they're my friend and I like them.
- [11:49]
Kieran
Yeah.
- [11:50]
Scott White
Plod ultimately is a model that people like spending time with. And being a model that you like spending time with means that you will give it more information, it will be more proactive in working with you, and it will feel more like that collaborative force that you end up doing more meaningful things with. And so that is a component that is harder to see in terms of features that you build, but is a lived experience of how people think about working with the product. That's really important to us.
- [12:18]
Kieran
Yeah, that's actually two good points. So when I think about it, actually I was obviously like a lot of people originally, pretty heavy user on ChatGPT. And what pulled me over to Claude initially was the content creation. It was just much, much better, was far better write an assistant, it was just a better collaborator. Like I felt it was just like a better fit for me in the way I wanted to kind of go back and forth. And it's interesting. The tool, like an AI assistant can be sticky for those reasons because there's no other product in software. You're like, well, you know, I just kind of like feel it's a better fit for me. Like my personality, I work really well with it. Where most software products, the way you decide to use them is quite binary. Like it has the feature set that I want. Whereas because Claude and AI itself is such a broad usage. Right. Like you can't distill it into has this feature, that feature. Although I think artifacts was another one of the reasons I stuck around, like really got me sticky. I love that feature. And so if you actually had said to me like, well, how do you think we're differentiation? I would have said it's pretty obvious you're building for enterprise. Like I feel like the features you're building for enterprise. We have a pretty large squad in HubSpot building AI across all of our go to market teams. And I'm. That's one of the pods that I lead. So I spend a bunch of time with other companies just learning about specifically how they're applying it to go to market, not how they're applying it to like product. Because I think that's a pretty common thing for most companies and maybe a slightly separate thing. And I would still say like the majority of companies are not even at the kind of getting started stage, they're still at the like, what do I even do here? They're doing the prospecting, they're doing the cold outreach. And so what do you think are some meaningful use cases either now or even you think ahead to next year that you think like enterprise companies should seriously consider Claude and AI for?
- [14:04]
Scott White
Yeah, this is a great question. I think about this on almost two parallel tracks that I've seen. I think there are use cases that exist where Claude is and will continue making people dramatically better and operating faster than they have historically for tasks that are sort of still being done today within these roles. And so here I think about things like content creation, what you were saying before. I've seen a lot of repeatability in marketing teams centralizing a lot of their content around their style guide, their Personas, their audiences, and merging that also with product documentation, what they actually support internally, what their products are, the specs that they're writing about, things that they want to launch in the future and merging these things into the actual messaging that you're writing about that product for these various audiences, often internationalized in different countries. Like these are things that maybe you would have done always or you wouldn't have done them as well because doing that would be really challenging and require a large marketing team that you now can do way faster with way higher quality with centralized information. And so there are use cases like that that also exists for things like sales teams. I've actually, I did a bunch of user research last week with our sales team. One thing that was really compelling to me was one of our better account executives, one of the best account executives I've worked with. I've actually worked with him at multiple companies. He said, in my old role, I would have made one really good strategic account plan to drive a strategy for winning or expanding an account in a quarter. I would try to have each one of my account executives create one of those that really could be strategic and hit it out of the park. He's now like I do 20 with clock. I've 20x'd the output of actual strategic account plans that I can prepare for executives to increase their leverage and engagement in my deals. And that's something that I was not able to do historically. I think this also exists in software development, how people are writing code and like Claude, writing a lot of the code, understanding a lot of context about your existing code base so that it can help you write that code and also merge it with internal documentation around what you're building. Specific features like these are things that might not exist in a point solution externally that you're using specifically for coding, like in your ide, but it's actually important to have that context around what you're actually trying to build. So these are a few of the use cases that exist sort of today. The thing that's also parallel to this though is it's not just about jobs that are evolving because they have access to this new tool or capability that changes the way they think about it. It's also people expanding their scope at the same time into like other job functions and like the lines becoming blurred, blurrier between how job functions are structured because people now have expanded sets of capabilities. So I see things like designers internally writing a lot more code than they typically would have or they would never have written code before. But now the delta between them being able to push the things that they have in their brain into production is so much smaller. They can now do that. You know, salespeople helping think about our marketing campaigns and our marketing strategy based on what they're hearing in the field or synthesizing product feedback from the field that they give into the product team that traditionally like wouldn't have really happened or would have needed specific people to exist in the organization to cross that bridge. But now that bridge is just cross between those two people. Engineers creating like low fidelity mocks, using artifacts to communicate to other people what they want to build and then actually get it prioritized and built. So the thing that's interesting to me also is how these job functions will sort of overlap in Venn Diagram circles. And I think we're going to start to see a lot more of that next year as things become more connected and more capable. But I think also talk about the enterprise thing. I think you're right. Deploying this for professional development means deploying this within businesses. And doing that well also requires solving for the unique needs of these large businesses where you can't just now give like everyone access to every tool. You need to be cognizant of security and compliance and administration and permissions and role based access control as you do these things. So as you're progressing towards this future where job functions look more Venn Diagram overlapping, it's actually really important to partner with the enterprises where this change management is going to happen so that you can also build the foundation underneath it to support gracefully moving between those two phases.
- [18:59]
Hey, it's Kip. If you listen to this podcast, you know how much I love keeping up with the latest and greatest in technology, but very few podcasts actually give you a dose of the future. The A16Z podcast is the exception. It's hosted by our friend and frequent guest Steph Smith. The chart topping show brings on movers with a track record of being both early and right. Like Apple co founder Steve Wozniak or the CISOs of OpenAI, Anthropic and DeepMind. Even the very first CTO of the CIA. From the science and the supply of GLP1s to drone delivery to the economics of deepfakes, Go ahead and eavesdrop on the future. Check out the a16z podcast wherever you get your podcasts.
- [19:51]
Kieran
Right. I have so many follow ups I'm going to try and stack rank them here in my head because I want to kind of go back to the marketing use case and maybe just show an example of that in action with Claude so people can really understand the value. The thing that you said is that because I'm like looking across the entire go to market as AI has been integrated into a bunch of places and the thing you said about AI actually changes functions like the roles. It really does because if you break roles down into a series of tasks and AI can start to do a bunch of those tasks, then a person can actually do a multitude of different roles if they actually understand how to do AI. And then I would love to kind of get into maybe the other thing you said which is like getting started into the enterprise can be much more complex and I think part of the problem is this cold start problem and I wanted to kind of riff on that with you going all the way back to the marketing portion here. So we kind of had talked previously before and I give you a little tee up to One of the things I started with Claude was these kind of writing styles, right? Like I would create these different writing styles and I truly believe that in the future you won't look at like this email or that email or this blog post or this piece of content. You'll just look at like the performance of different writing styles across your go to market and you know what kind of performance that this writing style have versus this writing style and Claude actually have that feature. I actually did a video, people loved it that Claude now have that feature built in if you don't mind. My only tiny piece of feedback from someone who obsesses over writing style guides. I think they could be a little more comprehensive for an enterprise use case. Like they're too short basically, but I still love them. Again, what I love about Claude is I can flick between them, right? I was going through uploading this PDF writing this. Oh, I'll try a different PDF now. I can literally just flick between them. So I edit them myself and put them in so our audience know that one. So that's why I'm kind of going over. We've taught that one a lot and taught them how to do this Claude, the other one that I want to maybe just riff with you on, because it's another example of Claude doing the thing that you said, which is like ingesting data and then being able to do something. For enterprises when you have writing styles. The other thing that an enterprise would really love is like audience segments, like these kind of audience style guides. And so what it would be able to do is take all of your unstructured data and then be able to say, well, this is your ideal customer profile. This is the kind of way that you can consume content at the top of the funnel, middle of the funnel, bottom of funnel. And the cool thing about Claude is it can basically take the unstructured data, create a first pass for you, and again, just allow you to select now the writing style and the audience style guide. And I'm just curious, like, how do you even start to figure out what one you should focus on? Because there's so many different use cases. Like, I give you an example of one that may or may not be interesting, but, like, how do you decide, like, writing styles is the thing we should integrate. This is the thing we should integrate. Because enterprise has such a plethora of different use cases across so many different functions.
- [22:24]
Scott White
Yeah, it's a really great question. And it's a big challenge for an organization that's building something that's so broadly applicable. And we want it to be so broadly applicable to hit our vision. But it's also important to offer specificity to companies so that they understand and have a mental model for how they should adopt and use something. So this is something I think about a lot. And I previously worked at a company called Airtable and that was very similar where we were building a very flexible application building platform. And we saw this too. To me, there is this balance that exists where if you look at use cases, ultimately a lot of them share some fundamental building blocks. And you can sort of think about this as like, we are building a LEGO kit and this LEGO kit has like a combination of, you know, one of the Legos is sort of like the knowledge that Claude has. One of the Lego kits is like its ability to augment or like the prompt around the outputs that it has. One of them is like the specific form factor of the output that you are getting. There are all of these LEGO blocks that enterprises though, are ultimately looking for the Millennium Falcon. And there might be, you know, Eiffel Tower or the Millennium Falcon or particular like LEGO kit. And we need to provide to people good jumping off points to understand how you can pull these primitives together to actually accomplish the specific thing that you're trying to do, like achieve the Millennium Falcon. And in many instances, it also might require building the piece that is also specific to the Millennium Falcon kit.
- [24:00]
Kieran
Right.
- [24:01]
Scott White
So if you're an engineer, you can't just have a generic integration that the document repository might be necessary but insufficient for you to write a lot of code. You also might need the GitHub integration, and that is like a primitive that is largely just ingesting context, but it's doing it specifically for engineers. And that looks more like the Chewbacca LEGO piece that's going in the Millennium Falcon LEGO kit. So I think for us, it is going to be about making sure that we're building the right primitives that we think are horizontally applicable, that can be applied to a lot of use cases, but making them composable enough to bundle them together to be really meaningful for specific audiences. And we only understand which audiences we should be really solving for based on feedback with customers. And that's why partnering with enterprise is so important, because you have this direct line of feedback. I am in shared Slack channels with these people where I can answer their questions and see what they're trying to do. And then that shortens the loop from me being like, okay, this is how we need to bundle this. This is what templates we should provide, this is what specific integrations we should provide, or specific flavors of building blocks that we should provide to people to then get them to solve a very specific use case. So that's a little bit of how I think about it. I still think we're a long way to go though, on pulling these things together into a connected experience with better starting points that can address the cold start problem that you were talking about. I think right now we've done a pretty good job, in my opinion, of starting to build the right layer, the right foundational layer. But we have not done a good enough job yet of connecting these things in a way that people feel like they can start and really understand. Like, how do I solve this for content generation, for marketing, in a way that's accessible on the first time that you see it?
- [25:50]
Kieran
Yeah, that makes a total of sense because you're really trying to figure out what are the building blocks of enterprise, how does enterprise work get done? And then probably have a product that's flexible enough that if you have the building blocks, you can then create the nuance that you need to apply them to your function. And there's a spectrum, right? Like there's probably a spectrum of least complex to most complex in that if you're an engineer, it's somewhat easier to say we just need to be a great coding assistant. I'm sure there's more to it, but you kind of know what an engineer is trying to do day in, day out. Whereas a marketer can be doing a spectrum of different things from product marketing to email marketing to all these different things. And so there's way harder to decipher, like what are the core building blocks there. And so I'm curious to kind of get into the cold start problem that I love to talk about is are you to begin with prioritizing any kind of specific functions in enterprise or just like enterprise in and of itself? Like, how do you start to prioritize where to start?
- [26:41]
Scott White
Yeah, I think that a lot of this comes back to where we have seen early success and where we get pulled to inside of the enterprise and business context. And I think about a few that get me very excited just based on the really positive feedback that we've seen. I think software engineering is obviously one that has been really successful for us. I think if you look at the discussion around Claude 3.5 Sonnet and how amazing it is at coding, that is an area that we continue to be really excited about and we have a lot of customer demand and love for. But alongside that, we've seen a lot more use cases starting to take off within business. I think content generation for marketing, specifically, like marketing content generation on different specific marketing channels, where you're internationalizing it in multiple different countries. This is like a repeatable use case that we've seen and gotten really excited about. This is less of a job function, but it is more of a way of working people who are analyzing information within their businesses and using that to create recommendations or strategies or outputs. This could be people who are in accounting, this could be people who are in strategic finance. This could be consultants, this could be operations professionals. The shape of what they do is actually quite similar though. Like they're pulling together a lot of disparate information, analyzing and synthesizing that information and providing recommendations or outputs or models based on that. And the inputs are very different. Right. It could be stuff that you're getting from the Internet. It could be stuff that you're getting from financial models internally. It could be your own existing enterprise data on how people are using your product, or it could be account research. But ultimately the shape of what they're doing is very similar. That is another one. And then sales, I think, is actually another one that I'm super excited about, which actually interestingly follows a similar shape that we've seen where people are finding a lot of information about their accounts. They're creating strategies to target these specific accounts and how to go after them. Then they are, of course, meeting with those customers, managing those communications, and then downstream. Like, they, of course have to work with their CRMs, HubSpot in order to actually make sure that they're driving visibility and repeatability in their business. And we're excited also to explore this space and how we can help. We don't want to replace any systems of record. That's not what we want to do. But we do want to help people work better with their systems of record.
- [29:08]
Kieran
Yeah. Yeah. What's really interesting is when you start to really think about AI and jobs is again, all jobs are a series of tasks. And when you start to apply AI to different jobs, you realize that the roles are separated into different functions because of domain expertise that you apply to the same set of tasks. But if you don't need the domain expertise because AI has the domain expertise, then you actually start to figure out that a lot of these roles can just be aggregated into the same role, because if they have AI, they're able to just change the inputs. They have domain expertise, that AI, and they do the same series of tasks. The other thing I'm thinking about, as you're kind of talking about the different use cases, the thing that I find, like, really hard to figure out, or I would find really hard if I was in your shoes, is like, what do you want people to build on top of Claude? Again, it's such a broad set of things you could build for. But then there's all of these other companies wanting to build on Claude. How do you think about the balance between what you all build and then what you happy for other developers and builders to build on top of Claude?
- [30:07]
Scott White
Like, my vision for Claude is again that it can be this virtual collaborator that can help you understand the context around your world, communicate with you in ways as you would a collaborator that you work with today and capable of helping you do things and act on your behalf, which ultimately means helping work in the systems that exist today. And so in many instances, it's not about creating, competing or replacing the systems that exist today and where people are working, it's about complementing and making you work better with those systems, with the information that underlies those systems. So I think about it as like, we want to be a place where you go to work with all of this. But at the same time, I think offering choice is quite important. We have choice in the sense of we're available through CLAUDE AI, we're also available through our API. And different companies are going to have different considerations around the solution that they want. An example of this is maybe something like HIPAA compliance, right? We're not at the stage of being HIPAA compliant as a, you know, product, but you want to build a knowledge repository of patient information for doctors to create notes. This is just a hypothetical example. Cloud AI probably is not the solution that you want to use for that. But because we have an API and it's available on something like, you know, bedrock, you can now build an on prem service that leverages our API to manage how your doctors are writing notes using our API. So I think choice allows the market to select based on a number of different criteria where they are in terms of their maturity and their stance on security, compliance, administration, their cost considerations and the frontier of what they want to build. It allows them to mix and match different solutions that we can provide. Because we offer choice. We have some companies where for a subset of their use cases, they might want to build something in house on our API, but then for a broader set of knowledge worker use cases, they're super excited to adopt Claude AI. So I think we always want cloud AI to be the place where people can start. But the reality is that there are specific systems of record, very deep workflows that we're cognizant that we never want to replace, but we want Claude to help you start those things, interact with those things, understand those things more effectively, and it should be able to do that. But if you have a very deep domain specific thing that you're trying to get done, it might make sense to build into a point solution. Or if you have a specific stance on security and compliance, you might want to do it there. Or if it's too expensive to do this, you might want to build it yourself. These are all sort of the dynamics or things that why it's helpful to offer choice 100%.
- [32:40]
Kieran
A lot of people are want to know, like, how do you get using it every day? Like I do, like, it's a cold start Problem. Right, because you worked in Airtable, I worked in Zapier. Both of those companies likely had such a variety of ways that you could start people. So when you're trying to figure out oh, what do I onboard this person onto, how do I get them to activate AI is like magnitudes order of greater than that because it's like whatever, whatever you want to do. And so part of your, I guess, challenge will be how do I get people in companies using AI more? What are some things that you're thinking about to get people into the usage every week, every day of Claude and AI?
- [33:15]
Scott White
Yeah, it's a really great question and I think it looks different for individuals who are adopting something versus people who exist in a business and are coming into a place where there is already like an architecture of how other people are using the system. I think that there are some commonalities between them though. I think a lot of cloud AI ends up being around a combination of context and intent. What is the information that you're working with to try to do something? What does CLAUDE need to understand? And then what is your intent with how it's going to use that information to do something on your behalf? There's also something interesting where you could infer intent from context. In many instances, like if you have enough context, then you can say here are some good starting points for things that you could do if your intent was X, Y and Z. So I think something to me that I think about is how can Claude, in both the business context and the personal context, even if you want to adopt it and start using it for work, how can it get the necessary context from you in terms of what is timely and relevant? Or top of mind to then provide inspiration for you on how it can use that context to get something done for you to make your life easier to accomplish a task that might be timely and relevant. So that is one way that I think about it in businesses, a lot of that context already exists, right? You've already adopted Claude. You might have a bunch of projects, you might have a bunch of ways that other people have used it. And I think there it's about surfacing valuable things that other people have done who might be relevant to you based on what your function is, what kind of information you work with, and giving you just the jumping off point to say, let me just click a button, understand, learn, and then build a repeatable habit or loop out of that for an individual where a lot of that structure doesn't exist already. I think it's about understanding the basic foundations of them and their goals and what they're trying to accomplish and the world and the context within they are doing that so that we can formulate a recommendation about how they should start based on the intent that we're able to infer.
- [35:16]
Kieran
Yeah, yeah. So it's basically like the more you can get people to add context, the more valuable those suggestions are going to be. Whether you're in a HubSpot and it's able to suggest things to you, or you're in another app, or you can actually give it a bunch of your previous content and it can write something very similar. Like in terms of the user experience there, do you think that user experience is always going to be, oh, I have to remember to jump back to Claude and go into my assistant or have you. You aren't really thinking through that problem or like you think that happens in the workflow in some way? Like, how do you think about how the user experience looks for that?
- [35:54]
Scott White
Yeah, I think this is something that we're earlier on, but we are thinking about a lot. And I think if you look at things like Computer Use and the sort of like API that we released earlier this year around that.
- [36:07]
Kieran
Scott, could you just, for the listeners, explain what that is, because they might not know what you can do with it.
- [36:12]
Scott White
Yeah, absolutely. So Computer use is this interesting model capability that's available in an early sort of preview via our API, which allows Claude to take screenshots on your computer, identify coordinates, and then click on those coordinates based on the task that you're trying to accomplish. So if I say I want to go and, you know, search for something that I can buy on Amazon or something, I want to find a particular object, I can ask Claude to do it. It can then control the environment, navigate to the browser, open the browser, look for Amazon, and then click on those coordinates. It's still very early, but that's the kind of thing where you can imagine right now Claude AI is really existing in this web application. Who knows in the future if it has access to more of those capabilities? It does change the form factor of how you think about where it is in your life. I do think about Claude AI right now as being the place where a lot of people go back to. And we do want to make it easier to reach for things like our desktop application, things like our mobile application. Making those continually more accessible, I think will be important. But I do think meeting people where they are is crucial. And that might mean in other point solutions, it might mean on your desktop, it might mean while you're walking and you want to speak to it, we do want to make it easier to reach for in those instances. And we also want to eventually make Cloud more proactive in reaching to you when it needs to, to make that also easier based on the context and the content that it has. So I think that will be important for us.
- [37:46]
Kieran
Yeah, that will be huge. I can imagine. Maybe I'm off base here in that it already can do this in some way, but I think at some point the AI assistants and we're specifically talking about Claude, so we'll stick to Claude, but we'll be able to automate a bunch of the work and the way that I suspect it will do that. If you're a knowledge worker today, most of your work is happening in browsers. And so if it can capture what you're doing day in, day out, it can probably suggest workflows that it can build and automate for you. And I think, you know, in that case you have to capture that information. But I do agree as well that it should be its own destination. And I can see, like, Claude is doing a good job because you have projects, you have artifacts. But the big unlock for me is when you can integrate it. Let's say I can have it integrated into like, Slack and email and all these different comms channels and I can just tell me what to do. Like, what is my unread Slack? What is like, you know, hey, these are the things you have to do today, Kieran, like, that sounds so simplistic. That is like the most impactful thing you can do for B2B knowledge workers is collate all of the comms, parse it and structure it, and then prioritize it. And I would say, like, every single person would be 10 times happier in their job if they could do that stuff and they could focus on the work. And so I think that's one incredible use case that I'm sure at some point will be available in Claude.
- [39:05]
Scott White
Yeah, absolutely. I think this goes back to that pillar I talked about on Claude, understanding. Like, you can really only prioritize the information for someone if you have access to the information. And so we're very excited for Claude to continue to get more information about you and your world and also merging it with your personal preferences so that it can then create more tailored recommendations of what it can do for you using all of that information. And so that's absolutely something that I'm excited about. I have maybe 65 unread Slack messages right now. And so Prioritizing that is going to be a tough thing for me. My entire commute is always just that. I do a commute. I live about an hour away from San Francisco and on the train here, on the train home. All it is is clearing out the communications backlog to then prioritize how I need to be communicating with other people. So it is a huge challenge for me.
- [39:57]
Kieran
Exactly. Yeah. Same. It's huge. This is awesome. Scott. I'm going to end with some really self serving things and you can ask them or not because I am a power cloud user and I'm going to ask you for some things and then you can tell me. I can't tell you anything or we're not doing it or just like that's fine. Is Claude going to connect to the web? It's really the only time I really have to leave it is I'm trying to do something and it doesn't have up to date information. And again, because you're building for the enterprise, it may not be as important to you all. But I'm curious if you're able to share anything there or not.
- [40:28]
Scott White
Yeah, I think connecting to the web means lots of different things. It could mean searching, it could mean it easier for you to give it links. So I think this is something we're considering, we're excited about. I can't give you a particular timeframe, but it's definitely something I'm interested in.
- [40:41]
Kieran
Okay. My usage will go up even further. So that's cool. I think artifacts is one of the coolest feature you all launched that you launched the ability to publish artifacts publicly, which is so cool because like I have none. Geeky friends. Right. I shared Claude's analysis of this clean energy fund with my friend who does invest and doesn't really understand the value of AI and I sent him one of the artifacts that I published on Claude. He's not an AI user. He can look at it. He's like, what? I was like, how long did it take you to that? I was like, Claude built me charts, forecasted things, predictive models. I was like, it took me. I'm literally watching TV and this took me three minutes. He's like, okay, signing up for an account. The thing I think Claude is getting right is just there's these like really cool features. Like projects was a cool feature. Artifacts was a cool feature. Being able to publish your artifacts, such a cool feature. I think you've built an incredible product. It's really exciting to have you on having actually someone on who's built a product that has been really impactful for my life. Has been really impactful, I suspect, for listeners lives. And I just can't wait to see what you all launch next year. So appreciate you coming on, Scott.
- [41:45]
Scott White
Yeah, absolutely. Thank you for having me. This was awesome.