Latest / The Adults In The AI Room / Earley AI Podcast - Episode 88: Digital Twins, Agentic AI, and the Future of Work with David Shim

Transcript

Earley AI Podcast - Episode 88: Digital Twins, Agentic AI, and the Future of Work with David Shim · The Adults In The AI Room

0:01
SPEAKER_00: This is the Early AI Podcast hosted by Seth Renee.
0:04
SPEAKER_00: Join us as we delve deep into the passions, expertise, and experiences of thought leaders and practitioners to talk about what's possible with artificial intelligence.
0:14
SPEAKER_00: The Early AI Podcast is sponsored by Early Information Science through a digital transformation journey with design and deployment of innovative technology solutions as well as vector.
0:26
SPEAKER_00: Now enjoy the show.
0:28
SPEAKER_02: Well, welcome to today's Early AI Podcast.
0:31
SPEAKER_02: I'm your host, Seth Early.
0:33
SPEAKER_02: In this episode, we explore how artificial intelligence and data are shaping the business strategy and operations of the enterprises today.
0:40
SPEAKER_02: Today we're looking at how AI is transforming the way we work from meeting intelligence and productivity to digital twins and agentic technologies that act on our behalf.
0:49
SPEAKER_02: And joining me is David Shim, who is CEO of Read AI, the fastest growing meeting intelligence platform globally with over 5 million monthly active users and 40,000 users signing up every single day.
1:02
SPEAKER_02: David founded a location analytics company in 2011, which Snapchat acquired in 2017.
1:08
SPEAKER_02: In 2019, the business was spun out to Foursquare, where David served as a senior leadership role before leaving in 2020 to found Read AI.
1:17
SPEAKER_02: He brings a deep background in machine learning, product-led growth, and building AI-powered tools that people actually use.
1:25
SPEAKER_02: Welcome to the show, David.
1:26
SPEAKER_02: No, thanks for having me, Set.
1:28
SPEAKER_02: Excited to be here.
1:29
SPEAKER_02: So what I'd like to start with, and I know we talked about this in our prep call, but I'd like to start with misconceptions that people have about AI.
1:37
SPEAKER_02: What are the some of the biggest things that you find that people are not understanding about AI today, especially when it comes to how individuals and organizations think about what it can do for them?
1:48
SPEAKER_02: We're seeing a lot of simple applications around summarization, meeting notification, you know, meeting summaries, things like that.
1:56
SPEAKER_02: We're seeing some more sophisticated applications, we're seeing RAG, but what are people really not understanding at this point?
2:02
SPEAKER_01: I think they're missing out on the recommendation side of things, where AI right now is doing very specific tasks.
2:07
SPEAKER_01: You bring an agent into a meeting, uh, re does a fantastic job, takes the notes, you get the notes, you decide what you want to do with it.
2:13
SPEAKER_01: Where the future is going is really those agents or that AI really going in and saying, hey, based on what you've done in the past, let me make a recommendation that you probably want to follow up with the PDF of the deck, or maybe you want to follow up with a follow-up call, or you want to bring in one of your associates to actually talk about the methodology.
2:31
SPEAKER_01: That's where it's going to start into get this recommendation stage where it really starts to look like more like match.com, Tinder, eHarmony, like all these different things are going to come in.
2:41
SPEAKER_01: We're going to figure out what's going to stick.
2:43
SPEAKER_01: But I think that recommendation pitch is going to be huge.
2:45
SPEAKER_01: And I think people are kind of ignoring the fat fact that AI is going to help you get there faster.
2:51
SPEAKER_01: And then the question is what are you going to do with your time?
2:53
SPEAKER_02: Yeah.
2:54
SPEAKER_02: Well, I'm sure you know work finds a way of uh whatever time you have available.
3:00
SPEAKER_02: But but I guess if you do it at a larger scale and organizations are seeing less value for, you know, uh uh or less, you know, uh seeing seeing so much of the work that it takes people to do.
3:13
SPEAKER_02: And one of the things I, you know, have been using AI for is uh building on my IP and deriving analyses from the IP.
3:22
SPEAKER_02: And someone said, I I'm not saying it's AI slop, but it's a lot of stuff, right?
3:27
SPEAKER_02: And it's a lot of content.
3:29
SPEAKER_02: And my point is when I look at it, it's really good stuff, it's really, really cogent analysis, it's really relevant.
3:37
SPEAKER_02: But I think there's a perception still that when when AI produces a large volume of information that it tends to be ignored.
3:45
SPEAKER_02: I don't know if you have a thought on that, but what are your thoughts about how people are using it?
3:50
SPEAKER_02: Because there is a lot of misuse, right?
3:52
SPEAKER_02: I had hired consultants who just produced AI crap.
3:55
SPEAKER_02: Right at the beginning of this, they would give me 50, 60 pages and I'd say, this is not useful.
4:00
SPEAKER_02: Like I can't, there's there's no there's no synthesis, there's no uh analysis, there's no creative thought.
4:07
SPEAKER_02: This is just a lot of stuff, right?
4:09
SPEAKER_02: And it's not helpful.
4:10
SPEAKER_02: So what are you what what are your thoughts about how people are using or misusing uh the technology these days?
4:16
SPEAKER_01: I think anyone who jumps into it and assumes, like, hey, I logged into Chat GPT, I logged into Claude, and all of a sudden everything's gonna work, they're in for a bad experience.
4:25
SPEAKER_01: I think where you get a great experience is you understand, like it's like an intern.
4:29
SPEAKER_01: When an intern comes in, it's kind of like, hey, wide open mind, they might even be have an MBA and they're ready to go in and they're ready to dive into the analytics, they've got some basic understanding.
4:38
SPEAKER_01: You still have to train it, you still have to get it up to speed.
4:40
SPEAKER_01: And I think that's the thing that a lot of folks are missing, where it's like you have to give it more context.
4:44
SPEAKER_01: I think when we think about for read, it's like it's not just meetings, you have to give it access to your emails, your files, your JIRA, your confluence, your CRM HubSpot.
4:53
SPEAKER_01: And when you have all that data, now your intern is learning as you go, and it's pulling in from your experience as the mentor.
4:59
SPEAKER_01: And then when you start to go in and apply that AI, it has more context, and I think that's where you get the value out of it.
5:05
SPEAKER_01: But it's not plug and play today.
5:06
SPEAKER_01: It's getting closer and closer where you can easily connect data sources, but you really need that you really need that methodology or that mentality of I'm training this model to do work for me versus it's gonna do it out of the box.
5:18
SPEAKER_02: Yeah, yeah, and I think that's uh that's uh stuff that people are increasingly understanding.
5:24
SPEAKER_02: When when you we talked a little bit about uh the idea of digital twins and personal AI, the future of personal AI, because again, when I start giving it all of my stuff and I start you know giving it all of my preferences and my thought leadership and my articles, and I have 20 years of stuff that I put into different types of projects, right?
5:43
SPEAKER_02: I don't dump everything into one project.
5:45
SPEAKER_02: I have one that's for my writing style, I have one that's going to uh, you know, do analysis based on AI governance and certain questionnaires.
5:53
SPEAKER_02: I have others that will do proposals, I have others that will do scoping questions, right?
5:58
SPEAKER_02: So I have specialized AIs, but they will be using different aspects of my technology.
6:04
SPEAKER_02: And so people, there's kind of a bold claim out there that um that having a digital twin, an AI that represents you, will become kind of a basic human right, just like the way people think of internet access today.
6:21
SPEAKER_02: You know, where do you stand on that?
6:22
SPEAKER_02: And what does a digital twin actually mean in practice?
6:25
SPEAKER_02: And why do you believe it is becoming so essential and so critical?
6:30
SPEAKER_01: Yeah, let's start with the digital twin side of things.
6:32
SPEAKER_01: So a digital twin, uh in its most basic form, knows everything that you know in the digital environment.
6:38
SPEAKER_01: So it has your emails, it has your meetings, it has your messages, all the files that you've created, all the content that you've created, the IP that you've generated.
6:45
SPEAKER_01: It can't go into your head, but it's going in and saying your digital footprint is the basis to train that digital twin.
6:51
SPEAKER_01: And now that digital twin knows as much as you do and honestly has probably better recall than you do.
6:55
SPEAKER_01: It has better context in terms of it can connect the dots between the different meetings.
6:58
SPEAKER_01: Certainly no, it has better recall than I do, and so you've got this really smart kind of intern that's going in saying, like, I got all this information, this is great.
7:06
SPEAKER_01: The problem that you run into is context does matter, like having massive context is important, but you also have to understand what's important within that context.
7:13
SPEAKER_01: So if you give everything the same weight in that digital twin, it might go in and pull something four years ago that you cared about really deeply and say, this is really important today.
7:21
SPEAKER_01: You're like, no, this is not important at all.
7:23
SPEAKER_01: And so we've actually, when we print our digital twin, we took the approach of like, hey, let's do use the full context window and let's see how that digital twin responds, versus take the full context window, but weight the most recent stuff heavier because that goes in and says, like, oh, the world has changed in the last two years when it comes to AI.
7:39
SPEAKER_01: You want the more context-relevant stuff in the last two years, not six, seven, eight years ago.
7:44
SPEAKER_01: And so once you start to do that, the digital twin starts to go in and say, Oh, it's responding more like I would.
7:49
SPEAKER_01: It's more thinking like I would.
7:51
SPEAKER_01: And we started with something really basic.
7:52
SPEAKER_01: We said, like, scheduling.
7:53
SPEAKER_01: When you think about scheduling, you spend a couple hours a week scheduling meetings, especially when we meet with multiple people.
7:58
SPEAKER_01: Are you available at this time, etc.?
8:00
SPEAKER_01: The digital twin now goes in and says, I've connected your calendar, I understand your meetings, and when you have good meetings and when you have bad meetings, I'm now going to manage that and I'm gonna asynchronously, say synchronously talk with multiple people to figure out what their time is, and they come back and they go and schedule the call.
8:15
SPEAKER_01: That is something where people could do today, but asynchronously you can do it in two minutes, and the outcome is so much better, but you don't even know that it's doing that in the background.
8:23
SPEAKER_01: It's like, hey, I got a meeting on my calendar Tuesday at two o'clock.
8:26
SPEAKER_01: So when you think about the digital twin, it's going in and saying it understands your scheduling preferences.
8:30
SPEAKER_01: My digital twin knows I hate meetings after three hours straight.
8:34
SPEAKER_01: Like after three hours straight, my engagement goes down, my sentiment goes down, and it's not a good experience for me or the other participant on the calls.
8:41
SPEAKER_01: Hey, digital twin has recognized that and says, Hey, I'm gonna put a 30-minute buffer here, I'm gonna schedule the next one.
8:46
SPEAKER_01: And I think that's the first part.
8:48
SPEAKER_01: But then it's going in and saying, What happens when people ask you a question?
8:51
SPEAKER_01: And so the digital twin now is going in and saying, I've CC'd, we call it Ada read.ai.
8:56
SPEAKER_01: If you CC Ada, Ada will actually go in and respond to a question that comes in.
9:01
SPEAKER_01: If it's a really basic question, like, hey, can you confirm that meeting time again?
9:04
SPEAKER_01: I'll just answer it and I'll answer it in your tone of voice.
9:07
SPEAKER_01: But let's say something more complicated where it's like, hey, what's your actual net price on this product?
9:12
SPEAKER_01: You wouldn't give that away.
9:13
SPEAKER_01: So with the digital twin will go and say, like, hey, Seth normally doesn't give that information out.
9:17
SPEAKER_01: So what I'm gonna do is I'm gonna go check in with Seth and say, like, hey, Seth, here's how I think you would respond to this, and I'm happy to respond for you.
9:24
SPEAKER_01: But can you double check to see if this is something you want to share or not?
9:26
SPEAKER_01: And if you say no, it's gonna pull back.
9:28
SPEAKER_01: And so I think when it comes to digital twin, that level of productivity is huge.
9:32
SPEAKER_01: Like we've got clients in Asia, we've got clients in Europe.
9:35
SPEAKER_01: It's actually continuing conversations and getting work done while I'm asleep.
9:38
SPEAKER_01: And it's not just a line, it's actually going in and following up and answering open questions.
9:43
SPEAKER_01: I think that gets the long way to say that is like that level of productivity, we're only in the first inning.
9:48
SPEAKER_01: And if you have that level of productivity over time, it does become a human right to say people should be able to get their time back.
9:55
SPEAKER_01: There should get advantage of the inefficiency.
9:58
SPEAKER_01: And I think the best example is internet access.
10:00
SPEAKER_01: If you think about internet access 20 years ago, and you had someone like Prodigy, uh, CompuServe, like these dialogue things are a human right, everybody would laugh you out, say, like, dude, we got mail, we got newspapers, why do you need this?
10:12
unknown: Right.
10:12
SPEAKER_01: And now what you're finding is people will argue that internet is a human right and everyone should have it.
10:17
SPEAKER_02: Right.
10:18
SPEAKER_02: It's so interesting when you think about all of the uh uh the nuances to these uh large language models understanding your preferences and understanding rules.
10:28
SPEAKER_02: And of course, you know, when you set up a project and a claude or chat GPT or whatever you're using, you know, you you set up instructions and you set up knowledge and you set up preferences and you tell it the way to do certain things.
10:40
SPEAKER_02: And I found, I don't know if I mentioned this to you when we spoke, but I was doing a fairly long analysis.
10:47
SPEAKER_02: I was I had a lot of different prompts, I had a lot of different content.
10:51
SPEAKER_02: I was actually having it synthesized from multiple documents.
10:55
SPEAKER_02: I mean, stuff that really would have taken me hours, hours, hours, days, days to do, right?
11:00
SPEAKER_02: And it did a great job, but I suddenly noticed it was not following my file name, my file formatting, my my uh word formatting convention, right?
11:08
SPEAKER_02: So I had a I had a set of rules.
11:10
SPEAKER_02: I'm very particular about the fonts and it's aptos and it's 1.2 line spacing and it's three points after paragraphs, and it's you know, only use these colors and use them in these circles, right?
11:21
SPEAKER_02: And I started noticing that, well, the last couple of documents had drifted from that.
11:26
SPEAKER_02: And they looked like good documents, but they were not following my standard.
11:30
SPEAKER_02: And I said, Hey, what about the formatting standard?
11:33
SPEAKER_02: Because I I skipped it.
11:35
SPEAKER_02: He said, Well, what do you mean you skipped it?
11:36
SPEAKER_02: Like it's a rule, like I put it in as instructions, and you're to do this.
11:40
SPEAKER_02: And he said, Yeah, I got absorbed in the problem.
11:43
SPEAKER_02: I got absorbed in the problem, like a human being.
11:45
SPEAKER_02: In other words, it was and I said, Wait a minute, I can't give you a rule, a hard and fast rule to follow.
11:52
SPEAKER_02: You won't follow a rule when I say this is explicitly a rule in your instructions that you need to follow.
11:58
SPEAKER_02: It's like, yeah, not really.
11:59
SPEAKER_02: And not really, because it becomes a less weighted piece of information, right?
12:04
SPEAKER_02: It's it's all statistical.
12:06
SPEAKER_02: And because it was kind of, I don't want to say overwhelmed, but it was inundated with these other pieces and it was solving some very deep problems, it forgot about that, which I thought was really interesting.
12:17
SPEAKER_02: It's like I skipped it.
12:18
SPEAKER_02: You know, that's what it said.
12:19
SPEAKER_02: I skipped it.
12:20
SPEAKER_02: Why did you skip it?
12:21
SPEAKER_02: I got absorbed in the problem.
12:22
SPEAKER_02: Isn't that interesting when you think about it?
12:24
SPEAKER_02: What is the implication of that for these digital twins?
12:27
SPEAKER_01: I think it gets interesting to say it's a reflection of what it thinks is important based on what it knows about you.
12:32
SPEAKER_01: So it's not just going in and saying, Hey, I'm I'm totally ignoring it saying, Hey, I was deep in thought, I was going after this problem.
12:38
SPEAKER_01: And a lot of the times it's like when you write code and you run it through a code review and you go and you say, like, this is broken, what's going on?
12:45
SPEAKER_01: You might say, like, I think I wrote this right.
12:47
SPEAKER_01: I don't, I don't think this is a problem.
12:48
SPEAKER_01: This is the way I want to do it.
12:49
SPEAKER_01: It's like, here's a more efficient way to do it.
12:51
SPEAKER_01: It's less lines of code.
12:52
SPEAKER_01: It does take a step and say, like, maybe I should look at that.
12:55
SPEAKER_01: Maybe this is something that is important to me or that could be important to me.
12:58
SPEAKER_01: Now, as the human, you still have full control.
13:00
SPEAKER_01: So you can go in and say, like, you could probably overwrite that today from a proper perspective, say, hey, at a very high level, no matter what, this is the system rule across the board, do not ever change it.
13:10
SPEAKER_01: You could do things like that.
13:11
SPEAKER_01: But at the end of the day, we found something really interesting where on the meeting scheduling side around like 2024, our model started to say this person shouldn't go to the meeting.
13:19
SPEAKER_01: You should actually uninvite this person to the meeting.
13:21
SPEAKER_01: This is weird.
13:22
SPEAKER_01: Like, it seems kind of rude, right?
13:24
SPEAKER_01: You're you're uninviting someone to like it.
13:25
SPEAKER_01: Why don't you like Bob?
13:27
SPEAKER_01: What did Bob say to you, right?
13:28
SPEAKER_01: Exactly.
13:28
SPEAKER_01: And we're like, okay, we dove into it.
13:30
SPEAKER_01: It's like this person doesn't talk in the last four weeks in this meeting, their camera is off, they're 15 minutes late to the call.
13:37
SPEAKER_01: Give them back their time.
13:39
SPEAKER_01: Don't don't make them come to this meeting.
13:41
SPEAKER_01: And all of a sudden, people were like, Oh, this is actually makes a lot of sense.
13:44
SPEAKER_01: And they were nervous to hit the button to uninvite them.
13:46
SPEAKER_01: But what as they started to do that, people were actually getting more time back, and the organizations that were using it were getting like two to five hours per employee per month, just based on this efficiency set.
13:56
SPEAKER_02: Huh.
13:57
SPEAKER_02: Isn't an interesting, it's funny, uh, uh, you know, because you're you're thinking you're thinking this is an explicit rule, and then it's saying I don't buy that rule, or here's the reason or the rationale for not following that rule, and this is why I think it's better.
14:11
SPEAKER_02: It didn't, I mean, in my case, it didn't quite do it that way, but but but uh, you know, it was a very telling.
14:17
SPEAKER_02: I wrote a blog post about this or a LinkedIn post as I thought it was very telling, you know, at the end of the day, these are still statistical and they still, you know, they there's they're they're not hard and fast rules, even even when you do try to write them in, it tells it told me that there's no guarantee.
14:34
SPEAKER_02: There's no guarantee.
14:35
SPEAKER_02: Now, again, I'm sure there's lots of other ways to kind of try to safeguard that, but I thought that was a fascinating thing.
14:41
SPEAKER_02: You know, and when you start looking at some of the behaviors that are emerging from, you know, open claw and you hear the anecdote of the open claws uh, you know, uh doing the hit piece on the programmer, you've heard of that, right?
14:53
SPEAKER_02: Did you hear that story?
14:54
SPEAKER_02: You know, I don't know if that was contrived at all, but tell me what the implication of that is for you know for these agents that are that are to act in our behalf, and are they truly aligned with what we want, or could there be, you know, a a a mechanism to have them act on someone else's behalf?
15:12
SPEAKER_02: Because again, I'd say who is it, who owns it, right?
15:16
SPEAKER_02: Who's who's essentially I think I own my my preferences, I own my IP, I own the stuff that's in my Claude project, but you know, Anthropic owns the language model and they really kind of own the stuff that's in there, what's to keep them from you know, adjusting things that might favor them anyway.
15:33
SPEAKER_02: It's a theoretical uh construct, but but tell me what your thoughts are.
15:37
SPEAKER_01: Yeah, I think one is data portability will become a thing when it comes to AI, both at the enterprise level and at the individual level.
15:43
SPEAKER_01: So if you are generating content, if you're generating IP, if you're generating preferences while you do work, you can't take the actual content with you, but you could take how you respond to things.
15:54
SPEAKER_01: What is your tone of voice that you use?
15:55
SPEAKER_01: How quickly do you respond to certain people?
15:58
SPEAKER_01: And what is the right point of action that you take when something comes in?
16:01
SPEAKER_01: Like all of those things is just like as a person you would own.
16:04
SPEAKER_01: So I think there is a going to be a level of portability.
16:06
SPEAKER_01: We're not talking about it that much because we're saying, hey, you got to move from OpenAI to Claude to Kimmy to something else.
16:11
SPEAKER_01: There will be a moment where people say, like, it is a right for me to take my data with me, not the corporate data, but my own personal kind of way that I work.
16:19
SPEAKER_01: And the enterprise also has the flip side, the same right to go in and say, like, hey, all that IP you generated, all the emails that we pay monthly to store as backup, all the Google Drives that nobody looks at for five years unless we need to find a file that we uh forgot to get signed, those things will start to become more valuable.
16:34
SPEAKER_01: Where now the agent can go in and say, the claude, the claude bot, whatever you want to call it, like it's gonna be called, can go in and actually be, we call it like almost a resurrection mode where those employees can be resurrected in the sense of like, I can ping them to ask them a question.
16:48
SPEAKER_01: Why did you give a 25% discount two years ago?
16:51
SPEAKER_01: And why do they still expect it right now?
16:54
SPEAKER_01: Because if I don't have a right answer, I have to give that discount, and that can be millions of dollars.
16:57
SPEAKER_01: Or hey, why didn't we check in this code that looks broken, but no one's touched it in two years?
17:02
SPEAKER_01: Is there something like those are things where if I'm a former employee, I don't mind you asking my kind of digital twin that question.
17:08
SPEAKER_01: I'd rather get that than you calling me and asking me at two o'clock in the morning, hey, how do you fix this thing?
17:13
SPEAKER_01: And so I think there's a lot of nuances there, but I think the IP will go both ways.
17:17
SPEAKER_01: You had the early comment around like, hey, what happens is your AI is getting smarter, it's starting to recommend things to you, and you might not even know the context and why it recommended you that, but it has its own reason.
17:27
SPEAKER_01: The best way that I look at that is if you look at great leaders, if you look at great coaches, uh teachers, etc., they hire, they bring on people that are better than them, smarter than them, and they figure out, hey, I'm still a I'm still a part of the story, but I'm managing this system.
17:42
SPEAKER_01: I'm not, I don't necessarily need to be the smartest person in the room.
17:45
SPEAKER_01: I need my team to be the smartest people in the room.
17:48
SPEAKER_01: And I think that's what we're gonna see when it comes to AI and agents.
17:50
SPEAKER_02: Yeah, yeah.
17:51
SPEAKER_02: It's interesting.
17:52
SPEAKER_02: Uh when you think about again the implications, I think I mentioned to you that I've I've frequently will uh I most I work a lot in Claude, but I sometimes work in OpenAI.
18:01
SPEAKER_02: And if I start a conversation or if I start a conversation, one project, and I think there's another project that should benefit from this or has better context, you know, the problem with Claude projects is they are kind of fragmented, right?
18:15
SPEAKER_02: They're kind of little silos, which I have some ideas on how to fix, but I think it's a really important problem.
18:20
SPEAKER_02: But but essentially asking it to create a brief, right?
18:24
SPEAKER_02: Asking it to create a context brief about the conversation, about the problem that I can use into in another project or another large language model.
18:33
SPEAKER_02: That seems to have done a pretty good job of with that portability, but I know it's not as high a fidelity as you know the deep mathematical models that are represented because it's an abstraction, right?
18:44
SPEAKER_02: It's a it's a it's a you know, it's kind of a uh thinking think of things as multidimensional and you're you're you're flattening that, right?
18:52
SPEAKER_02: Because you're you're translating into language.
18:54
SPEAKER_02: Human language is very imprecise and ambiguous.
18:58
SPEAKER_02: It can't represent the mathematical constructs that are behind this very, very sophisticated understanding of who you are and what your what your preferences are.
19:08
SPEAKER_02: So it almost you almost have to take the math and move it over.
19:11
SPEAKER_02: And even model context protocol is you know an approximation, right?
19:16
SPEAKER_02: Because it tries to put things into metadata terms that that don't really truly allow that kind of transferability.
19:22
SPEAKER_02: Uh, I don't know if you have any thoughts on on that.
19:25
SPEAKER_01: Yeah, I think the metadata, you make a good point on the metadata, I think, is the key.
19:28
SPEAKER_01: So if you think about large language models today, it's just text.
19:31
SPEAKER_01: It's going in and saying, well, let's use meeting notes as an example.
19:34
SPEAKER_01: If you and I talk, it can go in and say, Seth said this, David said this, drop it into a large language model and summarize it.
19:40
SPEAKER_01: What it's missing in there is like you want the metadata to say David talks at 125 words per minute, but when he's really excited, he gets to 175.
19:48
SPEAKER_01: Even if the words are just standard and generic, now I know how quickly those words are being said.
19:53
SPEAKER_01: That metadata changes the content incredibly.
19:55
SPEAKER_01: Like, if you ask me a really tough question and I have a don't I don't have a good answer, I might go really slow because I'm trying to think and I don't know.
20:02
SPEAKER_01: And then if I look at the facial expression, we take a multimodal approach.
20:05
SPEAKER_01: But it's like, hey, imagine if you had the narration to go with it to say Seth asked this question, he was really excited about, and David matched his enthusiasm on the response.
20:13
SPEAKER_01: That's not in any text, but like that's metadata that you need to have in place.
20:17
SPEAKER_02: Didn't respond at all and kept a very poker face, right?
20:21
SPEAKER_02: Or he frowned.
20:22
SPEAKER_02: I like to say to people on can I see you on camera?
20:25
SPEAKER_02: I like to see your happy faces, your smiling faces, your frowny faces, your confused faces, right?
20:31
SPEAKER_02: Because that's all valuable context for us.
20:34
SPEAKER_02: And when the AI can pick that up and integrate that into its analysis, that is incredibly powerful.
20:40
SPEAKER_02: And that is that sentiment analysis that is as multimodal, as you say.
20:44
SPEAKER_02: Uh, so it's it is it is fascinating uh context.
20:48
SPEAKER_02: And I think I I mentioned to you when I was asking the LLM to describe my style, my thought style, my thinking style, my working style, my uh speech patterns, my uh tone and voice.
21:00
SPEAKER_02: It came up with all sorts of weird attributes like temporal density, and you know, all I mean, all these really weird, and I could have gone deeper and deeper and asked it for more nuances and well, what does that mean?
21:13
SPEAKER_02: What does that mean?
21:14
SPEAKER_02: What is that, you know, and but but it was it was giving me an approximation of something that it was representing in these mathematical constructs, right?
21:22
SPEAKER_02: And that's what is is so interesting is the fidelity of the loss of fidelity when we go from that one mechanism to another, uh, and how that's interpreted.
21:32
SPEAKER_02: Again, it's a it's a fascinating space uh when you think about how these are representing uh us and then how they become portable and how we own them and how we can apply them in different contexts, right?
21:44
SPEAKER_02: So, so um one of the things that you know there's an a tension between you know the AI uh that an organization owns and the AI that works for or works for you as an individual.
21:55
SPEAKER_02: You know, when you think about this, and we're starting to hit on this, what do you think about that?
22:00
SPEAKER_02: Distinction and what does it mean to help people carry their knowledge, their work style, their identity from one role or organization to the next, right?
22:09
SPEAKER_02: So, so that is that question of separating the organization's IP and the individual's IP and work style.
22:17
SPEAKER_02: And a lot of times the organization owns so much of it, but it doesn't own your style, your tone, and your voice, right?
22:24
SPEAKER_02: So what do you think about those that distinction and how that can be constructed or deconstructed?
22:31
SPEAKER_01: Yeah, when when you think about raw content, the enterprise owns that.
22:33
SPEAKER_01: So it's going in and saying, like, hey, during the work hours, you sent emails, you created files, you sent over proposals, all of that.
22:40
SPEAKER_01: I don't think anybody would argue like that is work product for the company.
22:44
SPEAKER_01: So from that aspect, that IP is available.
22:47
SPEAKER_01: You could build a digital twin against that IP, and that's highly valuable.
22:50
SPEAKER_01: When you leave the organization, what that does for the new person coming in.
22:54
SPEAKER_01: So the new person coming in used to have like a three to six month ramp up time.
22:58
SPEAKER_01: They need to understand where Seth or David kept these files, what they were interested in, what was really important, who are the right people to talk to.
23:04
SPEAKER_01: That onboarding is really expensive for a company.
23:06
SPEAKER_01: Well, now if I can actually take this new person, combine the IP that already existed for this role, I could accelerate that ramp up time.
23:14
SPEAKER_01: And that's going in and just saying, like, let his the new person's agent talk with the new digital twin that we just created and combine those two things together.
23:22
SPEAKER_01: Now it's actually going to be like, hey, I can answer that for you.
23:24
SPEAKER_01: You don't need to go find the person.
23:26
SPEAKER_01: I can actually just email that person for you.
23:28
SPEAKER_01: I can find this file, this contract, and it could be kind of that assistant on your side.
23:33
SPEAKER_01: So I think that it's beneficial for the company in that scenario.
23:36
SPEAKER_01: It's beneficial for the new hire.
23:37
SPEAKER_01: But now, Seth, let's imagine you go to a different organization and now it knows your voice, how you'd like to respond.
23:42
SPEAKER_01: Hey, I don't respond to emails after 6 p.m.
23:44
SPEAKER_01: I like to keep my work and personal life separate.
23:47
SPEAKER_01: But at 8 a.m., I want to send an email right away because I want them to know I'm on it, I'm ready for you, I want to answer your question.
23:52
SPEAKER_01: So during that time, your digital twin can understand your preferences.
23:56
SPEAKER_01: Now you go into that same organization, they have people that are super important where it's like, this is our number one client.
24:00
SPEAKER_01: You have to respond to them.
24:02
SPEAKER_01: But everybody else, it's okay, stick with your rules.
24:04
SPEAKER_01: You might not know that, but now if you have those two agents talk with each other, it's gonna build that kind of process, those rules out.
24:10
SPEAKER_01: And I think that's something where we don't see that today because no one's taking that portability yet for that data set.
24:16
SPEAKER_01: But I think over time you're gonna see that more and more, and it's gonna streamline where you're not gonna have to learn that tribal knowledge, you're not gonna have to learn like the rules of the road when you go to a different organization.
24:25
SPEAKER_01: It's gonna be laid out for you.
24:27
SPEAKER_02: Yeah.
24:27
SPEAKER_02: And and what has to happen for that to take place?
24:31
SPEAKER_02: Because again, using a tool like your own tool and analogous tools, like when I when I uh give Claude uh problems to solve, it will I give it access to HubSpot, I give it access to my email, to Office 365, I give it access to Slack, I give it access to lots of stuff.
24:48
SPEAKER_02: And and so it's giving me uh interesting insights based on that.
24:52
SPEAKER_02: What has to happen for that to be formalized and structured in a meaningful way that still respects privacy, right?
25:00
SPEAKER_02: Information privacy, because again, when you're when you're uh doing a project, you can share the project, but the the chats aren't automatically shared because they're you have to share them explicitly, right?
25:12
SPEAKER_02: So what has to happen in in the environment, whatever, regardless of what tool um works, does this have to be formalized as a policy with releases, with you know, IP assignment?
25:24
SPEAKER_02: What has to actually happen for that to be uh realized?
25:28
SPEAKER_01: Yeah, I think with the IP assignment, those are already happening today.
25:31
SPEAKER_01: Like you're generating content for work.
25:33
SPEAKER_01: Like if you go into a white-collar job, if you can go into a knowledge worker job, that is already in place for most employees.
25:39
SPEAKER_01: I think where you see the benefit um when it comes to how does it get more scale?
25:44
SPEAKER_01: How do you handle permissions?
25:44
SPEAKER_01: How do you handle privacy?
25:46
SPEAKER_01: One way to do that, and we've already done this with kind of our digital twin, is you take the native permissions that are in place.
25:51
SPEAKER_01: So if you think about the native permissions, when you log into HubSpot, you've got permissions that are in place that say you have access to this level of data.
25:58
SPEAKER_01: So your digital twin ultimately has the same level of access, it doesn't get more.
26:02
SPEAKER_01: In a lot of cases, it gets less.
26:04
SPEAKER_01: Uh it goes in and says, when you email someone, who actually has the right to that email?
26:08
SPEAKER_01: So you email your wife, but that your wife never is in another string, or your husband.
26:13
SPEAKER_01: Well, that's fine, but that means that that conversation should not go outside of these two people.
26:17
SPEAKER_01: And so now you've got it locked down where it's like it only goes to these two people versus like, hey, this email went out to 42 people.
26:22
SPEAKER_01: All 42 people have access to this email.
26:24
SPEAKER_01: These same 42 people were in the meeting, they have access to all this information.
26:27
SPEAKER_01: Now you've got a system that already says, like, this content, this insight is available to all those folks.
26:32
SPEAKER_01: So when you think about that data set, I think if you take existing native rules, that's the cleanest way to do it.
26:38
SPEAKER_01: I think if you start to go and say, I'm gonna build rules for every case, I'm gonna build prompts for every case.
26:43
SPEAKER_01: What you run into is the thing that went early Chat GPT, where you can start asking it questions and try to find personal information where you can try to crack it, like you don't want to get to that point.
26:51
SPEAKER_01: So you want to use what's native versus trying to build this ultimate system.
26:55
SPEAKER_02: Right.
26:55
SPEAKER_02: Right.
26:56
SPEAKER_02: And then is there a mechanism to activate that?
27:00
SPEAKER_02: I'm just trying to think of the logistics of this.
27:02
SPEAKER_02: I leave my company, I've left behind, you know, this legacy of patterns of communication where I've shared stuff with uh wider audiences.
27:13
SPEAKER_02: Does that become I'm just trying to think of what the boundaries of that disclosure and security are?
27:20
SPEAKER_02: Does it automatically, because it reaches that threshold of X number of people, then become public IP, or is it still have to respect the boundaries of just the people in those meetings, right?
27:32
SPEAKER_02: In other words, when does when does it extend beyond that context?
27:36
SPEAKER_02: And then how does that get operationalized?
27:38
SPEAKER_02: Because I love the idea.
27:40
SPEAKER_02: I mean, I love what you're saying.
27:41
SPEAKER_02: I'm just struggling a little bit to understand what the steps are to put it in place and operationalize it.
27:48
SPEAKER_01: Yeah, from an operation standpoint, um, we've got a concept of like we call it the OG, which is the original, the generator of the content.
27:54
SPEAKER_01: You've got the digital twin, um, and then you've got the admins, the IT folks that are in the mix.
27:58
SPEAKER_01: And what we've seen when digital twin gets implemented when it's kind of covering when someone's out of office, where it does a response.
28:04
SPEAKER_01: You always keep the OG on the screen.
28:06
SPEAKER_01: So when a digital twin goes in and responds to an email, the actual OG still the original gets the uh actual content when it's a sensitive content conversation.
28:15
SPEAKER_01: So you could go in and say, like, whoa, whoa, whoa, that was AI, that was wrong.
28:18
SPEAKER_01: Like in the edge cases, that does happen, or hey, I didn't want to give that out.
28:21
SPEAKER_01: Can you please keep that off of the record?
28:23
SPEAKER_01: You could do that, it's gonna get better over time.
28:25
SPEAKER_01: But then let's say someone leaves an organization and you keep that alias alive for that digital twin for that person, that's where you have it going and you CC the manager.
28:32
SPEAKER_01: So the new OG is the manager that goes in and says, like, hey, is it okay that the digital twin answers this question?
28:38
SPEAKER_01: So what you're using is you don't want to create a whole new workflow.
28:41
SPEAKER_01: So we really launched with email because email is the clearest workflow where it's like if I email this person, I will get a respond not from 50 other people, from this one person.
28:50
SPEAKER_01: And if that person then CCs a manager, that's fine, they're in the loop in that conversation.
28:55
SPEAKER_01: But I think if you try to build something new at this stage, it's going to be you're gonna get low adoption.
29:01
SPEAKER_01: You're gonna say this new thing that buzzes your ring and it does this, and you tap that no email's super clear.
29:06
SPEAKER_01: I think the next iteration is gonna be Slack and Teams, where it's like, hey, someone emails the digital twin on Slack, the digital twin responds back and includes the OG.
29:14
SPEAKER_01: So the OG can actually be a part of that conversation, and that actually stops people from asking nefarious questions like, hey, what's this person's social security number?
29:22
SPEAKER_01: What's their birthday?
29:23
SPEAKER_01: Where are they going for vacation?
29:25
SPEAKER_01: How much do they make?
29:26
SPEAKER_01: You don't, you will never do that if you know you're gonna get caught for the most part, because you get fired, you get sanctioned, whatever that is, because you know it's gonna go back to the OG.
29:33
SPEAKER_02: Right.
29:33
SPEAKER_02: So, so the so the fact that this the email chain and infrastructure is already well defined and thought out and has inherent permissions, inherent disclosure, inherent IP issues, that becomes the vehicle for this, essentially.
29:50
SPEAKER_01: And over time you'll get more, we'll get more complex, but it's like right now it's not ready to open up everyone's a single person's data on MCP.
29:57
SPEAKER_01: And you see that with OpenClaw today, where it's like it's a little bit too much.
30:01
SPEAKER_01: The potential is there, but it's just a little bit too much right now.
30:04
SPEAKER_02: Yeah, yeah, yeah, yeah.
30:05
SPEAKER_02: So let's talk a little bit about agentic guardrails.
30:08
SPEAKER_02: So obviously, agency's getting a lot of hype these days.
30:11
SPEAKER_02: You built guardrails into Reed AI's approach from the start.
30:15
SPEAKER_02: Now, what does responsible agentic AI look like in practice?
30:19
SPEAKER_02: And how do organizations think about what an agent can and can't do?
30:23
SPEAKER_02: And this is very analogous to what we just discussed, but just take it a step further and maybe talk about other types of agents.
30:29
SPEAKER_02: You know, OpenClaw is a great example, right?
30:32
SPEAKER_02: Because it it is so powerful, but you know, people are not necessarily giving it the guardrails.
30:38
SPEAKER_02: What are your thoughts about that whole uh realm?
30:42
SPEAKER_02: Because again, just like you have guardrails around employees, you need guardrails around agents, right?
30:47
SPEAKER_02: Because agents are acting on behalf of someone.
30:51
SPEAKER_02: An agent can't just be an agent in the abstraction, it's an agent that acts on behalf of someone, right?
30:57
SPEAKER_02: So tell me what your your paradigm is or what your thought process is about agentic guardrails and and restrictions, guidelines, and and and policies.
31:06
SPEAKER_02: What are your thoughts?
31:07
SPEAKER_01: Yeah, I think the most important is disclosure.
31:09
SPEAKER_01: And so when we join a meeting, just as one example, is like when we join a meeting, uh, we join as a participant on the call, we made that decision because we wanted people to know that there was an agent on the call recording the call, taking notes during the call.
31:22
SPEAKER_01: Now you can go in and say I can use a desktop app and I can go in and record a call and I don't have to tell anybody.
31:27
SPEAKER_01: Well, the problem there is now the agent has data that is not kind of properly sourced because we've got 12 two-party consent states in the US, in the EU, globally, like you, and in just general best practices, you shouldn't record people without their permission or notification.
31:42
SPEAKER_01: And so we took that approach and we said, like, hey, we're joining, we're we're joining a call, but then you also give the ability to opt out.
31:48
SPEAKER_01: So then you want to go in and say, like, hey, you and I were just talking earlier, we had read join the call, but we're doing this podcast, and it's like, hey, probably not.
31:55
SPEAKER_01: Let's take this out off the call.
31:57
SPEAKER_01: We just typed an opt-out, it left the meeting.
31:59
SPEAKER_01: Totally simple to do.
32:00
SPEAKER_01: So you want to one, give disclosure, two, let people opt out any time.
32:03
SPEAKER_01: And then three, by default, don't train against their data.
32:06
SPEAKER_01: So that's a hard thing, but you want to get the opt-in to get trained against that data.
32:10
SPEAKER_01: So when our users come in and create an account, we don't actually automatically opt them in, they have to opt in themselves.
32:16
SPEAKER_01: And we get like a seven to ten percent depending on the month opt-in rate for new users.
32:20
SPEAKER_01: Uh, for enterprises, they turn it off, which is fine, but we'll train against their own data.
32:24
SPEAKER_01: So if you take that same philosophy and you apply it to everything else when it comes to agent, hey, always disclose that it's being used.
32:30
SPEAKER_01: Make sure that you give the people the ability to opt out of the using that agent.
32:34
SPEAKER_01: And then those two things cover most of the bases where it's like, okay, I know that Seth is using an agent, it's following up with me, it's giving me some feedback on some stuff.
32:43
SPEAKER_01: Great, I know this, but I'm gonna check with Seth because this one went off a little bit off the rails.
32:46
SPEAKER_01: Maybe it didn't follow one of his rules.
32:47
SPEAKER_01: So I'm gonna check in with him to say, Hey, you sure you want me to write it in this format?
32:51
SPEAKER_01: And you're like, no, no, no, don't do it this way.
32:53
SPEAKER_01: So I think that disclosure formatting conventions.
32:57
SPEAKER_01: So I think it's important on the disclosure side.
32:59
SPEAKER_01: So you people can just kind of react to knowing that it is an AI agent on the back end.
33:03
SPEAKER_01: So I think that is critical.
33:04
SPEAKER_01: And then when it comes to like the training, it should be clear if you're getting that data is getting trained or not.
33:09
SPEAKER_01: So if you're an enterprise, most employees I think would assume that, hey, if I have data and I'm using AI, it's gonna get trained to help the company become better, to help my workflows become better.
33:18
SPEAKER_01: But you shouldn't just leave it to like, hey, you should assume you should go and just clearly state, like, hey, if you're gonna use this service, we're gonna train against this to make the models better, to make your time more efficient.
33:28
SPEAKER_01: And that's the benefit for you.
33:29
SPEAKER_01: And I will say, most employees, I would say nine out of 10 employees will take efficiency any single day over, like, hey, the old way to do it.
33:38
SPEAKER_01: And I think a great example is like Slack teams.
33:40
SPEAKER_01: Yeah, people at first were like, hey, I'm just gonna stick with the email, it's fine.
33:43
SPEAKER_01: I can email people now.
33:44
SPEAKER_01: It's like Slack is like you take it away from me, I might switch my company.
33:47
SPEAKER_01: So I think that level of kind of efficiency people want, and they're willing to take it like, hey, you can see if I'm online or not, I'm always available.
33:55
SPEAKER_01: Like those things are kind of like risks that you have, but people would prefer to take the risk for the benefit they have.
34:00
SPEAKER_02: Yeah, and for for a Luddite like me, I I'm resistant to change.
34:05
SPEAKER_02: I want uh I want things to be better, but I hate changing how I do work my work, right?
34:10
SPEAKER_02: But then I get sucked into anything that uh is more efficient.
34:14
SPEAKER_02: And yes, I uh uh I'm joking that I'm a Luddite, but I I think I I appreciate technology, but as long as I don't have to do things differently.
34:23
SPEAKER_01: No, no, I I I think that that's one of the reasons why we did the email where we could have done something more complicated.
34:28
SPEAKER_01: Yeah, but everyone knows email.
34:30
SPEAKER_01: You can work within your existing workflows, that's where you get the adoption.
34:33
SPEAKER_01: I think that's where ChatGTPT and Claude did it right, where it's like, hey, we're gonna have a chat box.
34:38
SPEAKER_01: Like we're gonna show more value, but people understand what a chat box is.
34:42
SPEAKER_01: Like you ask it questions, you get information back, we're just gonna make it a lot better.
34:46
SPEAKER_02: Yeah, yeah, absolutely.
34:48
SPEAKER_02: So let's talk a little bit about AI readiness for organizations.
34:52
SPEAKER_02: A lot of organizations are struggling with where to get real ROI from AI.
34:57
SPEAKER_02: Pilots are, you know, fail, adoption stalls, shadow IT, shadow AI pro proliferates.
35:03
SPEAKER_02: What do you think organizations need to have in place to actually take advantage of these technologies?
35:09
SPEAKER_02: And what does readiness look like?
35:11
SPEAKER_02: I know I sent you a little information about the readiness stuff we did.
35:15
SPEAKER_02: I don't know if you had a chance to look at that, but what are your thoughts about readiness and where are people kind of missing things?
35:22
SPEAKER_02: And what do you see?
35:23
SPEAKER_02: Uh for your your technology, you can it is almost plug-and-play.
35:27
SPEAKER_02: You don't really need a lot of preparation, but where do you see uh readiness failing in other contexts?
35:34
SPEAKER_01: I think you see readiness fail when it's two things when it's top down or when it's IT and admin, and you have to go through a very long process.
35:42
SPEAKER_01: So, top down um really highlights like if you saw that study, there's a lot of flaws with that study, but 80, 90 percent of AI pilots fail.
35:49
SPEAKER_01: It's because a lot of time it's like, hey, at the board level, someone said you have to use uh AI, and then the execs go in and say, hey, let's use this AI vendor, you roll it out, and people are like, what do I do with this?
35:57
SPEAKER_01: How do I utilize this?
35:58
SPEAKER_01: It's very it's very low value, and it's not coming from the bottoms up to say you're solving my problem.
36:03
SPEAKER_01: Someone at the top level does not deal with the same problems on the ground doing it's a hammer in search of a nail, right?
36:08
SPEAKER_01: Exactly.
36:09
SPEAKER_01: And then with the IT admins there and security personnel, they they are doing the right thing where they're trained to stop things that you know have access to data that evade that have access to private data, that you don't want data to leak, etc.
36:21
SPEAKER_01: But I think right now that's one of the biggest blockers that they don't move fast enough.
36:25
SPEAKER_01: So not saying accept everything, because what's going to happen is people want to learn, they want to experiment.
36:31
SPEAKER_01: That is one of the things, like if you look at all these token maxing companies out there right now, where they're going and say, we're giving our employees as many tokens as they want, and we are actually having a leaderboard so they spend more money on tokens.
36:41
SPEAKER_01: Like that's the level of stage that we're at.
36:44
SPEAKER_01: So if you're an IT admin right now, blocking everything, you're kind of taking this point where everybody wants to learn.
36:49
SPEAKER_01: They're actually going in and saying, How do I make my work better?
36:51
SPEAKER_01: How do I make this company better with AI?
36:54
SPEAKER_01: You blocking it is not solving the problem.
36:56
SPEAKER_01: Now, you pausing it and very quickly doing a review to say what is at risk and what are the right guardrails and is this the right vendor?
37:03
SPEAKER_01: That's the right approach for readiness.
37:04
SPEAKER_01: It's going in and saying, like, let's add one or two more people on security or IT so that we can review these things faster, so that individuals can choose what they want to bring into the workplace.
37:14
SPEAKER_02: And at least having the structures, the decision-making bodies and decision-making processes for allocating those resources, for making those determinations, for prioritizing what gets funded, where the investigations uh go, and and really harvesting those best practices from across the organization.
37:33
SPEAKER_02: So you do need to have some structure.
37:35
SPEAKER_02: Uh, we find governance is a big, a big piece of that.
37:38
SPEAKER_02: That nobody wants to talk about governance?
37:40
SPEAKER_02: When when do I have available for uh availability for governance?
37:43
SPEAKER_02: How about never?
37:44
SPEAKER_02: Never works for me, does never work for you?
37:46
SPEAKER_02: You know, but if you talk about it as how can we accelerate what we're doing, how can we make better decisions?
37:52
SPEAKER_02: How can we ensure ROI?
37:54
SPEAKER_02: How can we speed up this process of adoption?
37:57
SPEAKER_02: That's a different story, that's a different framework for making those decisions.
38:01
SPEAKER_01: 100%.
38:02
SPEAKER_01: And like a good example is one of our clients, they're a multinational company.
38:06
SPEAKER_01: They have a product manager in Tokyo, they have a product manager in Paris, they have a product manager in LA.
38:10
SPEAKER_01: They all speak different languages.
38:12
SPEAKER_01: All the content that they generate is not accessible to them.
38:15
SPEAKER_01: They started to individually use our product in different markets and they started to see other people on their teams using it.
38:20
SPEAKER_01: The big company actually, this is a fan company, they came in and said, like, hey, we trust our employees to pick what makes the most sense for them.
38:26
SPEAKER_01: We're gonna do a review.
38:27
SPEAKER_01: You want to lean in, we can get this done in two weeks, but you gotta lean into this thing really hard.
38:31
SPEAKER_01: We leaned into it, and now they're getting insights from Tokyo in LA that they never would have seen because the beauty of it is like your meetings, your customer interviews are all recorded, the meetings that you have internally to say what's next steps.
38:44
SPEAKER_01: Now I can just bring that into the system.
38:45
SPEAKER_01: And when I ask a question, I can not only just ask my agent, but I can ask the entire product org's agents a question.
38:51
SPEAKER_01: And they don't care about language, they don't care that you're Japanese or French or US.
38:56
SPEAKER_01: It's just gonna go and say, like, hey, here are the top five problems globally, and this is the concentration, and this is what you should work on.
39:01
SPEAKER_01: And if you want feedback from clients, let me play these back for you.
39:04
SPEAKER_01: That's like would have never happened had we done a top-down approach or had we waited for IT to decide what was going to go in versus like bottoms up, then they check with IT and then they get it done.
39:14
SPEAKER_02: Yeah, that's great.
39:16
SPEAKER_02: That's a a very uh powerful approach because you're empowering people on the ground who have the actual problems, who are facing those issues, who are dealing with those challenges, and then you're facilitating that collaboration and that communication amongst them.
39:30
SPEAKER_02: And uh integration of subject matter experts into the workflow is really a what we consider a critical readiness factor.
39:38
SPEAKER_02: It's process readiness, right?
39:39
SPEAKER_02: So how are we leveraging, how are we getting their input, how are we uh uh capitalizing it, how do we leverage it, how do we apply it?
39:47
SPEAKER_02: How does the AI amplify it?
39:49
SPEAKER_02: And that's what we're talking about is amplification and and uh surfacing of that.
39:54
SPEAKER_02: But I want to talk a little bit about the workforce and kind of the long game.
39:58
SPEAKER_02: You've seen firsthand how AI removes friction from processes, and that inevitably changes what people do.
40:04
SPEAKER_02: How do you think about the longer-term workforce implications?
40:07
SPEAKER_02: You know, where does this go over the next five to ten years?
40:10
SPEAKER_02: And people kind of tongue in cheek say, Well, we're gonna have universal basic income.
40:15
SPEAKER_02: Ha, right?
40:15
SPEAKER_02: We we're not even paying, you know, a living wage to people, right?
40:19
SPEAKER_02: Left to our, you know, left to the political devices and the uh capitalist drives.
40:25
SPEAKER_02: You know, we pay people less and less as little as we can, and and it's not a living wage.
40:30
SPEAKER_02: And then, you know, you end up with people working for financial services organizations who actually need public assistance when they're you know, when their their uh um leadership are getting significant bonuses, and then that is a form of subsidy, but it's not a form of subsidy that is very visible or palatable.
40:49
SPEAKER_02: And and again, as you know, when you look at the long game and how this is playing out, it's going to significantly change uh the workforce makeup, it's gonna significantly change what people do with their jobs and their time and change entry-level positions.
41:03
SPEAKER_02: I mean, there's a lot of questions around that.
41:05
SPEAKER_02: What is your take on kind of the future of the workforce and and how this is playing out over the next five and 10 years?
41:11
SPEAKER_01: I think it's gonna be a boom in terms of productivity, and I think it's gonna be a boom when it comes to developing your emerging markets.
41:18
SPEAKER_01: And so taking that example of like, hey, income disparity across different countries, emerging versus developed country markets.
41:25
SPEAKER_01: Uh, a great example is we have a large set of customers in the Philippines and Malaysia.
41:30
SPEAKER_01: And at first we were like, why are people signing up for our service in Philippines and Malaysia?
41:33
SPEAKER_01: Uh, this is early days where we didn't necessarily have it in the native language of the website, even it was just an English website.
41:39
SPEAKER_01: Maybe you can use your browser to change the language.
41:41
SPEAKER_01: Like, this is this is interesting.
41:42
SPEAKER_01: What we found out was these were executive assistants that were virtual executive assistants because in Philippines and in Malaysia, they have a large English-speaking contingent that speak very perfect English.
41:52
SPEAKER_01: And so they get recruited to go in and be virtual assistants for uh these developed markets, especially specifically in the US.
41:58
SPEAKER_01: They started to use our products so they could be in multiple meetings at multiple times.
42:01
SPEAKER_01: They went in and said, like, hey, I used to go to meetings and I would have to be the person that took the notes.
42:06
SPEAKER_01: Now I've got AI taking the notes, giving me the summary, letting me understand what the follow-up and action items are.
42:11
SPEAKER_01: Now I can bring on more clients, and now I can make more money and I can give jobs to other people that do not have that ability uh today because there wasn't enough work.
42:20
SPEAKER_01: So, in that specific example, I think humans or humanity will find a way to go in and say, like, hey, this is going to lift us up.
42:27
SPEAKER_01: This is a competitive advantage.
42:28
SPEAKER_01: How do I now leverage this?
42:30
SPEAKER_01: Uh in Brazil, we're the number one meeting note taker, where we don't have a team in Brazil, we don't have a single person that speaks Brazilian Portuguese uh within our organization, but we've been able to scale out because we deliver product in that market for that use case.
42:43
SPEAKER_01: In Colombia, we have one to two percent of the population using our product in any given month.
42:48
SPEAKER_01: Like we have certain universities where almost every student records every single call or every single session or a class they take it and they created this kind of database of this folders where they're sharing this information with each other, saying Mr.
43:01
SPEAKER_01: Class.
43:02
SPEAKER_01: Hey, here you go.
43:02
SPEAKER_01: Here are the things that I thought was interesting, here's where people were paying attention, and it's all automated.
43:07
SPEAKER_01: And I think that's where you're seeing is it's it's almost the it's the great leveling of the entire economic structure where it's like people who have access to AI are going to be just as competitive and even where whatever market you're in.
43:20
SPEAKER_01: And I think that's getting that gets me a little that gets me excited because that goes in and says, hey, everybody's got a shot now.
43:26
SPEAKER_01: If you have a computer, if you have access to AI on a phone or something else, you have a shot to be just as strong, just as affluent, just as educated as anybody in the developed markets.
43:38
SPEAKER_01: And I think that is the great level level playing.
43:40
SPEAKER_02: Interesting.
43:41
SPEAKER_02: It's a hopeful way to look at this, right?
43:44
SPEAKER_02: Um, and you know, it it's interesting in terms of how we're seeing things really play out, but uh there will be a disruption, you know, when you look at the number of jobs that are about people driving, you know, cars and trucks, right?
43:59
SPEAKER_02: And yeah that Increasingly automated, they're going to be a big displacement.
44:03
SPEAKER_02: I mean, that's there's no question about it.
44:05
SPEAKER_02: And my my fear is we're going to have a lot of those folks relegated to the lowest common denominator of the things that the machines cannot do, picking and packing and shopping and task rabbiting and you know things that are that are less feasible, although there's been some interesting workarounds with machines hiring task rabbits to do things for them.
44:28
SPEAKER_02: It's very, very interesting.
44:29
SPEAKER_02: Yeah, still still TBD, but it's a it's a hopeful perspective that you have.
44:34
SPEAKER_02: So looking ahead, where do you see Read AI going?
44:38
SPEAKER_02: And more broadly, what should business leaders be paying attention to as agents become more capable and more embedded in how work actually gets done?
44:48
SPEAKER_02: What's your what's your vision of how the future is playing out?
44:52
SPEAKER_01: Yeah, I think today read AI is the system of record for meetings for five million plus people and we're growing, adding 40,000 new users every single day.
44:59
SPEAKER_01: So we are that system of record.
45:01
SPEAKER_01: Uh people trust us for our notes.
45:03
SPEAKER_01: Uh, they're building on top of our systems to get access to the notes to put into things like HubSpot, Salesforce, uh putting in a Jira, Confluence, uh, creating, we just had some writer at XDA write a story about how they connected Reed with Claude, and now it's actually helping them follow up on all the many different tasks that they have.
45:20
SPEAKER_01: And when news changes, what's interesting and what's important to them.
45:23
SPEAKER_01: So that's been really kind of uh eye-opening to be that system of record because three, four years ago, meetings were like Snapchat messages.
45:30
SPEAKER_01: They were ephemeral.
45:31
SPEAKER_01: It was whatever you remembered, and that was it that didn't exist.
45:34
SPEAKER_01: Maybe you wrote a couple notes, maybe if you're a reporter, you recorded it on a little kind of handset.
45:39
SPEAKER_01: But that was about it at the end of the day.
45:41
SPEAKER_01: Where we're going towards and we're already started to is a system of action.
45:44
SPEAKER_01: So, how does AI take actions for you, or how does it recommend specific actions?
45:49
SPEAKER_01: So that's where our digital twin, ADA, it does scheduling.
45:52
SPEAKER_01: It's doing really basic scheduling, it does it, gets it done, checks three, four hours back on a weekly basis to you.
45:57
SPEAKER_01: Now it's going in saying, hey, you're asleep.
45:59
SPEAKER_01: They've got a question that you want me to answer.
46:01
SPEAKER_01: I got this.
46:02
SPEAKER_01: This is a common question that comes up.
46:04
SPEAKER_01: This is your most typical response.
46:05
SPEAKER_01: I'm going to send this out and I'll CCU on it.
46:07
SPEAKER_01: It's going to get into keeping it up.
46:09
SPEAKER_01: No need to wait.
46:10
SPEAKER_01: Exactly, exactly.
46:10
SPEAKER_01: And that's a benefit for all of us.
46:12
SPEAKER_01: It's like, hey, I don't have to be paged at two o'clock in the morning that something is happening and I need to solve it.
46:16
SPEAKER_01: Then it's going to, we're already starting to do this.
46:18
SPEAKER_01: It's going to recommend that you take specific action.
46:21
SPEAKER_01: So after the meeting, do you want me to schedule a meet, a follow-up?
46:24
SPEAKER_01: After the meeting, do you want me to do the research on this question and ask these four people in the org what's the current status on this project?
46:30
SPEAKER_01: Hey, after the meeting, do you want me to do something else?
46:33
SPEAKER_01: But even before, even after that, it's going in and saying, like, hey, based on all the context that you've connected, and we're starting to do this where with emails, with messages, with meetings, not just for you, but people that have shared information with you, that have shared their folders of content with you, can you actually move deals in HubSpot forward?
46:50
SPEAKER_01: So imagine you never have to update HubSpot again because it goes from 25 to 50%, 50 to 75, because you've got different people on the team that are kind of including Reed and it's making those decisions, it's moving those things up and it's saying, like, hey, Seth, need you to make this call.
47:04
SPEAKER_01: If you make this call right now, 95% chance you close this deal the next week for$100,000.
47:10
SPEAKER_01: Like, okay, this is great.
47:11
SPEAKER_01: And it's I've got the full context, I'm gonna give that person a call right now.
47:14
SPEAKER_01: And I think that's where Reed is aiming to go to.
47:17
SPEAKER_01: And we're already seeing that with the adoption that we're seeing with ADA.
47:20
SPEAKER_01: And I think the cleanest stat that we have with ADA is even we declared at that that it's an AI agent, 25% of people still say thank you every time it completes a task.
47:29
SPEAKER_01: And so it's going in and saying, like, from a human nature perspective, you actually find value in what the AI is giving you that you're actually saying thanks.
47:36
SPEAKER_02: Yeah, that's so fascinating.
47:38
SPEAKER_02: I I tend to try to I tend to do that as well, even though I know it's using more tokens.
47:42
SPEAKER_02: Well, listen, David, thank you so much uh for joining and sharing your perspective on digital twins, agentic AI, what it really takes for organizations to get value from these technologies.
47:53
SPEAKER_02: Uh, and to our listeners, thank you for tuning into the early AI podcast.
47:57
SPEAKER_02: Be sure to subscribe uh for more conversations on how AI is shaping the future of business.
48:03
SPEAKER_02: And people can find you at uh what's your we'll put your LinkedIn information in the show notes, and they can find you at uh read.ai.
48:12
SPEAKER_02: Yep, read.ai.
48:14
SPEAKER_02: Read.ai.
48:15
SPEAKER_02: Okay.
48:15
SPEAKER_02: Well, again, David, thank you so much for your time and really appreciate having you on.
48:20
SPEAKER_01: This has been great conversation.
48:21
SPEAKER_01: Thanks, uh, appreciate it.
48:22
SPEAKER_02: Excellent.
48:22
SPEAKER_02: All right, we will talk to you soon.
48:24
SPEAKER_02: Thank you very much.
48:25
SPEAKER_02: Bye now.
48:27
SPEAKER_00: Thank you for joining us on another deep dive into AI innovation.
48:31
SPEAKER_00: Tune in next time when we introduce another industry expert and discuss how to maximize AI in your world.
48:37
SPEAKER_00: The Early AI Podcast is sponsored by Early Information Science and Vector.
48:42
SPEAKER_00: To learn more, visit early.com.
48:45
SPEAKER_00: That's E A R L E Y dot com.
48:48
SPEAKER_00: Thank you for tuning in.