Introduction: What’s Possible With AI Today

We're going to take you through what is possible right now today.

Capturing Real-World Inputs From the Room

And what I just went around, as some of you saw, as I said, please can I take your picture? Please can I take a picture? And I took a picture of about 10 to 15 of the handouts.

And now we're using AI.

From Photos to Text: Transcription With Gemini

In this case, I'm using Gemini. So what I did up here is I said, please can you transcribe all of these photos?

And what's cool is that it can take everything written, even poor handwriting, and immediately transcribe it into what people wrote but that's not the most interesting part of

what we're gonna do here the most interesting part of what we're gonna do

here is we're gonna start playing with it and so I'm gonna start working with Gemini like a teammate rather than a tool to see if I can surface some interesting insights from what the group said so let me get started here I'm

using something called monologue which doesn't seem to be working so you know know, live demos, that's super cool. Let's see if I can use Nope. Okay.

Oh, yeah, it's working. All right. So here we go.

Prompting for Insight, Not Just Transcription

I'm going to tell Gemini this, I want you to be Sammy the synthesizer. And you are incredible at reading messy inputs, and spotting what matters.

So I just did an exercise with a group of about 100 executives, where I had them fill out both, you know, answering the question of today I use AI for this and tomorrow I use AI for this is what I'd like to be using AI for.

So you've already transcribed it. I want you to analyze it. Give me three bullets of where this room is at right now.

That's the first bullet I want. I want you to high level it into where this room wants to be. And then I want you to synthesize those two things and say, what is the most important thing that their answer reveals that's standing between these two?

And I don't want to see any of the transcription. So this should be turning. Oh, it's just saying, thank you. Great. All right.

Well, I've got a secondary thing I can do here, which is I have the, uh, Jeremy cover for me.

The, uh, the danger of a live demo is sometimes your transcription software doesn't work.

A Core Mindset Shift: Treat AI Like a Teammate

This is quite normal, but there's nothing magical about words. That's that should be like a bumper sticker quote for the rest of the day. There's nothing magical about words.

And what what you probably noticed in how Devin was just speaking to AI was just, he was using words to tell a teammate what to do.

1And what we've observed is the folks who outperform with AI don't treat AI like a tool. They treat AI like a teammate.

And if it sounded, I bet most of you didn't even realize that Devin was attempting to interact with AI, right? Because the way he was speaking was so casual and informal and not robotic, right?

But that's the way Devin talks to AI. I've seen Devin talk to AI. Devin treats AI like a teammate.

And what you'll see, and hopefully we'll get the voice transcription thing working. Yeah. But what you'll see is

when you see folks demoing today, you're going to see folks treating all of these different technologies like a highly capable teammate that will thrive under good leadership, that will thrive under good management.

And we can talk about what some of those practices are later. That's right.

Early Findings: Current Use Cases vs. Opportunity Areas

Okay, so this just went through and in like a blink of an eye, basically, I had a backup version of this prompt that in the blink of an eye, it went through here and it identified what are the most common current use cases for today, such as content generation, refinement, summarization of synthesis of unstructured data, and then the biggest opportunity areas that people are looking at. agentic and automated workflows complex financial operational reconciliation I'm

not we actually don't have time to go through all of this right now but it's really interesting how it also calls out some surprising or non -obvious patterns

here such as like one executive is using AI as a board of directors and instead of just generating a presentation they're similar simulating the adversarial environment I've seen this work super well in many different

Building a Source-Grounded Workflow With NotebookLM

instances but here's the thing I don't have time to go through all this right now so what I'm actually gonna do is I'm gonna take this entire answer and I'm

gonna put it into something called notebook LM and who here has used notebook LM before all right so you all have seen how if I create and for those

Why NotebookLM: Constraining the Model to Your Sources

of you who don't know what notebook LM is basically what you can do with notebook LM is a normal large language model is pulling from everything that's that's possible on the internet, 1but Notebook LM is completely just on the source.

So I personally have copied text here. So with this source now, I can query Notebook LM for just what this room just said.

So this is the analysis of what this room just said right here.

And what I'm gonna do is I'm gonna click audio overview, and what this is gonna do is it's gonna generate generate a deep dive of basically the trends, what people are doing today, what they want to be doing tomorrow, and then we'll be able to listen to it as it's generating.

You can see it's generating over there.

But then the next thing that I want to ask is basically to say, okay, everybody here has identified what it is they're trying to do and what they'd like to be doing more of.

Turning Insights Into a Research Brief

so what could be what could be a great deep research report that we could run that would inform and surprise the attendees here so that they could leave with a newfound sense of direction and what to do give me concrete and specific

examples now one thing that you can see that I'm doing here is just moving fast and misspelling things and maybe you all do this but normally my normal mode is to just be talking out loud with it as Jeremy said I talked to it like a teammate and so what's happening right here is I'm saying hey what could be us

we're gonna run a deep research report and how many of you have run a deep research report before okay a few hands coming up and basically what this does is it's meant to put the McKinsey million -dollar report out of business. No offense to anybody who's from McKinsey.

And so what I've done is I've said okay great here you know and so it's giving me based on the analysis of this executive summit data here's the most striking and impactful deep research report that we can run. So I will take this here and then I'm gonna

to go, um, you can run a deep research report in many different places. In this case, I'm going to run it, uh, in Gemini. Um, I prefer the ones in Gemini, but you can do it in chat GPT.

You can do it in, um, Claude. And so I'm just going to paste this in and I'm going to say, yeah, let's go pro.

Importantly, Devin got the AI to draft the research brief, right? He didn't actually bother with what's the exact brief. He treated AI like a highly capable brief writer that he's now handing to another AI.

He's basically right now, no offense, Devin, he's the copy paste monkey between different AIs, basically, right? But he's taking one AI output and then he's feeding it into a different model. And this is an important role to acknowledge and appreciate in this moment.

That's right. I'm the copy paste monkey right here.

And so some of you have experienced this before but I just want to kind of keep backtracking and saying what it is we're doing right now I went everybody in this room came up with what is it that we're

doing right now with AI what do we wish we'd be doing tomorrow and then I took pictures of it I used AI to scan it all and then I had it transcribe it I had it run an analysis on what are the trends what are the patterns what are the blind

spots and then I brought that over into notebook LM and notebook LM is a a specific platform that is only going to be trained on the data that I give it, nothing else. And then I had Notebook LM give me a deep research query that I brought into Gemini. And so now

here's the research plan, which is it saying, okay, you know, and then so what Gemini will do is it'll say, hey, here's what I think I'm going to do. I'm going to send a team of agents out

To go research organizational redesign frameworks AI governance and all the rest and I'm just gonna say start research And that's gonna go off and cook for ten minutes or so While we go do some other fun things

You can see it's already starting to send a team of agents out to go run that but while that's running I would actually like to go back to where I was in notebook LM and Let's see here.

Do you want me to just take a beat on this while you're doing that?

Deep Research vs. Chat: How Agentic Research Works

So one thing about deep research for the roughly 60 % of people whose hands weren't up, deep research is not querying a language model. The way language models work, not to get technical, but they're predicting the next word, right? And so when you ask you IGPT or Claude or Gemini or Grok or whatever, a question or give it a query, it's basically predicting the next word and then the next word. That's not just clever UI, how text kind of unfolds. That's how the language model works.

When you use a deep research tool, whether it's Gemini's, which you're seeing Devin use, or Chad GPT's, or Perplexity's, or whatever, that's not a large language model in the sense that it's predicting the next word. What it's doing, even though it's in the same interface, it's kind of confusing, but it's ingesting your research query, and then it's synthesizing based on what the company, whether it's Google or OpenAI or Anthropic, what the company said, this is the way we commission research agents. We do, for example, we send agents to do three Google searches and see what comes back. And then we read the top 100 links, and then we refine our research query. And then we go and conduct another 100 links, and we read them and we refine, right? Anthropic or OpenAI or Google or whatever it is has taught a team of agents how they undertake research. search.

And what you see Devon's doing right now with deep research is those agents are following the process that Google or OpenAI or Anthropic have given them to go not predict the next word, but to go undertake meaningful web search and then synthesis and then redirection of its plan in order to ultimately serve up to the user, Devon, a citable resource that is grounded in actual factual, factually verifiable information.

So even though you're interacting through the same interface, it's important to know that what's happening with these deep research queries is really meaningful research that's informed by a lot of, call it, pre -training. It's not just asking ChatGPT a question and you're getting a hallucinated response.

That's exactly right. Thank you, Jeremy.

From Analysis to Executive Artifacts

Okay, so we are just going flying right along here on on this demo, and what I want to show you now, I'm back in Notebook LM.

And so remember, Notebook LM is only trained on what people in this room said, that's it.

So I asked Notebook LM earlier to create my deep research report prompt, and that's running, and now I'm back here and I changed it up, and I said,

Drafting an Executive Slide Outline

could you be Priya the presenter and create an executive eight slide presentation that I can bring to people in my company that shows them exactly where this room is at?

and again you can see I'm just misspelling things because with AI you move fast and misspell things it knows what you're doing and so Priya came back

Generating the Deck in Gamma (On-Brand, Fast)

here and now I'm gonna show you all gamma raise your hand if you've used

gamma here before okay so some hands are going up that's 20 % for those in the front roughly 20 % of the room okay so gamma is a really really cool tool that has saved me so much time and so basically what I can do with gamma is I that can create a presentation like instantly.

And we're gonna watch this get created here. So it asked me some questions. In this case, I'm gonna say paste in text. And then I'm gonna paste in the slideshow

that I just had Notebook LM create for me. And this is based on everything the room said and the analysis of it.

And then I can say, preserve this exact text, or I can have it generate notes from an outline. And then from here, what's gonna happen is I'm gonna actually choose my own template.

So this is the MindStone company template here. And I can say things like preserve, let me see if I can make this slightly bigger. I can say things like preserve.

To preserve all the text, I could have the AI condense it down. I'm gonna keep it at preserve. And then let's do some illustrations in here.

And then the coolest thing is just watching it work. So you click generate, and uh it says i'm back online i hope i am okay so then this is the slide deck outline in gamma

and this is just going and oh no the co -pilot era did you see that yes yep yep love it love it um and so this is based completely on what everybody in this room just said i used image to transcribe it. I had it run an analysis on it.

And then I, based purely on what this room said, nothing else, I created a presentation, which is in the MindStone look and feel format here. And so we can just take a quick look at it.

Bridging the AI gap from co -pilots to autopilots. Love it. Not bad. Not bad, not bad.

And a lot of the room is in the co -pilot era, which I'm super sorry for you on that. No offense to our sponsor. So now it's like, this is pretty good.

And if I wanted to, I could actually engage with the agent here and I could say, hey, can you add more words to each of these areas and beef them up?

And so now what's going to happen is the agent is now it's asking, let's just say this current current card. So the agent is saying, yeah, I can add some more words, because that's what we always want to do in slide decks, right?

And it asked me, like, which card do you want? And I'm just saying the current card. And now we'll be able to watch in real time as it's going through and it's adding some more words here.

Now, I, of course, wouldn't normally do this, but it's just cool to see, like I say, actually pare it back down again. in.

And importantly, Devin is treating Gamma here like a teammate. He's not, if he were making this presentation, he'd be looking at it going, how do I reduce the number of words, right?

But if he has a highly capable team member who's responsible for the slide deck, he looks at it and goes, we need less words on that slide. And that's all he has to tell the teammate.

And the teammate is is fully capable of executing his vision as a manager of the AI. Right.

And so I spend all of my day talking to my AIs like this. I use voice -to -text constantly, and I'm just saying, actually, no, okay, so this part of this slide I think is good. This part I want to change.

Oh, this makes me think of this. Move this all around. And then the AI teammates are just constantly churning through it and moving it around.

So that's Gamma, and it's super cool how you can have it built with your exact template. You can export it, right, to Keynote or PowerPoint or PDF. That's right.

So what I will oftentimes do is I'll export it directly into Google Slides. And now it's asking me to authenticate, so we don't have to worry too much about that. But, yeah, I'm not going to open the app. But I promise you it works.

Audio Synthesis: Turning Notes Into a Podcast

um and so so just to recap we took photos we use image to transcribe it we got a corpus of everything we used it to create a deep research report which is uh currently cooking and we also brought it in and we asked it for a podcast and for it to make a podcast based on what this room said so let's try and play this and you might need to up the volume a little bit

But they are quietly training it to replace their entire badminton style. There's no gap between the public hype and that private reality. It's huge. It really is. So welcome to the deep dive.

Today, we have our hands on a truly fascinating data set. It comes from a digital transformation strategist who spends their days guiding major enterprise companies through these massive tech shifts. Right. They recently compiled these raw, unfiltered responses from a summit of enterprise executives. just to see what the actual practical adoption looks like at the very highest levels. Exactly.

And the mission for this deep dive is to cut straight through the butlers. We want to look at the raw data of what these leaders are doing in this very second, what they are desperately wishing for in the near future, and crucially, the massive strategic blind spots they are currently completely ignoring. Because looking at this data through the lens of organizational strategy, it's just, well, a surprising narrative that emerges. Yeah, we're seeing a distinct split between what leaders are doing today and the systemic overhauls they are hungering for tomorrow. Okay, let's unpack this.

Where is the baseline right now? What does the today of enterprise AI actually look like? Well, currently, the data shows that adoption is highly tactical. It's individualized, and it's very isolated. Isolated, huh? Meaning executives are leaning on AI for immediate discrete productivity gains. They are treating it like a highly capable, localized assistant for very specific tasks. Ah, rather than an integrated company -wide system. Exactly. It's a localized point solution right now. Right.

So we're talking about content generation and summarization. Okay. So that goes on, and it's 19 minutes.

And so what I'll oftentimes do is I'll commission reports, or I'll get long synthesis of data, and I'll bring in a notebook LM and create an audio version of it, and then walk around around my block and absorb it through audio. And I find it's a really effective way to learn new things. And you can even kind of raise your hand and then have conversations with the podcast partners about, oh, what about this part? Or what's the most interesting part of this data? And they'll talk back to you. So that is one of the coolest parts of Notebook LM.

Deep Research Results: A Citable Report

Now, I also want to bring back that we've been having a deep research report that's been cooking for us. Let's see if it hatched.

What do you think, Jeremy? You think it hatched? Okay. If it's Gemini, probably.

Yes. All right.

So Gemini came through. And so now, remember, I know I've been like going here and there and here and there, and I'm going to recap what it is I've done.

But we brought in everything that this room said, and then we asked it, hey, what would be a 20 or 30 page deep research report that we could run that would give relevant information to everybody in this room that would help

them and so it sent a team of agents off and it's saying the strategic blind spots of enterprise AI and now we can see here that there is a lot of really good information and let's see if it'll let me export to Docs here it's gonna ask me to open my app again so while that's loading you can see over here it's creating this document and if you scroll to the bottom it has citations

look at all the various websites it went through to research to find information here I'm seeing at least 30 different websites Gartner you know all different kinds Stanford I'm seeing Stanford law American progressives it's really been going the extra mile to research for us and not just regurgitate kind of the standard LLM response.

So yes, you can see it out here. Here's our deep research report. Let's see how many pages it is. 17 pages.

And what I'll oftentimes do with a deep research report like this is I'll take it back in the notebook LLM and I'll make a podcast out of it. And then I'll listen to it and talk to it.

But all of these works cited really has made it for me that I can have highly nuanced questions I want answers to, and I can send the AI agents off to get that exact answer that I'm looking for.

End-to-End Recap: The Full Pipeline

Okay, so let me recap some of the tools that we just went through. I took a bunch of pictures, we used image, and we transcribed all the images, and then we asked it to create a synthesis

of everything that the group has said, where we're at, where we're going tomorrow, what stands in the way and then from there we brought it in a notebook LM so that everything that we created from there was based just on what this room said not leaking out into the rest of the world from there in notebook LM we

asked it to help us write a brief for a research report and then we took that brief for the research report for a research report that would do what the research report would go out and so we have all the information of where we're we're at here in this room, but the research report would go out into the world, and it

would ask, what are some other case studies that we need to find out? What's going on that would help the people in this room bridge the gap from where they are today to where they want to be tomorrow?

Thank you, Jeremy. And so it created this research report for us.

We also asked it to create an executive -ready slide deck presentation, and then we brought Brought that over into Gamma, and it created it immediately for us. And we watched it work.

And then we came back, and we said, wow, I can't believe everything we were able to do with just what was built in this room. And a podcast. You forgot the podcast. Forgot the podcast. And we listened to the podcast as well.

And there's even more things you can do in Notebook LM here as well. So you can ask all kinds of questions. You can create a video overview. review. You can create infographics.

Conclusion and Reflection Prompt

That's the time that I have for this right now. But I just wanted to give you a quick sense of what's possible, and especially working with it like a teammate, giving it feedback, asking questions.

I didn't demo this for you, but I will oftentimes have it interview me and say, what questions do you have for me? So what I want to do now is I

want to go, and I just want to basically ask everybody to say, what from what I just showed you do you want to basically what do you want to incorporate into your workflow what was super exciting for you that you saw so turn to someone used next to you and say what would you love to start doing based on what you just saw and

we're gonna play music you've got about three four minutes that's my question for you what do you want to start doing based on what you just saw and thank you