×
Evaluate Your Team's AI Readiness

Take our official assessment and explore training paths to build secure, compliant AI workflows.

Access Portal

AI at Work Podcast

Ep #13 AI Ads, the Era of Zero-Click, and NotebookLM Upgrades

AI in Business, Featured Podcast

Audio episode

Listen to this episode

Summary

Elijah and Kevin trade war stories from the prompt trenches, and land on a counterintuitive rule, more context can make outputs worse. They unpack why Gemini “Gems” feel powerful but brittle, and how NotebookLM just changed the game by becoming a reusable source library that Gems can tap. From there, it turns into a practical playbook, curate your notebooks, keep lightweight brand style guides handy, and use purpose-built “skills” when you need up-to-date platform know how. The second half pivots to ads, trust, and a looming zero click future where the chat converts the sale.

Takeaways

  • Overprompting backfires fast, give the model clean context, then get out of its way.
  • NotebookLM as a shared source library plus Gems for repeatable workflows is a strong combo.
  • Keep brand guidance simple and current, a lightweight style guide beats a bloated brand book.
  • “Skills” are a practical workaround for stale model knowledge, wire them to living docs and repos.
  • Ads inside chat pushes us toward zero click buying, which changes brand control, trust, and attribution.
Chapters 00:00 Navigating AI Prompting Techniques 02:32 Exploring Google Gemini and Notebook LM 05:53 The Power of Notebook LM in Content Creation 08:30 Integrating Skills and Custom GPTs 11:51 The Fluidity of AI Tools and Their Ecosystem 14:43 Understanding AI Explainability and Inference 17:39 The Future of AI in Complex Projects 33:49 The Future of AI and Metaglasses 34:41 AI SuperBullets and Advertising Dynamics 38:35 Black Mirror and the Reality of AI Subscriptions 40:05 Google’s Advertising Strategy in the AI Era 41:24 The Cost of AI Advertising and Consumer Intent 42:54 Zero Click Engagement and Its Implications 44:31 E-commerce and the Shift in Consumer Behavior 47:13 The Role of Experience in Shopping 50:01 The Future of Travel Planning with AI 53:25 The Importance of Differentiation in a Commoditized Market 56:44 Navigating the New Advertising Landscape 01:00:27 Traction Points and the Future of Brand Engagement

Explore practical AI implementation: https://assessment.ascendlabs.ai/

Book a conversation with Kevin: tidycal.com/kevinwilliams

Show Notes & Links: https://www.spark6.com/podcast

Submit listener questions:

[email protected]

[email protected]

Check out Kevin’s stuff:

AscendAI: https://ascendlabs.ai/

Follow Kevin on LinkedIn: https://www.linkedin.com/in/kevinguywilliams/

Check out Eli’s Stuff:

SPARK6 Agency: https://www.spark6.com/

Sign up for FREE AI Framework Friday Newsletter: https://www.spark6.com/newsletter

Follow Elijah on LinkedIn: https://www.linkedin.com/in/elijahszasz/

Transcript and citation details

Transcript

00:00 Play from here What's up, Kevin?

00:13 Play from here Eli, how's it going?

00:14 Play from here All right, man.

00:15 Play from here I just saw in our chat that our producer asked me for some more photos of myself.

00:20 Play from here And I'm wondering if it's going to make me look more or less human in the intro.

00:24 Play from here I'm hoping even less human.

00:25 Play from here Maybe the more reference photos I provide, the more robotic and weird AI on Candy Valley

00:30 Play from here are.

00:31 Play from here You know, to other people...

00:32 Play from here Like when you prompt too much, like you give it too much context, you're like, I had a

00:34 Play from here better output when I just gave it one sentence.

00:36 Play from here You know, that's actually a great takeaway right there.

00:38 Play from here I had this conversation with an executive yesterday that you really get in trouble quickly

00:44 Play from here when you over prompt.

00:45 Play from here When you try and be super prescriptive and say, I want one, two, three, and four.

00:51 Play from here And the LLM, which has like Ogilvy and McKinsey and everything else built into its hive brain,

00:57 Play from here is like, yeah, okay, that's what you want.

01:00 Play from here I'll give it to you.

01:01 Play from here But you generally get better results by just backing off a little bit, making sure you

01:05 Play from here have like all the right pieces of context and then letting it do its thing.

01:08 Play from here That's what it's for.

01:09 Play from here It's super trippy.

01:10 Play from here I just, I had this exact experience yesterday building a gem in Gemini.

01:15 Play from here And, you know, I turned, I turned on the dictation and did a long rambling.

01:19 Play from here Like these are all the things I want in the instructions.

01:22 Play from here I said, well, let me ask Gemini how it feels about these instructions before I create the gem.

01:26 Play from here And it said, let me give you something better.

01:28 Play from here And I swear it cut like 70% out of what I said.

01:31 Play from here And I wondered, how did some of the important context lost?

01:34 Play from here And so I tried two different gems and its version was better.

01:38 Play from here It was just giving me better output as it, as it cleaned it all up.

01:41 Play from here Super weird.

01:42 Play from here So I hate to jump in with the practical here because we've got a bunch of fun stuff to talk about.

01:48 Play from here We have nothing but time, Kevin.

01:49 Play from here I've got all day for you.

01:52 Play from here This actually, I have to give the nod to this, this client who had discovered something that I hadn't become aware of,

01:58 Play from here which is the linkage between Google Gemini and Notebook LM.

02:02 Play from here Have you linked into this yet?

02:04 Play from here Explain a little bit more because I've been deep in the Notebook LM weeds lately and I got stuff for you too.

02:09 Play from here Okay.

02:10 Play from here Okay, great.

02:11 Play from here So I'm still very frustrated by the Google universe, by the way,

02:14 Play from here just working with it on a workspace level versus an individual account level is such a different deal.

02:20 Play from here Everybody has become, not everybody, but most people have their first LLM exposure to chat GPT.

02:26 Play from here And the memory function of chat GPT is quite good.

02:29 Play from here And it learns about you and such.

02:31 Play from here And that's the same is true about Gemini right up until you use a paid account.

02:36 Play from here And as soon as you use a paid account, it just nerfs its memory because of security issues.

02:42 Play from here So when you're creating a chat, you basically have to recreate that context.

02:47 Play from here And what we've been telling people to do is to build gems based on their company identity and their individual identity.

02:55 Play from here And then they operate within that gem.

02:58 Play from here So for those who don't know, a gem is like a custom GPT within Gemini.

03:03 Play from here Or a cloud project.

03:04 Play from here Or a cloud project.

03:05 Play from here You can give it custom instructions and you can give it documents.

03:08 Play from here But because the system won't remember you, you need to have these context documents that are inside the gem.

03:15 Play from here And that are creating guidance for you.

03:18 Play from here So that's been really frustrating to walk people through because it's not at all intuitive.

03:24 Play from here And it's not.

03:25 Play from here The loss of memory over time.

03:26 Play from here Well, the loss of memory over time.

03:28 Play from here And yeah, once you close that chat, you basically lose that memory of the chat.

03:35 Play from here Right.

03:36 Play from here You can't spin up another thread in that same gem and have it pick up where it left off.

03:40 Play from here No.

03:41 Play from here And you can't just go into the normal chat interface and assume that it knows anything about anything.

03:46 Play from here Because it won't.

03:47 Play from here You can use Google Drive documents, which can be more dynamic.

03:51 Play from here But you can only put 10 in.

03:53 Play from here Now, at the moment.

03:54 Play from here That will probably change over time.

03:56 Play from here And those documents can be really huge.

03:58 Play from here So we were telling people, hey, you want to create a Google Doc that has all of the information about your brand.

04:06 Play from here And what you do.

04:07 Play from here And et cetera.

04:08 Play from here And then have your team add those files to your gem.

04:11 Play from here And then have them add information about themselves and the things that they like.

04:16 Play from here And how they like to communicate.

04:17 Play from here We've talked about this before.

04:19 Play from here But even doing an oral interview with yourself.

04:21 Play from here Of like, you know, this is the way I like to communicate.

04:24 Play from here This is what I do.

04:25 Play from here Et cetera, et cetera.

04:27 Play from here That's all great.

04:28 Play from here But you have to actually build that structure.

04:31 Play from here The unlock is that Notebook LM is now includable as sources in gems.

04:39 Play from here So this wasn't a thing two weeks ago.

04:42 Play from here This stuff moves really fast.

04:44 Play from here But you can create a Notebook LM that's huge.

04:47 Play from here Notebook LM can have 300 plus documents in it.

04:50 Play from here That has a whole bunch of stuff going in.

04:53 Play from here And you can create one for your brand.

04:55 Play from here You can create one for your sales messaging.

04:57 Play from here You can create one for, you know, proposals, whatever it is.

05:01 Play from here And then you can create gems that call on that big data supply to do specific things.

05:09 Play from here And there's this argument now that I think isn't settled of if Notebook LM is so great at this stuff,

05:16 Play from here why do you use gems at all?

05:18 Play from here And I have an answer for that.

05:19 Play from here Do you?

05:20 Play from here I don't have an answer for that.

05:21 Play from here I didn't even know about this update.

05:23 Play from here So, well, let me start with this.

05:26 Play from here My first question to you is when I'm in Notebook LM, and for people who don't know,

05:30 Play from here Notebook LM, you can put a massive, massive amount of context, all sorts of documents.

05:36 Play from here You can paste information in.

05:37 Play from here You can upload files.

05:38 Play from here And then you can have various outputs, right?

05:40 Play from here You can either have a chat about all of the context inside that Notebook.

05:44 Play from here Or you can say, now give me a presentation.

05:47 Play from here Now make me a podcast.

05:48 Play from here Now give me an infographic, whatever it is, right?

05:52 Play from here But inside of that, you can checkbox all those different reference files either on or off,

05:56 Play from here depending on the output that you want.

05:58 Play from here So when you're calling that notebook from Gemini, can you do the same thing?

06:01 Play from here That is an interesting test.

06:03 Play from here Or is it going for the whole kit and caboodle, which could be a gazillion different things, right?

06:10 Play from here So the answer is I don't know.

06:11 Play from here But I'm pretty sure the answer is it's going to be the whole kit and caboodle.

06:15 Play from here It's going to be the whole thing.

06:16 Play from here Because that could be a lot.

06:17 Play from here It's a lot.

06:18 Play from here You need to think a little bit about your Notebook LM strategy.

06:21 Play from here The benefit being that you can now add multiple Notebook LMs.

06:26 Play from here And I've tested this with up to five.

06:28 Play from here And, indeed, it is pulling key sort of esoteric information from all five of those Notebook LMs.

06:34 Play from here To call from Gemini.

06:35 Play from here To call from Gemini or from a gem.

06:38 Play from here So you can add them as the gem resource.

06:40 Play from here Okay.

06:41 Play from here The entire Notebook.

06:42 Play from here The entire Notebook, yes.

06:45 Play from here But you're right.

06:46 Play from here This is going to require more curation.

06:48 Play from here Because it settles right back to what we talked about in the beginning here, right?

06:51 Play from here Is how much context is too much context without letting it do its voodoo that it does so well, right?

06:57 Play from here Like, you know, where is that tipping point?

06:59 Play from here With a million tokens of context, there's a lot that you can put in there.

07:03 Play from here There's a lot, yeah.

07:04 Play from here But getting more regimented about what are the LMs that we use?

07:08 Play from here And internally we have an LM that's dedicated to training.

07:12 Play from here We have an LM that's dedicated to proposals.

07:14 Play from here We have, you know, client-specific LMs, et cetera.

07:18 Play from here And it's really pretty cool to be able to match those data sources and be able to pull them either into a chat or into Gem.

07:27 Play from here Now, again, this lends the question of two weeks ago the best approach that we had was what I was describing earlier.

07:34 Play from here Build a gem.

07:35 Play from here Put the documents in it.

07:36 Play from here The whole thing felt really kludgy.

07:38 Play from here And it was like this can't possibly be the way that we have to do this.

07:42 Play from here And now that you have access to 300 documents and all of this information, that approach feels a heck of a lot less important.

07:49 Play from here And the answer at the moment has to do with functionality and what you're trying to achieve.

07:53 Play from here So things that you can't do in Notebook are use NanoBanana for image generation, although obviously the studio outputs are using NanoBanana.

08:03 Play from here But you can't just say, hey, make an image.

08:05 Play from here LM won't do code.

08:06 Play from here But can you call it?

08:07 Play from here Can you call your Notebook from Gemini and just enable the NanoBanana tool?

08:13 Play from here Yes, you can.

08:14 Play from here So in a gem, you can do create basically an extension of your Notebook LM that is for using the LM context and the information that's in there combined with like NanoBanana image generation.

08:30 Play from here You can do the same thing with code.

08:32 Play from here So another thing LM doesn't do is code output.

08:35 Play from here So if you have a need to output code that's tied to a specific project, I suspect what you're going to see people start doing is actually loading up Notebook LM with code bases such that they can interact with it that way really cost effectively versus OpenAI's codecs or cloud code or whatever you're using.

08:54 Play from here So that wouldn't be me necessarily, but I could see that.

08:57 Play from here So, yeah, there are things that you can still do in gems that you can't really do in Notebook.

09:02 Play from here So I think that the answer is kind of rolling all the way back to when OpenAI launched custom GPTs.

09:09 Play from here A gem is supposed to be a specific use case that is repeatable.

09:15 Play from here So it's a much narrower use case.

09:17 Play from here So you might have a brand book and a customer service SOP and such, but you have one gem that is designed to take in those LM pieces and respond very specifically to customer responses or whatever it is.

09:32 Play from here Direct to consumer brand, having a gem that's built on doing product listing, authoring, whatever it is that really understands your brand.

09:43 Play from here And now you have a lot more flexibility with the gems.

09:46 Play from here So you can build a complex gem that has the brand layer, that has the customer service layer, that has the sales layer, that is designed to do sales messaging, for example.

09:55 Play from here Listeners, if you find this exhausting to try to keep up with, you're not alone.

10:01 Play from here Listen, something to make note of is that the three Frontier Labs that we participate in the most, being Anthropic, OpenAI and Google, are doing so much so fast right now that personally, I feel like if I get too far into any one of them, all of a sudden I miss something in another one.

10:20 Play from here And I do use all three of these labs every day.

10:24 Play from here But, you know, I will tend to go deeper in one versus another, you know, Cloud Code, for example, right?

10:30 Play from here We got some Cloud Code stuff to talk about.

10:32 Play from here You can easily lose an entire day, two days, three days in that one product, just figuring out all that you can do with it, right?

10:40 Play from here And the same thing with Notebook LM, the thing I was going to share with Notebook LM is my wife is spinning up a new business based on this new kind of talk therapy model that came out relatively recently.

10:51 Play from here And I'm like, cool, what a great opportunity to see how many different AI tools I can kind of weave in here to help you get going.

10:58 Play from here And there's this one seminal piece of academic text that is really, really tough to consume.

11:05 Play from here It's not the kind of language you'd read and retain stuff, just citations broken up into every other paragraph.

11:11 Play from here And it's pretty brutal.

11:12 Play from here And I started dumping chapters of this into Notebook LM and turning it into a 45 minute podcast for her for each chapter.

11:20 Play from here And I can listen to it and I can retain this stuff and I can understand the concepts.

11:24 Play from here Whereas if I read a few pages of this academic text, it's like, yeah, I just have no idea what I even just read.

11:30 Play from here That in itself, like what a game changer to be able to completely flip the way that we consume complex information.

11:38 Play from here And inside, you know, so you've got your basic Notebook, right?

11:42 Play from here But then when you're going to prompt these different outputs for that notebook, in this case, a long form podcast, you can adjust the tone of that podcast.

11:50 Play from here You want it to be argumentative.

11:52 Play from here Do you want to be informative?

11:54 Play from here What do you want the link to be?

11:55 Play from here And you can put other prompts in there as far as, you know, like one of them that I added was at the end of each chapter.

12:02 Play from here Give me the 10 top takeaways and then give me a quick quiz.

12:05 Play from here Right.

12:06 Play from here And it did it incredibly.

12:08 Play from here I just so, so powerful.

12:09 Play from here Really awesome.

12:11 Play from here It's a really neat tool for sure.

12:13 Play from here So many different uses, even just came off of an ideation meeting and everybody on the meeting had come equipped with a Notebook LM slides.

12:22 Play from here And you almost have to get over the fact that they're gorgeous.

12:25 Play from here Not everybody is a brilliant designer.

12:27 Play from here Well, but you have to get over it because from a marketing perspective, these slides are really compelling.

12:33 Play from here Like this is the best slides that I've ever seen.

12:36 Play from here But they can be obscuring like really bad information because they're so pretty.

12:40 Play from here But, you know, are they actually telling the story you want it to tell?

12:44 Play from here And I won't get over the fact that from a presentation perspective, there's no way to edit your slides yet.

12:52 Play from here I know that they're going to fix this, but right now you have to jump through.

12:55 Play from here Didn't you find some little PowerPoint plugin?

12:57 Play from here Did that not end up working?

12:58 Play from here Yeah.

12:59 Play from here We use a tool called CODIA, C-O-D-I-A.

13:04 Play from here And I think it's about 10 cents a slide.

13:07 Play from here So for those of you who are using Notebook LM and you're just super frustrated by the watermark and, oh, it misspelled the client's name or it did little stuff.

13:15 Play from here For about 10 cents a slide, it will convert it into PowerPoint.

13:20 Play from here It takes some of the pizzazz away.

13:23 Play from here Like there's something that it's doing.

13:25 Play from here Now you're in PowerPoint.

13:26 Play from here That's the first problem.

13:27 Play from here You're in PowerPoint, but it turns every single element into its own object and it does strange stuff with text sizing and whatever.

13:34 Play from here It's still much better than nothing, but I just can't imagine that this is going to persist.

13:39 Play from here Like how is there not a Google Slides plugin to make all this stuff happen automagically such that you can edit it?

13:47 Play from here So, again, no relation with that platform, but that's just what I'm using.

13:52 Play from here And it works, sort of air quotes it works, but you will be frustrated.

13:57 Play from here So what I often do is I play output roulette with it.

14:01 Play from here So when I get to a prompt that I feel is getting almost client ready, I'll run it four times.

14:07 Play from here And because I guess there's a tip in here too, if you want the style to be consistent, make sure that you're adding a style guide to your notebook LM document store and not some fancy brand book that you spent a bunch of money developing.

14:23 Play from here It's just a Google Doc that has a list of like hex codes, fonts, et cetera.

14:30 Play from here And then you just refer to it and it does a pretty good job of adhering to it.

14:34 Play from here It doesn't like logos for whatever reason, but if you use a style guide.

14:39 Play from here Yeah.

14:39 Play from here it'll like just play games with your logo.

14:42 Play from here So I usually in my prompt will say, don't include a logo

14:44 Play from here and do reference the style guide,

14:47 Play from here but then I'll run it four times,

14:48 Play from here you know, draining small lakes in the Midwest.

14:50 Play from here I'm sorry, sorry world, I really am, this will get better.

14:54 Play from here But then being able to mix and match

14:57 Play from here from the individual slides, like what's the best of,

15:00 Play from here because it will follow the same pattern

15:02 Play from here of whatever the story arc was, because it's good at that.

15:05 Play from here And in set four, you can just grab slide seven

15:09 Play from here and drop it in.

15:10 Play from here So more and more, I'm getting away with that approach

15:12 Play from here versus trying to edit stuff.

15:13 Play from here When you said the draining the lakes thing,

15:15 Play from here this popped into my head that's been going around.

15:18 Play from here You had to have seen this.

15:19 Play from here It was like, I think I saw on TikToker Reels or something,

15:22 Play from here and it's this guy in his backyard and he yells out,

15:24 Play from here hey Google, why are tomatoes red?

15:27 Play from here And then it's him again in the backyard

15:29 Play from here and he starts to answer,

15:30 Play from here but then he jumps in front of himself

15:31 Play from here with two bottles of water and that's Gemini

15:34 Play from here just pouring out the water.

15:35 Play from here Let me tell you about all the different ways of chemistry

15:38 Play from here that a tomato, and it's like, no, I just wanted that.

15:41 Play from here And it's just pouring the water out.

15:42 Play from here For the record though, the data around water consumption,

15:46 Play from here because this comes up every time I speak

15:48 Play from here and it makes me feel terrible.

15:49 Play from here I hear about it all the time.

15:52 Play from here It was the Empire of AI book.

15:54 Play from here And there is a stat in there about water consumption

15:58 Play from here that is frankly wrong.

15:59 Play from here That does not make this an environmentally friendly

16:01 Play from here technology. The multiplier was off

16:02 Play from here by an order of magnitude, I think.

16:03 Play from here Yes, it was off by an order of magnitude.

16:05 Play from here So, I mean, there were plenty of issues with this,

16:07 Play from here but like, it is not quite as dire

16:10 Play from here as the media has led people to believe.

16:13 Play from here Again, not great, but not quite that bad.

16:16 Play from here So, I want to ask you about,

16:20 Play from here so this new development with notebook LM,

16:23 Play from here I imagine you're probably going to be using it even more now

16:25 Play from here and I think I will too.

16:26 Play from here And you talked about a custom GPT or a gem or a project

16:29 Play from here for being these very narrow use cases that are repeatable.

16:32 Play from here Do you see most of those starting to go away for you?

16:34 Play from here Because I find myself having less of those

16:39 Play from here and more of the deeper thinking projects, right?

16:42 Play from here That require more sophisticated output.

16:44 Play from here Oh, it's such a good question

16:45 Play from here because I think where we're heading

16:47 Play from here is giant spaghetti bowls of data

16:49 Play from here that have windows into the right pieces

16:51 Play from here and to a large degree, trusting the models

16:54 Play from here to make sense out of it.

16:56 Play from here And with these huge context limits

16:58 Play from here that like Google has a million tokens plus,

17:01 Play from here like it's much more plausible.

17:04 Play from here I like operationalizing GPTs

17:07 Play from here and I find them easier to operationalize than other things.

17:11 Play from here This is the tool that you use to do this.

17:13 Play from here And particularly if you're dealing with sort of

17:15 Play from here younger or less experienced staff,

17:17 Play from here it's just easier to have-

17:18 Play from here And then sharing those tools too, right?

17:20 Play from here And then sharing them.

17:21 Play from here But I will admit, I do have my GPT panel up

17:25 Play from here on another screen here

17:26 Play from here and there are like 10 or 15 custom GPTs that are there

17:30 Play from here that most of them are super forlorn.

17:32 Play from here I did open one that I built

17:34 Play from here to help me manage my accounting platform,

17:36 Play from here which is an SMB platform called Xero.

17:39 Play from here And I opened it and it was set to GPT 4.0,

17:42 Play from here which was so long ago.

17:44 Play from here And I haven't done any-

17:45 Play from here I have one of those too, that's so funny.

17:47 Play from here Like a little Xero assistant.

17:49 Play from here Well, it's my claim to fame.

17:52 Play from here I haven't checked in a minute,

17:53 Play from here but the last time I looked, it had about 6,000 uses

17:56 Play from here because I put it into the public store

17:58 Play from here like right after custom GPTs were launched.

18:01 Play from here In any case, it's not great.

18:03 Play from here It's not great.

18:04 Play from here It hasn't been updated.

18:05 Play from here This isn't the way I would do it.

18:06 Play from here I would actually do it as a Xero skill.

18:09 Play from here So we haven't talked about skills here

18:12 Play from here and it's a little bit of a rabbit hole,

18:13 Play from here but be aware that increasingly OpenAI

18:17 Play from here and Anthropic specifically,

18:20 Play from here allow you to develop skills for repeatable tasks.

18:23 Play from here And skills for repeatable tasks

18:25 Play from here are like this little bucket or a little box

18:28 Play from here of the things that it needs to understand

18:31 Play from here and it's a little bit like a sales process

18:33 Play from here where you have a certain platform or a certain approach

18:36 Play from here and it actually can be like a sales process too,

18:38 Play from here or how do we communicate with clients?

18:40 Play from here But I tend to use it for more technical things

18:43 Play from here where I'm trying to really encapsulate

18:46 Play from here the best of information about how to use

18:49 Play from here like a vibe coding platform like Lovable.

18:51 Play from here I have one for Lovable.

18:52 Play from here And it's more up to date than the training of the models.

18:57 Play from here So instead of the model like going out

19:00 Play from here and trying to get whatever to come up

19:01 Play from here with how you do something in Lovable,

19:03 Play from here it goes straight to the skill,

19:05 Play from here which saves on tokens, which saves on water,

19:07 Play from here which allows you to get closer to the result faster.

19:11 Play from here And that has much higher utility to me

19:14 Play from here than having as somebody who's doing

19:16 Play from here really dynamic things, right?

19:17 Play from here Than having like a GPT

19:20 Play from here that's trying to advise me on Lovable.

19:23 Play from here Right, yeah.

19:24 Play from here Yeah, so what other skills are you using?

19:26 Play from here Where are you using skills

19:28 Play from here where you wouldn't be using

19:29 Play from here one of these other custom tools or a notebook?

19:31 Play from here I mean, you can correct me if I'm wrong, by the way,

19:33 Play from here or if my gut is wrong.

19:35 Play from here I have them set up for all of the vibe coding tools.

19:37 Play from here So in my rotation right now is Cloud Code, Lovable, Repl.

19:43 Play from here I also have one that's configured for N8n,

19:47 Play from here our favorite workflow tool.

19:48 Play from here Xero, the accounting platform, HubSpot, et cetera.

19:52 Play from here So a lot, basically.

19:54 Play from here I'm using a lot of different skills

19:56 Play from here and that will probably go away too at some point.

20:00 Play from here But for the moment, they're just dynamic too.

20:03 Play from here And I love the fact that, again, we're non-technical here,

20:06 Play from here but you can take a skill

20:09 Play from here and you can call it into Cloud Code,

20:11 Play from here which is really pretty cool.

20:14 Play from here So if you want Cloud Code to understand Lovable

20:17 Play from here in a way that the Anthropic training set

20:20 Play from here won't understand Lovable,

20:22 Play from here you use a skill such that it can access it.

20:26 Play from here And now it understands better

20:28 Play from here how the dots are going to be connected

20:30 Play from here between your deterministic backend flow

20:33 Play from here and whatever cool thing you're building with Cloud Code.

20:36 Play from here Yeah.

20:37 Play from here Back to the point of it's all moving very, very quickly.

20:40 Play from here It is.

20:41 Play from here I mean, it's hard to get your arms around this stuff.

20:44 Play from here And you had made mention of Google's ecosystem

20:49 Play from here and how kind of disconnected everything is

20:53 Play from here in regards to where are the entry points

20:55 Play from here to use these different tools?

20:56 Play from here How do they work with each other?

20:57 Play from here How do you, you know, now a custom gem can call a notebook.

21:02 Play from here I didn't see any like big,

21:03 Play from here I remember, I think I saw a blog post

21:06 Play from here or a tweet or something referencing it.

21:09 Play from here I'm like, oh, I meant to talk to Kevin about that.

21:11 Play from here Then I never did.

21:12 Play from here And then also randomly,

21:14 Play from here sometimes like I see things just get turned off.

21:17 Play from here Yesterday, I was trying to use the Google Labs beta

21:20 Play from here as a tool call.

21:22 Play from here I can see it and it's disabled

21:23 Play from here and I just can't use it anymore.

21:24 Play from here And I asked Gemini, why can't I use it?

21:26 Play from here Said, oh, well, it's these betas are rolled out

21:28 Play from here to certain users.

21:29 Play from here I'm like, no, no, no.

21:30 Play from here I was using it yesterday.

21:31 Play from here Now I can't use it today.

21:32 Play from here Just somehow turned off from my account.

21:34 Play from here No idea why, right?

21:35 Play from here So these things are very, very fluid right now.

21:38 Play from here All of these tools and their availability

21:41 Play from here and which ones are best for different things.

21:44 Play from here I think you touched on something really important

21:46 Play from here when you mentioned your company's not style guide

21:49 Play from here but just all the brand attributes, right?

21:51 Play from here You can have so many versions of things like that

21:53 Play from here of these documents that just can live

21:56 Play from here in your Google Drive, for example,

21:58 Play from here like just as Google Docs.

21:59 Play from here And it is really important to keep those updated

22:02 Play from here because as all these changes roll out

22:04 Play from here and all of these different tools become available

22:06 Play from here or they change features,

22:07 Play from here the source material still matters.

22:09 Play from here And like having that ready to go

22:11 Play from here and then jump into the, you know,

22:12 Play from here who knows tomorrow Anthropic might have something better

22:16 Play from here than Notebook LM.

22:17 Play from here I mean, we really can't tell at this point

22:19 Play from here but having all of your source data ready to go

22:22 Play from here and jump into the other product is always super valuable.

22:24 Play from here So you just said something that again,

22:26 Play from here I just feel like I need to hit over and over and over again.

22:28 Play from here You asked Google Gemini

22:30 Play from here about something that happened yesterday.

22:32 Play from here Gemini doesn't know what happened yesterday.

22:34 Play from here None of the models know what happened yesterday.

22:37 Play from here The models know what happened six months ago.

22:39 Play from here So it becomes really, really challenging

22:42 Play from here to get contemporary information,

22:44 Play from here to work with the models

22:46 Play from here because the models don't know anything

22:47 Play from here about their current state.

22:49 Play from here What if we're calling a web search?

22:51 Play from here What if you're basically just pouring water out

22:53 Play from here to do a search?

22:54 Play from here That's basically what you're doing.

22:56 Play from here And to be fair, Gemini knows how to get information.

23:00 Play from here So theoretically it's going to go out

23:02 Play from here and it's going to ask its buddy Google

23:04 Play from here of how, why this happened.

23:06 Play from here And it got a snippet of information

23:08 Play from here and it knows what to do with that snippet of information

23:11 Play from here but it's essentially sitting there in a vacuum.

23:14 Play from here It's not something that it can compare

23:16 Play from here against all of its training information

23:18 Play from here about the dynamic aspects of what you've asked it.

23:21 Play from here It just knows the thing.

23:22 Play from here And people definitely don't understand this yet.

23:25 Play from here When we're doing training, they're like,

23:27 Play from here well, it seems like it's all up to date.

23:29 Play from here Well, it is, but it lags at least six months behind.

23:32 Play from here I actually don't know

23:33 Play from here what the current training cutoff is of Gemini

23:35 Play from here but it's probably around August of last year

23:38 Play from here is where most of them are sitting.

23:41 Play from here So if you're trying to get something new,

23:44 Play from here that's where these skills come in

23:45 Play from here because a well-built skill can do something clever

23:48 Play from here like access and GitHub repository on FAQ

23:54 Play from here on its frequently asked questions.

23:56 Play from here So some random person at N8N

23:58 Play from here goes and updates something in that Git

24:01 Play from here and because your skill is referencing the Git,

24:04 Play from here you're going to get the latest and greatest information.

24:06 Play from here Whereas if you just ask Gemini about how to do it,

24:10 Play from here it's going to only know how to do that thing

24:12 Play from here as of six months ago,

24:14 Play from here or it's only going to know what some bro on the internet

24:17 Play from here has put forward in a blog post.

24:19 Play from here What about deep research?

24:20 Play from here You know, when you're seeing it run through

24:22 Play from here all of the different sites,

24:23 Play from here it's kind of building out citations.

24:26 Play from here That's not six month old data.

24:27 Play from here That's going in real time,

24:29 Play from here trying to find a bunch of stuff to collect for you, correct?

24:32 Play from here Yeah, and it's an important difference,

24:33 Play from here but it's using search.

24:35 Play from here Okay, so it's an agent.

24:37 Play from here Like deep research is the first functional agent

24:41 Play from here that does many, many things

24:43 Play from here that most people are exposed to.

24:45 Play from here Most other agents kind of suck so far

24:47 Play from here because they don't have any orchestration

24:49 Play from here as far as what they're supposed to do,

24:51 Play from here but Google and all of the deep research functions

24:54 Play from here are very well built as far as what their mission in life is.

24:57 Play from here So it's multiple agent steps.

24:59 Play from here The first is a plan of what the user actually wants

25:02 Play from here and they usually ask questions about it.

25:05 Play from here The next is a search function

25:07 Play from here where they gather all of the information

25:10 Play from here and through gathering the density of information,

25:14 Play from here now they can apply their LLM skillset

25:17 Play from here of looking at that giant pile of information

25:20 Play from here to synthesize the pile of information

25:22 Play from here and then come out with a report that makes sense.

25:25 Play from here And yeah, you want to make sure that you're instructing

25:28 Play from here your deep research to stick to the newer material for sure,

25:32 Play from here because at some point there's no point

25:34 Play from here in doing a deep research.

25:35 Play from here You can just, if you're trying to do, I don't know,

25:39 Play from here something like a research report on the Battle of Waterloo,

25:42 Play from here something that's been long established.

25:44 Play from here Really you don't necessarily need to do deep research on it

25:47 Play from here because it's so well coordinated within the LLM itself

25:51 Play from here as far as how the LLMs consider

25:54 Play from here the various intricacies of it.

25:55 Play from here It's where you're stringing together pieces of information

25:58 Play from here that aren't naturally, naturally air quoted,

26:02 Play from here like trained into the models

26:04 Play from here to create associations between them.

26:07 Play from here So you're sucking it all into a big bucket.

26:09 Play from here That's agent number two.

26:11 Play from here And agent number three is looking at the bucket

26:13 Play from here and it's making sense of it based on its lens

26:16 Play from here as a research assistant that can gauge the different pieces.

26:21 Play from here So it's incredibly valuable

26:22 Play from here because the output will be unique,

26:24 Play from here particularly if you're doing it right.

26:26 Play from here Like, you know, I want a usage guide

26:28 Play from here for left-handed screwdrivers, you know,

26:31 Play from here based in, you know, for plumbers in Moldova

26:35 Play from here or whatever the heck it is.

26:36 Play from here And you're pulling things from-

26:37 Play from here Very popular use case.

26:38 Play from here Very popular, especially the left-handed screwdriver.

26:41 Play from here But now you have this context piece.

26:45 Play from here And then the next step is to use that somewhere.

26:47 Play from here So we'll generally put that into a notebook LLM.

26:50 Play from here By the way, you can do deep research right in notebook LLM.

26:54 Play from here So as like a project hub,

26:56 Play from here like we're working really hard on deciphering

26:59 Play from here what, you know, the Google verse it means

27:02 Play from here as far as like product IDs and crazy stuff like that.

27:07 Play from here So we're constantly doing research

27:09 Play from here on the latest and greatest technologies

27:12 Play from here and accumulating all of those into a notebook LLM

27:14 Play from here such that we can work with it

27:16 Play from here in a way that's more constrained.

27:18 Play from here So that's the other, other, other attribute

27:21 Play from here of notebook LLMs that people need to remember.

27:23 Play from here And reasons to use an LLM versus reasons to use

27:26 Play from here just Gemini or a Gem is that a notebook LLM

27:30 Play from here is constrained by the data that's in it.

27:32 Play from here So it only really can consider the information that it has

27:36 Play from here or the information that you've added.

27:38 Play from here Whereas if you have a,

27:40 Play from here take a deep research report in Gemini,

27:42 Play from here now you can just go out to the wild LLM universe

27:45 Play from here and ask other questions.

27:46 Play from here What I'll commonly do is I'll do that

27:48 Play from here and then put that the research findings

27:50 Play from here into notebook LLM as reference material.

27:53 Play from here Yeah.

27:54 Play from here This stuff just starts to bend your mind after a while

27:56 Play from here when you think of how far you can take

27:58 Play from here any one of these single directions.

28:00 Play from here It's pretty wild.

28:01 Play from here Here's this made me think of something as well.

28:03 Play from here When you were referencing all of just the kind of agent

28:07 Play from here stacks and the processes that these run through

28:10 Play from here as you string along all of these different,

28:12 Play from here like a research project, for example,

28:13 Play from here when you are seeing all of the text come up

28:17 Play from here as something is running its report,

28:20 Play from here I've heard varying things as to, you know,

28:23 Play from here that's really just a glorified loading screen

28:26 Play from here to give the users something to read while time passes

28:30 Play from here or no, this model is actually thinking

28:33 Play from here and you're seeing, you know,

28:34 Play from here you're seeing all the sequential steps happening in the UI.

28:38 Play from here Have you heard anything about this?

28:40 Play from here Look, on one side,

28:41 Play from here one of the biggest weaknesses of AI is explainability.

28:46 Play from here So there is definitely an aspect of explainability to this

28:51 Play from here such that theoretically you can explain

28:54 Play from here why the LLM did what it did.

28:57 Play from here And this is super important

28:59 Play from here in like high risk data applications.

29:01 Play from here If you're doing anything that's involving hiring,

29:04 Play from here financial decisions, insurance, healthcare, et cetera,

29:08 Play from here if something goes haywire,

29:09 Play from here you need to be able to explain it.

29:11 Play from here So-

29:11 Play from here If you put in a prompt like that,

29:13 Play from here those are the first little messages

29:14 Play from here that you always see come up.

29:15 Play from here Oh, the user seems to be asking some healthcare questions.

29:19 Play from here Make sure that-

29:19 Play from here that you say that this is only for informational purposes.

29:22 Play from here I'm not, you know, user seems very troubled.

29:25 Play from here Maybe tell them to call 911.

29:26 Play from here All sorts of random stuff come in and I'm not saying I was like, you know,

29:30 Play from here doing anything harmful, but it's interesting to see what messages come up.

29:33 Play from here So the fact is though, my understanding is that that isn't just display theater

29:40 Play from here because if it was just display theater, it wouldn't tick that box.

29:43 Play from here So it is the sequential prompts and actions that are declarative of what

29:51 Play from here the LLM is doing at that moment.

29:53 Play from here And does it have to do it that way?

29:56 Play from here Actually, I don't, I don't think it does, but being more, they call it, you

30:00 Play from here know, inference, inference oriented and moving process wise through things.

30:06 Play from here I believe it does give it better results.

30:08 Play from here And parts of this tie in with what we were talking about previously about

30:11 Play from here like the anthropic constitution and what all of this means, because now you're

30:17 Play from here leaving more than breadcrumbs behind you.

30:19 Play from here You're leaving like a bread road behind you of like everything you're doing.

30:23 Play from here And it's not just a series of magic outputs and the LLM is going to be using

30:27 Play from here that process for how to work with you in the future as well.

30:32 Play from here Yeah.

30:32 Play from here I thought that'd be so cool to actually see that with people, you

30:36 Play from here know, as you're talking to them, like, what are those little things going on

30:38 Play from here in their head as you're having these conversations?

30:40 Play from here So I actually took all of our transcripts and I put them into a

30:43 Play from here model and then basically had it output like a Kevin internal processing log.

30:49 Play from here It was pretty amazing.

30:50 Play from here Like during, during, during the times, during the times that you talk, it said

30:54 Play from here things like a loading skepticism model already loaded.

30:57 Play from here Why is it loading again?

30:58 Play from here That was one of them.

30:59 Play from here I thought that was good.

31:00 Play from here No way.

31:01 Play from here Really?

31:02 Play from here Confidence, confidence detected running additional verification loop.

31:06 Play from here This sounds impressive.

31:07 Play from here That is rarely a good sign.

31:08 Play from here Attempting to imagine this in a quarterly planning meeting, failing, pausing to ask

31:13 Play from here who is, who is on call when this breaks at 2 AM, this is either the future

31:17 Play from here or a very well-funded mistake.

31:18 Play from here Adjusting sweater vest, preparing measured response.

31:21 Play from here I liked that one.

31:22 Play from here You put that one in there.

31:24 Play from here I swear I didn't.

31:25 Play from here This was Gemini.

31:26 Play from here This was all Gemini.

31:28 Play from here Pausing to choose words that won't alarm anyone unnecessarily.

31:31 Play from here That's one of the funniest things I've heard.

31:33 Play from here So maybe, maybe that'll be a feature in the next Metaglasses, right?

31:36 Play from here We can put those on.

31:37 Play from here And as we're talking to somebody and it's listening to the conversation in real time,

31:41 Play from here we can see all these little thinking modules going through the processes.

31:44 Play from here I'd quite enjoy that.

31:45 Play from here I hope they're as good as the ones that Gemini did for you.

31:47 Play from here So let me get this clear.

31:48 Play from here You're doing videos in there or are you putting the transcripts?

31:51 Play from here Transcripts, yeah.

31:51 Play from here What do you know about my sweater vest?

31:53 Play from here I haven't worn a sweater vest in years for the record.

31:55 Play from here But I make comments on your sweater vest.

31:57 Play from here You do.

31:58 Play from here Yeah.

31:58 Play from here Or Patagonia vest, whatever.

31:59 Play from here Patagonia vest, you know, I'm that guy.

32:01 Play from here Interchangeable.

32:03 Play from here All right.

32:03 Play from here So listen, now that we got that out of the way, how'd you enjoy

32:06 Play from here all of your AI Superbowl ads?

32:08 Play from here Boy, so I have mixed emotions that, all right, first, Anthropic ran a bunch of ads

32:15 Play from here and they were basically ads about ads because Anthropic has been positioning

32:21 Play from here itself as the real, real soulful AI that's going to solve all your problems.

32:27 Play from here The people's AI, as it were.

32:30 Play from here And meanwhile, OpenAI, like opening up the floodgates to ads, at least in the free

32:35 Play from here models, and anybody who knows anything about online marketers knows that

32:39 Play from here digital marketers run everything.

32:42 Play from here That's just the way of the world.

32:43 Play from here Like once they get themselves a new toy, they're going to

32:46 Play from here lever it to the nth degree.

32:48 Play from here And whatever the intent is of OpenAI, it is going to get

32:52 Play from here perverted in some way, shape, or form.

32:54 Play from here And the Anthropic ads were really just, just jabbing Mr.

32:58 Play from here Altman right in the eye, where they would have like, you know, a

33:01 Play from here young man who's doing pull-ups.

33:03 Play from here Who happened to look like Altman?

33:05 Play from here He did look a little bit like Altman.

33:06 Play from here He did.

33:07 Play from here Yeah.

33:07 Play from here I think, I think that was intentional.

33:08 Play from here That was another jab.

33:10 Play from here Yep.

33:10 Play from here Um, asking how to like put on some muscle.

33:13 Play from here Yeah.

33:13 Play from here Asking how to put on some muscle and like this, like trainer appears and they did a

33:17 Play from here really good job in casting because current generation GPT does have a

33:22 Play from here very strange pedantic vibe to it.

33:25 Play from here It's like this sycophantic, not, it's not as bad as it was.

33:29 Play from here It's not as sticky, but it's very formulaic.

33:32 Play from here It's still sycophantic though.

33:33 Play from here Yeah.

33:33 Play from here It still is.

33:34 Play from here And this like giant gym trainer appears and is giving like good advice to the

33:38 Play from here person and then interjects with like, try muscle milk, or it was like an

33:43 Play from here insole or something that added a push.

33:45 Play from here It was an insole to make him taller, which I think was yet another jab.

33:48 Play from here Yeah.

33:48 Play from here And it was, it was brilliant.

33:50 Play from here It was brilliant advertising in that you could see the user just melt.

33:55 Play from here Like the acting was really good.

33:56 Play from here I mean, what do I care about commercial acting?

33:58 Play from here But I thought it was, it was really well done.

34:01 Play from here That in each example, it was, it was stuff that hit

34:04 Play from here right to the core of how people use this stuff.

34:07 Play from here As far as getting really like kind of intimate sentiment, sensitive advice

34:12 Play from here and what this is all going to mean.

34:14 Play from here So that's sort of the way of the land there.

34:16 Play from here I'm going to say it was a bit of an own goal at the end of the day, because you

34:20 Play from here and I get this and you and I feel this, but I don't think that the population

34:25 Play from here at large really gets these nuances so much or even knows that open AI is going

34:32 Play from here to wear this when it's just high.

34:35 Play from here It's like a joke in the joke, in the joke that nobody got.

34:38 Play from here I mean, they spent, yeah, exactly.

34:40 Play from here Like, like four people in the know, this was like really hilarious and, and like,

34:45 Play from here like pointed and it's not wrong.

34:47 Play from here Like that's, this is, this is all going to, as I said a second ago,

34:49 Play from here it's going to get perverted.

34:51 Play from here But if you're like a random person and you haven't used many of these tools,

34:57 Play from here this just convinced you that this was just yet another, like, like fraud being

35:02 Play from here perpetuated on Americans to like pull their attention into yet another platform.

35:08 Play from here Like you probably decreased the overall feeling about AI on a macro

35:13 Play from here level, not specific to open AI.

35:16 Play from here You just made a bunch of people more suspicious about AI.

35:18 Play from here So good job guys.

35:20 Play from here But they were funny.

35:21 Play from here They were really funny.

35:22 Play from here They were really funny.

35:23 Play from here For the people, for the people who spend a lot of time in this world and do

35:26 Play from here understand all the nuance, they're really funny.

35:28 Play from here The, the guy talking to his therapist about how to better communicate with his

35:32 Play from here mother and then the therapist says, well, and if that doesn't work and you still

35:36 Play from here want to maybe get to know a hotter, older woman, you could try cougar.com or

35:41 Play from here whatever it was, and he's like, what would you like me to create you a profile?

35:44 Play from here So I'm, I'm a big fan of the, the, the British show black mirror, and I'm

35:49 Play from here sure you know a bit too, and there was an episode, I don't know, a couple of

35:51 Play from here seasons ago about a woman who gets an implant and it saves her life.

35:56 Play from here And then basically the company goes out of business or gets sold to somebody else.

36:00 Play from here And now she can't afford to pay the subscription and without

36:04 Play from here it, it was all on a SAS model, right?

36:05 Play from here It had limitations.

36:07 Play from here Like she couldn't drive outside of a certain, yep, geography.

36:11 Play from here And there are all of these different updates that were different

36:14 Play from here tiers on the price plan.

36:16 Play from here And yeah, yeah, but she was a teacher and basically it was an ad supported model.

36:21 Play from here And just out of nowhere, she'd start spouting about like Limpton

36:25 Play from here iced tea or whatever the product was.

36:27 Play from here And obviously it ruined her life because that's the way everything in black

36:30 Play from here mirror ends, but boy, talk about getting very cheery, cheery show.

36:34 Play from here I was feeling really good.

36:34 Play from here I think that was even before LLMs, that particular episode.

36:38 Play from here So, so it's prescient in, in a ton of ways that when you add ads to these

36:42 Play from here ecosystems, like I, the point here is that there's a level of intimacy and

36:48 Play from here trust that isn't available on Google.

36:50 Play from here And I actually think that Google itself is keenly aware of this and I'm very

36:56 Play from here interested to see what direction it goes.

36:58 Play from here On the one hand, it feels like open AI just gave Google a free pass

37:03 Play from here to like start running ads.

37:05 Play from here And Google does, I would say in most consumer perception, Google does

37:08 Play from here nothing better than running ads.

37:10 Play from here The pushback that I feel like I'm like 20, 30% on the pushback on this,

37:14 Play from here because Google's an ad machine.

37:16 Play from here Google knows how to do ads better than anybody else.

37:18 Play from here Google's going to add.

37:19 Play from here That's what Google's going to do is the picture here is way bigger than ads.

37:23 Play from here Yeah.

37:23 Play from here The global ad business for Google is $200 billion, but when you start

37:27 Play from here looking at this broader market opportunity, as far as LLMs for

37:32 Play from here productivity and integrated into everything, you're into the

37:35 Play from here trillions of dollars annually.

37:37 Play from here So if Google decides just to wash its hands of the ad model in favor of

37:43 Play from here leaning into a really trustworthy large language model, that might be

37:49 Play from here a really smart market decision to make.

37:51 Play from here Because I wouldn't advise anybody to use open AI in the free model

37:55 Play from here once it starts spouting ads, because there's just no way that

37:58 Play from here that's not going to get abused.

37:59 Play from here Yeah.

38:00 Play from here I mean, going back, Google is synonymous with ads.

38:04 Play from here You know, you kind of have like the duopoly of Google and meta for

38:07 Play from here advertising online and you're going to use one or the other, most likely

38:11 Play from here both, if you are especially some of these smaller brands, right?

38:15 Play from here These DTC brands, you have a Shopify site, you're probably in both

38:18 Play from here ecosystems and I agree, this is a bit tabula rasa, this whole platform.

38:23 Play from here You could go a completely different direction and it does feel like

38:27 Play from here they were just handed a bit of a hall pass with this by putting ads in.

38:30 Play from here Also to mention, the ads are predicted to be really expensive.

38:34 Play from here I don't know if you saw any of the preliminary pricing, but wow.

38:38 Play from here Wow.

38:39 Play from here Wow.

38:39 Play from here So to give people some context, they're talking about a $60 CPM average, right?

38:45 Play from here CPM is like a cost for a thousand impressions.

38:48 Play from here And, you know, so if you go to like LinkedIn, people are like, oh,

38:51 Play from here LinkedIn is super, super expensive.

38:53 Play from here That's, you know, it's around 30, 40 bucks per CPM.

38:56 Play from here YouTube is around 20, meta is around $8, TikTok's $5, Google

39:01 Play from here display ads are like two bucks.

39:04 Play from here So, you know, we're, we're talking a lot of money and the rationale behind

39:09 Play from here that is that they're saying that intent is super, super high.

39:13 Play from here If somebody is having a conversation about something inside of one of these

39:17 Play from here chats and then all of a sudden they're served a link where they can directly

39:20 Play from here buy from there, you know, there's also a lot of speculation as to it being a

39:24 Play from here totally different experience than seeing a blue link in Google search results.

39:28 Play from here You know, you would just purchase it right there inside of the chat.

39:30 Play from here All the hooks would be established, right?

39:32 Play from here That's also a really big deal.

39:34 Play from here You know, it's no longer like, did somebody now come to my website and

39:38 Play from here did they put it in the cart and did I get a conversion?

39:41 Play from here It's like, no, you're just converting right inside of the chat.

39:43 Play from here So that's the argument for that really high CPM.

39:46 Play from here Okay.

39:46 Play from here All right.

39:47 Play from here So we call that a zero click engagement.

39:50 Play from here This is my old marketing shops coming in.

39:52 Play from here So like when you toss up an ad, you want somebody to click on

39:56 Play from here it such that something happens.

39:58 Play from here And if you're moving to a zero click environment, then the ad unit is no

40:05 Play from here longer pushing you to the website, unless there's some reason to push you to the

40:09 Play from here website, it's just pushing you to that transaction and that changes the compact

40:14 Play from here between the ad networks like Google or now OpenAI out there and the brands that

40:21 Play from here they're serving, because how the heck does a brand like optimize a consumer

40:28 Play from here experience if the consumer never even makes it to their environment?

40:33 Play from here They're beholden to the black box that is Google or that is OpenAI, that is

40:38 Play from here o-anthropic to serve up the right stuff at the right time and because it's not

40:44 Play from here 10 blue links, so that's what, you know, traditionally Google would put up these

40:47 Play from here 10, 10 paid links, it's going to be like one or two options.

40:51 Play from here So that's what everybody is clamoring for right now.

40:54 Play from here And we're doing a lot of work in that respect of trying to help brands develop

40:59 Play from here the right traction points around, by getting the data out there, it just

41:05 Play from here changes the dynamic of, of how advertising has always worked.

41:10 Play from here And it's a little hard for me to play out whether that actually

41:13 Play from here makes sense for brands or not.

41:15 Play from here I mean, it's really early to even understand, is that going to work?

41:19 Play from here Is that behavior going to stick?

41:21 Play from here Because if you think about how it works right now, if you click on one of those

41:25 Play from here blue links, very often, one of them, if not more than one is an Amazon link.

41:28 Play from here And now you're in this massive ecosystem where you're comparison shopping,

41:32 Play from here you're looking at suggested products.

41:34 Play from here Uh, it's, it's feel like e-commerce has taken this direction where it's very

41:39 Play from here rarely one and done a great example is that myself and most people I know would

41:44 Play from here prefer to buy something through their prime account because they know the

41:47 Play from here shipping is going to be free.

41:48 Play from here And if they don't get what they expected, they can easily just go to

41:52 Play from here Whole Foods and drop it off and get their money back.

41:54 Play from here Right.

41:55 Play from here Which isn't quite as seamless with a lot of the DTC purchases.

41:59 Play from here So Shopify sites, I think what you just said is actually really telling

42:03 Play from here because you know, who doesn't add, allow open AI or Gemini access to their

42:07 Play from here catalog, Amazon, because I mean, Amazon love them or hate them.

42:12 Play from here They are wicked smart and they are clearly making a bet that it is a better

42:18 Play from here consumer experience for their people.

42:20 Play from here And for both, you know, remember there are Amazon, there are

42:24 Play from here two types of Amazon customers.

42:25 Play from here There are us buying weird widgets on Amazon.

42:29 Play from here And then in large respect, there are people selling those widgets

42:31 Play from here that are on the other side.

42:32 Play from here Those are both customers.

42:34 Play from here So Amazon is blocking the access to the rest of these protocols that Shopify

42:39 Play from here is opening its floodgates to.

42:41 Play from here So Shopify is, you know, the biggest e-commerce platform out there.

42:45 Play from here They're all in on this.

42:47 Play from here Amazon is saying absolutely not because they clearly are feeling like this is

42:51 Play from here going to cheapen the experience for their brand and they're probably right.

42:56 Play from here Now being Amazon versus being some also ran little like e-commerce seller who

43:00 Play from here sells like muscle rubs or whatever, like, you know, that's probably a different world.

43:06 Play from here And most of those guys are also selling on Amazon, by the way.

43:08 Play from here Well, most of them are selling on Amazon too, but anybody who's

43:11 Play from here selling anything online right now.

43:12 Play from here And by the way, selling online doesn't mean just selling a widget.

43:17 Play from here It means there's some sort of a transaction happening that could be,

43:20 Play from here you know, you're a dentist and you have an appointment setting tool.

43:24 Play from here You're a property management group and somebody books a property.

43:28 Play from here Like these are, these are handshakes.

43:29 Play from here You're an event coordinator and you're selling tickets.

43:30 Play from here Yeah.

43:31 Play from here You're event coordinators and you know, often there's money that's involved, but not always.

43:36 Play from here So you need to take a breath and look at what's going on here and imagine what

43:40 Play from here your business looks like in a zero click environment, like what role are you

43:45 Play from here playing and what does that do to you?

43:46 Play from here And for some, it could be amazing.

43:48 Play from here It could be absolutely amazing that, that great.

43:51 Play from here I basically have no website and I've somehow won the LLM game.

43:54 Play from here And now I just dominate my particular vertical forever, or at

43:58 Play from here least for the next little bit.

43:59 Play from here But.

43:59 Play from here It's, we're slowly and we meaning again, recovering digital marketer, we're

44:04 Play from here slowly losing the tools that we've had to have granular control over how our

44:12 Play from here brand appears and where and ownership of the customer and I deliberately just

44:16 Play from here slowed down because I'm thinking through it as your bot would say, is he dubious?

44:21 Play from here I don't know.

44:22 Play from here Wait, do you want me to pull up some more of them?

44:24 Play from here No, no, no, no, please don't, please don't.

44:27 Play from here Oh, here's one, safety, safety mode enabled.

44:29 Play from here And three out of 10, it's just a safety mode, safety mode for sure.

44:34 Play from here But if you're advertising on Meta, there's like different fields of talk,

44:37 Play from here but Meta wants you to basically use their AI engine to build ads and it's

44:42 Play from here then going to optimize them for you.

44:44 Play from here And okay.

44:46 Play from here But there's a reason that I want to do the things that I want to do with my

44:49 Play from here brand and abdicating that maybe it's okay if you're, if you don't have a clear

44:54 Play from here brand vision, but losing control of your customer that way I think is going to

44:59 Play from here be a mistake for pretty much any brand in the longterm, winners will gloat

45:04 Play from here because they will be big winners and losers are going to really, really have

45:08 Play from here trouble being surfaced in the future.

45:11 Play from here So I'm so curious, Kevin, about how, how much of this is going to stick

45:16 Play from here and how much of it is just a test, right?

45:18 Play from here Like, is there going to be some, some, some, some blowback if all of a sudden,

45:22 Play from here you know, these mega companies are trying to shepherd consumers in one direction

45:27 Play from here and they just don't buy it, right?

45:28 Play from here And then, no, no, actually we liked just doing it this other way.

45:32 Play from here We're not going to buy directly inside of this chat.

45:35 Play from here It's really a bit of the wild West in that regard.

45:38 Play from here I just, just like how I said, most people I know, myself included,

45:42 Play from here would rather just go to Amazon, love them or hate them just because

45:45 Play from here it's such an easy experience.

45:47 Play from here And if things go sideways, it's easy to turn that around, right?

45:50 Play from here It's easy to just pack that up, send it back and you're done.

45:53 Play from here And I think it's worth mentioning too, that, you know, Amazon has its own

45:56 Play from here intelligence layer on top of its marketplace as well.

45:58 Play from here Have you, do you use that often, the Rufus?

46:01 Play from here Rufus?

46:02 Play from here Yeah, I think it's very basic.

46:04 Play from here I think it's clever in that it's, it's focusing the user's

46:07 Play from here attention, but not excessive.

46:09 Play from here Norm, I'm a big shopper by the way.

46:11 Play from here So it's, so it's like, I am not, or you are a big shopper.

46:13 Play from here I'm not, I'm not at all.

46:14 Play from here I don't, I don't get really any pleasure from shopping.

46:17 Play from here So there's, there's this as a user behavior aspect too.

46:20 Play from here And there's a reason that, that, that brands maintain flagship stores.

46:24 Play from here Brand building exercise, but it's also a tactile piece

46:28 Play from here of leisure for lots of people.

46:30 Play from here Whether it should be or not is kind of sort of a different societal

46:32 Play from here question, but people like to shop.

46:34 Play from here And if you take the shopping experience away from them, like now you're

46:39 Play from here just down to the transactional, like I need whatever plumbing thing that I

46:43 Play from here need, get it to me as fast as you humanly can for the lowest price possible.

46:46 Play from here There's a lot of that.

46:47 Play from here And I, you could argue that already happened with e-commerce, right?

46:51 Play from here It took away the shopping experience of going around and looking at the windows

46:55 Play from here and walking through a mall and all that stuff, and that really eroded quickly.

46:59 Play from here Now people's shopping experience is sitting there staring at

47:02 Play from here their phone or at their desktop.

47:04 Play from here And I'm not a very big consumer either, but I think it makes it more so that

47:07 Play from here when I do buy something, my over-controlled personality wants to

47:10 Play from here make sure I got the best one.

47:11 Play from here And so it wasn't that long ago in time that I would spend a lot of time, more

47:17 Play from here than I care to admit, reading reviews and then wondering if the reviews were

47:21 Play from here even real, with something like Rufus, you can just do that from a text window.

47:27 Play from here Just say, Hey, based on the 5,000 reviews, this product has, what are the

47:32 Play from here most common, yeah, bad things that people experience or what are the best things?

47:37 Play from here And we'll just give you those summaries.

47:38 Play from here Right?

47:39 Play from here So, so, so even that, like, does that take away the shopping experience?

47:42 Play from here Like I hated that part of the experience.

47:44 Play from here I'm a lot more utilitarian when it comes to, I need something.

47:47 Play from here I just want to find the shortest distance between these two points

47:50 Play from here to be able to buy that thing.

47:52 Play from here Whereas other people, I really think it's going to be just really a whole new

47:57 Play from here territory of what do people really feel comfortable with, what do they lean

48:02 Play from here into and what do they push back on?

48:03 Play from here It's a, it's an interesting time with, with advertising.

48:06 Play from here And I'm really, really curious about how quickly it's going to change

48:10 Play from here when these new things are rolled up.

48:11 Play from here You, you brought up, you brought up the meta ads with how Andromeda, which is

48:15 Play from here their, their AI powered ad platform now is rewarding AI generated creative assets.

48:20 Play from here Right?

48:21 Play from here It's actually, those are performing much better than these organic assets

48:24 Play from here where people got cameras and lights and actors and actresses and scripts.

48:29 Play from here And did ads and now they're just all AI generated and they're doing much better.

48:33 Play from here Is that going to be a forever thing or is that going to change?

48:35 Play from here And then all of a sudden organic media is going to be rewarded on ad platforms.

48:39 Play from here No idea.

48:40 Play from here No idea.

48:41 Play from here So you, I think you answered some of the macro questions yourself in there.

48:44 Play from here One is that, nobody knows anything about what's going on, but commodities

48:51 Play from here will be further commoditized that it is not a good time to be an also ran brand.

48:56 Play from here So in Amazon world, the joke is garlic presses.

49:00 Play from here It's sort of what Amazon influencers are always trying to show.

49:03 Play from here Like garlic presses as an example, like, yeah, do you need one?

49:06 Play from here Not really, but it's kind of nice to have.

49:09 Play from here And there are like 17,000 sellers of Amazon garlic presses out there.

49:13 Play from here Like it is not a good time to be that person because there's, there's no real

49:19 Play from here marketing opportunity to differentiate yourself as the user, if they want it,

49:24 Play from here they want it, they're going to snap it up.

49:26 Play from here And anything that's like transactional like that is going to be problematic.

49:30 Play from here I think that, that where this is going to show up first.

49:32 Play from here And I think I said this last week is in, is B2B actually.

49:36 Play from here So if you're buying, you know, if you're buying one hammer as a consumer,

49:41 Play from here like you're going to look at Rufus, you're going to kind of do some research, but if

49:45 Play from here you're a construction purchaser and you're buying a thousand hammers, no, you know

49:49 Play from here what your spec is, you're going to send your agent out.

49:51 Play from here The agent's going to buy the hammers.

49:52 Play from here Nobody else is going to even have a shot at it.

49:54 Play from here It's going to be price-based based on the lowest common denominator.

49:58 Play from here And then on the other side is experiential shopping.

50:01 Play from here So I don't, I don't see physical high-end retail going away anytime

50:06 Play from here soon that it's, it is a good experience.

50:09 Play from here You go into even like a Barnes and Noble, like it is a good experience.

50:14 Play from here You know, who reads books?

50:15 Play from here I read books, but it's a, it's a great experience.

50:19 Play from here Yes.

50:19 Play from here Actual paper.

50:20 Play from here You go into the bookstore.

50:21 Play from here Yes.

50:22 Play from here Didn't they all close?

50:23 Play from here Wait, we have like one left in sugar house.

50:25 Play from here I think there's a, there's one in park city too, but there's, there's a good

50:29 Play from here shopping experience.

50:30 Play from here It's, it's, it's all part of the brand.

50:33 Play from here That's great.

50:34 Play from here I think another area would be in travel in that.

50:38 Play from here Yes.

50:38 Play from here It's a primary use of the LLMs to help you travel plan.

50:42 Play from here But getting a better, getting an itinerary is one thing, getting

50:46 Play from here a feel for travel is another.

50:48 Play from here So I see my wife doing this.

50:50 Play from here She's part of, part of the pleasure of travel for her is the travel planning.

50:54 Play from here She really likes it.

50:55 Play from here And it is her favorite way to spend my time, but my wife's the same way.

50:59 Play from here If she really likes it, she really likes it, but she really loves using AI

51:03 Play from here now to hasten her, her exploration.

51:07 Play from here But then she's happy to go off on all the rabbit holes from the links

51:10 Play from here that it gives her such that she can get a better feel for this.

51:13 Play from here But when it comes to the commoditized part, like the plane ticket, like

51:17 Play from here we don't really care about car, like she just wants to get the

51:22 Play from here destination and the activities.

51:23 Play from here That's what she cares about.

51:24 Play from here Those are the commoditized parts, whereas the experiential parts

51:28 Play from here will still have a lot of value.

51:30 Play from here And again, that's not going to go away.

51:32 Play from here Those, those, those brands will not go to zero click, but zero click will

51:36 Play from here still be an issue for them because my experience is your commodity.

51:41 Play from here So you and I not wanting to do travel planning might be zero click travelers

51:46 Play from here where our spouses are going to be like high click travelers because they're

51:51 Play from here going to do things differently.

51:53 Play from here There's room for everyone except for an undifferentiated me too type brand.

52:00 Play from here Those, those are all really in trouble.

52:02 Play from here Yeah, I agree.

52:03 Play from here Those are, those are in real big trouble.

52:05 Play from here I know it'll be, it'll be interesting to see where it goes next.

52:08 Play from here Um, how this rollout goes with advertising, if it is going to really look the way

52:13 Play from here that Sam Altman says it's going to look, you know, to be, to be clear, it's not

52:18 Play from here supposed to be part of the chat conversation.

52:20 Play from here That's the narrative right now that this is more akin to a Google display ad that

52:25 Play from here comes up that it's even been questioned.

52:28 Play from here Is it going to be relevant to the conversation you're having, or is it

52:30 Play from here just going to be a bunch of display ads?

52:31 Play from here It's still really early to see where this is all going.

52:35 Play from here I don't know if you've heard more about it, Kevin, but that's, that's, that's,

52:39 Play from here that's kind of the last that I heard.

52:40 Play from here It's a very, very high level at this point.

52:43 Play from here It's very high level and it moves really quickly.

52:45 Play from here So we're advising brands to do unfortunately to cover their bases.

52:49 Play from here So we're not going to get technical here because it's alphabet soup as far as the

52:52 Play from here different protocols and schemas and things like that.

52:55 Play from here But the gist is that you have what the LLM sees out there in a, in a product or an

53:00 Play from here experience, and then you have like a data layer that has to be

53:03 Play from here maintained that is like a honeypot.

53:06 Play from here It's a, we call it a traction point for the LLMs and brands not only

53:11 Play from here have to now optimize their page.

53:13 Play from here And again, this isn't just products.

53:15 Play from here This is whatever you're transacting online.

53:17 Play from here So they've been optimizing their products for a long time, but they

53:20 Play from here have to optimize these data feeds, but they have to optimize their data

53:24 Play from here feeds in a changing landscape.

53:26 Play from here So there are five or six different approaches to this.

53:29 Play from here And the most ambitious brands are trying to accommodate all of them.

53:33 Play from here And at the end of the day,

53:34 Play from here Are you talking more about AEO right now, as opposed to paid?

53:38 Play from here Oh boy.

53:38 Play from here Yeah, that is that, but it relates because you can't, so AEO

53:42 Play from here is answer engine optimization.

53:44 Play from here I, my own like soap boxes that we're, we're confusing people by

53:49 Play from here just generalizing everything is AEO.

53:51 Play from here AEO is getting financed really fast and, and transacting.

53:55 Play from here GEO is generative engine optimization.

53:58 Play from here That's a broader approach that incorporates like your brand, your

54:01 Play from here Reddit, like all kinds of stuff that's going around and they're different

54:05 Play from here tactics and strategies that have to be employed in each, so soap box off,

54:08 Play from here different things, but they all play in to the paid universe in the LLMs in that

54:15 Play from here if you can't transact like zero click, then it's going to be really hard

54:21 Play from here to have an ad layer that works.

54:22 Play from here So you have to have a transaction layer for sure.

54:25 Play from here If in an answer based environment, even if there are like auction bids, so

54:30 Play from here then in advertising you, you sell real estate based on bids and availability,

54:35 Play from here it's all under the hood, but you're basically bidding against other people.

54:39 Play from here If you're bidding for those same people, even so, if you don't have those

54:43 Play from here traction points that your product is carrying, which in Google might be

54:47 Play from here through what's called Google Merchant Center, but if it's open AI, like they

54:52 Play from here don't play with Google Merchant Center.

54:54 Play from here So how do you give the traction points to the LLM such that if five garlic

54:59 Play from here press sellers are all bidding for the same person who really, really, really

55:02 Play from here wants this stupid product, like who gets the nod?

55:05 Play from here My bet is the one that gets the nod is the one that's loaded for bear with the

55:10 Play from here most traction points for the LLM with the lowest points of friction, because

55:14 Play from here why is the LLM going to send you somewhere where you end up in a high

55:17 Play from here friction environment?

55:18 Play from here Yeah.

55:18 Play from here When you say traction points, can you unpack that just a little bit?

55:21 Play from here Yeah.

55:21 Play from here Carefully with this is again, now I've now I know your bot is listening.

55:25 Play from here So it's a, it is this explainer Kevin, I guess.

55:31 Play from here Yeah.

55:32 Play from here Okay.

55:32 Play from here So I think the easiest is, is to think about just like, almost like you're

55:36 Play from here looking through the matrix of like zeros and ones and things.

55:39 Play from here And again, there's just, you have a, you have a shoe, you have a garlic press,

55:43 Play from here you have a garlic press.

55:43 Play from here All right.

55:44 Play from here And the garlic press has basic information that's already

55:47 Play from here available to the internet.

55:48 Play from here And it's already available on Google merchant center and is already available

55:52 Play from here as what you call a product feed that goes to have meta, et cetera.

55:55 Play from here And it's things like material, stainless steel, right hand or left hand, left

56:00 Play from here hand, you know, availability, shipping size, shipping weight, first available.

56:06 Play from here Like, you know, if you were the matrix, you would see all of this stuff.

56:10 Play from here Then as marketers, we put stuff on the page for like features versus benefits

56:15 Play from here of, you know, keep stinky garlic off of your hands by using a garlic press

56:19 Play from here might be a, you know, a benefit versus the future or FAQs and, and all of this

56:25 Play from here stuff helps if it's well-designed into the site in a regimented way.

56:31 Play from here So people want to geek out.

56:32 Play from here You can check out schema.org for example, is one of the standards

56:36 Play from here that people use to do this.

56:38 Play from here And it's like increasingly being built right into the metadata of the site.

56:43 Play from here But where it gets more exotic are in dynamics that LLMs care about

56:48 Play from here that normal shopping engines don't.

56:51 Play from here So things like occasion, what, when is this, what is a good occasion for this?

56:57 Play from here Is it for a birthday or whatever?

56:58 Play from here Sentiment, like what is sentiment is this particular transaction carrying?

57:04 Play from here And here's where garlic press is a bad example.

57:06 Play from here I do own a garlic press.

57:08 Play from here I'm a vampire.

57:09 Play from here Yeah.

57:10 Play from here Yes, I am a vampire.

57:11 Play from here And then relationship.

57:13 Play from here So pretty good example, a wonderful brand that we work with that does gifting

57:18 Play from here and they do get well gifts and the occasion is obvious.

57:23 Play from here It's like get well, the sentiment is like sympathy, sympathy, sympathy, love,

57:28 Play from here whatever, and the relationship is relatively, all that mushy stuff.

57:33 Play from here Well, I don't mean to say it that way, but you know, but this is

57:36 Play from here going to make sense in a second.

57:37 Play from here And then the relationship is like family, close circle, friend, boss, like, et cetera.

57:43 Play from here So, so you've defining that.

57:44 Play from here So now when you go to GPT and you say, Hey, my mom crashed skiing and she's

57:51 Play from here in the hospital getting surgery.

57:53 Play from here What would be a great gift for her?

57:55 Play from here I feel really terrible.

57:56 Play from here Now you have the occasion.

57:59 Play from here Somebody is in surgery.

58:00 Play from here You have the relationship.

58:02 Play from here It's a family member.

58:03 Play from here And you have the sentiment, which is love, sympathy, support, whatever it is.

58:08 Play from here And that's like catnip, or at least theoretically, that's like catnip

58:12 Play from here to draw the attention of the LLM.

58:14 Play from here So how now the trick is to figure out how to do that over 10,000 products.

58:19 Play from here Like, and what is the right sentiment that triggers the right response?

58:25 Play from here Because we still don't completely understand what's happening in that black box.

58:29 Play from here Exactly.

58:30 Play from here We're just taking guesses.

58:31 Play from here And then how does it get syndicated, not just to Gemini, but to open AI and to

58:36 Play from here anthropic and to perplexity and all.

58:39 Play from here of them. So this is some of the stuff we're helping figure out. But those traction points

58:46 Play from here are going to be really important and not enough people are paying attention to them in sort of

58:49 Play from here the mere mortal territory. The big, big, big brands, they're getting this, they have inside

58:55 Play from here information, whatever. But if you're dealing with people who are in the sub-100 million category,

59:00 Play from here there are very few that are paying much attention to this at all. And it's going to be abruptly

59:06 Play from here important, would be my bet. And then GPT says, if your mother doesn't recover from that ski

59:11 Play from here accident, perhaps I can connect you with another older woman on Cougar.com. I hear this did not go

59:18 Play from here well. We need urns. Just a whole pivot in that ad delivery. Just like the whole thing, it goes so

59:26 Play from here horribly. So in the audience, like if you haven't seen these, I think the Claude ads are really

59:31 Play from here interesting as far as catching that concern that people like us have.

59:37 Play from here The potential dystopian future that everybody fears when people are bidding for our attention

59:43 Play from here as we're just trying to find an answer to something, which is kind of how ads work.

59:47 Play from here Oh man, I just got another memo. Kevin says, lowering excitement to maintain trustworthiness.

59:52 Play from here I saw that one just pop up for you. That probably means that we've gone over an hour and it's time

59:56 Play from here to wrap up this episode. Can we build a Kevin bot for our audience? I'm not sure. Will it

01:00:01 Play from here wax on with techno babble for minutes and minutes at a time?

01:00:04 Play from here I would just watch the thinking module all day long. I think it would give me endless

01:00:09 Play from here entertainment. But that is it for this week. Always nice to chat with you, buddy,

01:00:13 Play from here and we'll see you next time. All right.

01:00:19 Play from here Thanks for listening. If you'd like to send us a question to answer on the show,

01:00:22 Play from here email me at elijahatspark6.com. That is S-P-A-R-K, the number six dot com, or Kevin

01:00:30 Play from here at Kevin at ascendlabs.ai. To learn more about AI strategy and training for your business,

01:00:36 Play from here visit Kevin's company, Ascend AI Labs at ascendlabs.ai. And to learn more about design

01:00:43 Play from here and development of custom AI solutions, visit my agency at spark6.com. You can also sign up for my

01:00:50 Play from here free newsletter, AI Framework Friday, by visiting spark6.com forward slash newsletter.

How to cite this episode

Credit the named speaker and AI at Work by AscendAI. Link to the episode or the passage you quoted.

“Ep #13 AI Ads, the Era of Zero-Click, and NotebookLM Upgrades.” AI at Work by AscendAI, 2026-02-10. https://ascendlabs.ai/podcast/ai-ads-the-era-of-zero-click-and-notebooklm-upgrades/

Recent episodes

View all episodes
×
AI Solutions Graphic
Free Resource
Get Started Now