Google Zero, Publisher Licensing and OpenAI's Sponsored Agents

0:01Hello everyone, welcome to our marketing to agents podcast. This is episode 13 and as always we have some exciting news from the industry and we'll dig deeper into some of the paradigm shifting things which are happening in the space. Now with me as always I have Lalit with me joining. Hey Lalit.

0:22Hey everyone. Yes.

0:24Perfect. And today we have two new guests. Vinay who heads the full U devstack CTO of Airmeet vine was with us in the previous episode as well. Vine great to have you here.

0:37Hey Ashwani, thank you.

0:39Hey and today we have a new guest Harsh. Harsh leads AirPulse an AI visibility stack company doing a lot of things. So we will hear a lot of things about what we are seeing in the AI visibility space as well. Hey hush great to have you here. Hey Ashwani, hey everyone. Nice to be here.

1:00Lovely. All right. So with this great set of u panelist we'll start our today's episode. We have a full docket of news and updates. So let's kick kick things off. So to start with everyone we are seeing a lot of news coming out from the overall publisher side. one of the biggest thing is about Google zero. We've heard a lot about it. We have talked a lot about it on this podcast. So we are so publishers as a whole are saying that Google clicks organic clicks are not going to come back and seems that space is going to slowly die down. some of the proofs of that is Google of course overviews AI overviews do not pass on the traffic to the publishers and as a mitigation strategy we are seeing more and more publishers entering into licensing deals and we have great news this week from multiple publishers you talking about how they monetize their data and how they monetize user access. So to kick things off u I'll go to Lith first. Hey, what do you think about? So, let's cover it from first the angle of what's happening with the clicks are is this the fear real that publishers

2:17Are feeling the pinch now that organic clicks of course are not going and in no future in no real future near future they see that coming back. So, let's start from there and then we dig deeper into how publishers are responding to that.

2:32Absolutely. Yeah. So I think u this this this kind of let's say fear has been there is a reason why you know people are discussing about this because it is very evident in all the let's say in search tracking tools whether it's Google analytics or Google search console right you're seeing that the the clicks that you used to get in let's say the pre2022 level right they are they are you know down drastically from that level right and U it's also indicating towards a irreversible change in the user search behavior where people are looking for direct answers and dare I say we are also moving from the answer paradigm to an outcome paradigm where people are actually just delegating a task to be done or an outcome that they need to their agents and then the agents are not giving coming back with an answer but basically you know getting the job done and then just updating that the job is done right so I think users have clearly sort of indicated that they they want more and more delegation more and more you know do it for me kind of a expectation and the publishers which which are still used to let's say expecting visitors to come monetize through ad traffic they are waking up right and u the search engines

3:53Like Google they have this in some sense they are they are in a very tricky situation where they have to balance publisher relationships also and also do what what the users actually want because the moment they dep prioritize any of these parts of this marketplace they will start losing the market share right so as we already know Google was slightly late in terms of getting into their AI mode as mainstream they had they had to tread that path very carefully because they're a public company lot of revenue at stake but you know hats off to them they actually pulled it off really nicely right and I think for publishers in my opinion And licensing is the is is a great way to sort of get their revenue back, right? And we discussed multiple in multiple episodes of the podcast. There are some flat licensing deals available. Then there are let's say you know paper use paper crawl kind of licensing deals available. There are longtail publishers and then there are very publishers. So so it's it's it's a landscape where there's no standard as of today. It's for large publishers I think these these labs are basically getting across the table cracking something very custom but for longtail I think there are newer

5:10And newer players which are emerging we talked about cloudfare quite a lot I think September 15 they sort of you know by default blogged for all ad monetized websites u you know the the bot traffic which is coming for let's say training or inference time visit so that's what I think I I think just adding to something there is a friend of mine who who writes blog on u India's ancient cities. So India although it's an ancient country there are very ancient cities also with ancient temples. So she has a very popular blog at least it used to be very popular in terms of traffic. So she has now at least understood this change and she's now publishing more and more content on YouTube. Right? So earlier it was a textonly blog but then with this dip in revenue she has shifted all her focus to YouTube because YouTube still has actually growing in terms of monetization. Perfect. U so let's before picking up on the licensing piece would love to you know get hers and vine's opinion about guys do you feel the search as we know it is slowly ending like the behavior of search as an intent which was captured and all lot of companies lot of traffic was built on that is that now ending

6:29And what lalit mentioned that now it's not just the answer but it's also the task to be done happening within the LLMs and within in that query itself. So do we feel that search as we know it is ending and the search intent which was so which was so popularly captured and built ad platforms built around that will slowly see u will slowly die out so I have a point of view about how the experience has changed over the past couple of years right so if we look at what happened let's say in 2021 22 23 right like peak covid when let's say we were all cooking at home and stuff like that. There used to be two kinds of search happen. One we used to look for what to cook on Google. Then once we figured out the second search engine for us was YouTube where we used to go and search how to cook and the how used to be explained in a video there. Now when I want to do something like that, I will just look at one place which is instinct and instinct itself will give me the recipe, give me what to order from Swiggy because it has already read my calorie requirements and will also give me the

7:47YouTube video which will take me through it. Right? Those so the search behavior which used to be multi-platform before, right? because I'm clicking on links to understand what is the recipe, what is the video, what is the what I need to read is now just a single chat interface where all the information is given to me in one place.

8:15Yeah, I think the way we all have done such for you know almost in our entire career till now it's already ended. It doesn't mean that you know search is ended in such it just moved the platforms right instead of as hush is saying instead of searching in multiple platforms now it is a single interface and everything is consumed on that single interface it's very exciting to see how these monetizations will happen whoever published in hashes example recipe books right we used to go into that website and see text recipes or video recipes now it is all consumed in the single interface and how how do they get paid in future it's a unknown for now perfect so I think we're all in alignment that the form factor will significantly change the way people were interacting with the systems now explicit search instead of that it's going to be a more holistic experience either provided by meta muse or like hir said something like instinct. So I think that experience is here to stay probably and that will subsume your search and you know different types of search which happen within that. Now

9:34Moving to Lalit what you were you were high oh go ahead hush sorry

9:37Sorry I just wanted to jump and another big change that we should highlight in searches search has become very patient as compared to what it was before right we were used to the fact that we will search something on Google get answers in 5 seconds like Google even had that metric right like 300 million pages rendered in 0.25 in 2 5 seconds. Now it is much more patient where I will write write a query or send a voice note to any of the agent that I'm using and I will just go away from it. So yeah the last shift I would suggest the search is much more patient now as compared to before. Perfect hush completely agreeing that could also be because now search like like I think you and mentioned it's now merging with the tasks as well. So it's not just a search, it's also things to be done and that's and that you're absolutely right takes longer and therefore we are becoming more patient with that whole experience now.

10:38Perfect. So

10:39On that note

10:41Yeah just want to add Google search console just launched something called a multimodel search filter. So earlier it used to be a text search and now they are actually segregating the traffic on image and voice text like the hush mentioned an example if somebody is using search with the image upload or voice they're segregating the traffic in that filter.

11:04Very interesting. Very interesting. Yeah. Would love to I think probably in our subsequent episodes we can deep we can dive deeper into just understanding how that is turning up results for publishers. Okay. touching on the second bigger aspect of this whole experience now. So with search traffic going down, so publishers are now finding out more ways of monetization. Nalith highlighted a few of those which could be memberships, which could be subscriptions, which could be licensing deals now. So coming to licensing deal, we've heard quite a few of these deals or discussions happening. We heard the first deal which OpenAI is cracking in India. We heard Reddit is now actively in talks about active licensing deal. we heard New York Times has a licensing deal on the radar as well. So what's the forum's overall opinion about these licensing deal? Is that would that could that become actually could that become also like an offshoot is could that also become like exclusive deals? Could that give value to certain LLMs and certain chat interfaces? U what do you what do we feel about that? I know Lalit you did highlight a bit about it. So I'll let Vine and then Harsh go and just share their opinions first about it.

12:19Go ahead H.

12:24I'm so sorry. So I have like a bit of a mixed opinions about licensing coming over here. Yes, there's a good part that you know New York Times is going up with it and there are some publishers like India today like in India would also be doing that but a large part of licensing also relies on what kind of IP are they trying to do. If you look at New York Times, right, it is a very high and thorough journalistic platform that writes big opeds and everything that can feed into the training of a model like JP 7 whatever, right? But I don't want my model to be trained on top 10 places to visit near near X location which most of the news publications already do. So it will be very interesting to see how these licensing deals pan out and what mod what content do these LLMs actually end up consuming. Like a perfect example of a licensing deal that almost happened and sorry a perfect example

13:45Of a licensing deal that almost happened and didn't go through is Disney and Sora. So Disney was going to license characters for OpenAI sort of and back in 2023 and at the end of the day it cancelled because OpenAI had certain specific things that it wanted and Disney was like no no you can't do that. So

14:12Yeah have much on this cont. Yeah, I think see the big significance for the licensing deal is because there is always new information, new data that the you know are as a civilization we keep you know discovering there's new research that is happening, new events in real life that are happening right and so and as our as consumers our reliance on our agents and chatbots continue to increase, right? we would expect them to have access to real-time information, up-to-date data and you know we don't want to be misinformed about an answer that we are so you know let's say [clears throat] in eagerly seeking right so the deals actually allow them allow the chatbots and agents to have realtime access right which we call as inference time access in our internal discussions right where someone is where the agent is actually searching the the web and then the publishers's web page is encountered And then to actually access the content and make it part of the answer or the workflow or the outcome they have to pay something right or or they should have the right to consume that. So that is

15:29What these licensing deals enable. And my very strong opinion is that this will open up a very large industry right of let's say niche data niche information because one of the things which not no not many people highlight about AI is that AI is a fantastic understander of sorts right understander of sort as in like it would understand each and every individual so deeply that we can potentially see a billion marketplaces right thrive because AI will do the job of let's say discovering the right buyer for a product which was earlier so esoteric that the creator of that product would not have thought of marketing it right or let's say a piece of content let's say a genre of comics right which or content which which someone is creating just out of interest or just out of let's say their passion but then AI will help them find the right you know consumers because AI knows each and every human that is interacting with them so deeply after especially after let's say we starting start we start to rely too much on our agents and then we open our worlds and information to them and they will then help us discover what we

16:47Like the best so I am very hopeful and very bullish on the a million marketplaces that it'll open up so it'll basically create niches like you know different genres of movies, different genre of let's say story writing. you know few episodes ago we talked about data marketplaces also where there is so much data and information which is unscattered like which is scattered over the internet and in real life and it a glimpse of that data only comes through a journalist doing primary research and you know writing an article about it but the primary data still remains either on the desktop or you know in those scattered sources. So there will be new companies which will sort of you know bring up data and under a licensing or a real-time licensing arrangement with these AI bots they will be able to monetize all of their effort. So very bullish about it.

17:39So that's where my one of the scenarios came in. So let's take a real life example. So let's say New York Times they enter into an exclusive licensing deal with Google Gemini. Okay. Now all of the New York Times like you said realtime inference that's happening only on Gemini now for New York Times. so for example let's say if I run a search query in or I chat and use chat GPD or claude I they would not have access to the New York Times news piece or that news article. Right? So does that also mean if these deals become more are more common then there will be platforms who will be good at something and some platforms will be good at something else. So for example maybe Gemini goes ahead and has ties up deals with all the news publishers whereas let's say claude goes ahead and ties deal with all the marketplaces data provider let's say autoarketplace or something else. So that means then each of these LLMs based on the publisher base that they have access to or the unique data that they have access to their they automatically then you know sort the persona who use them. So quick answer to that is I think this is just a recent or maybe a short-lived

18:57Phenomena that we look at multihoming is unavoidable in it right what I what I meant by multihoming is that as users we are not stuck with one chatbot we use all all these AI you know all all AI agents and all these you know AI providers quite like that publishers will also you know let's say go ahead and you know sign up with multiple you know AI engines And if someone really wants an exclusive right to information, then they would be paying a very big premium, right? And why would they be doing it? Because they want to maybe highlight it on their marketing. They want to capture demand for some consumers. So very easy way to understand this would be let's say sports event like you know broadcast rights. So a digital platform like a Netflix or a geostar right? they would go and bid much higher to get exclusive rights so that their user acquisition can jump up right so look look at it this from like a market capture dynamics and not a steady state equilibrium steady state I think publishers would want to have you know relationship with almost every consumption platform every consumption let's say AI engine but they might if if let's say there is a niche publisher and then there is a niche AI agent then those those matches would be very natural

20:16I also feel that publishers will require to you know have the relationships with all the agents because I think just if you understand the AI as a simple topic right at the end of the day it's reasoning on steroids now you cannot have reasoning on steroids if you don't have both both parts of the same you know like same statement right you cannot have a p open AI talking just with the right wing you know publishers and the Google talking just with the leftwing publishers. This will make AI an open an echo chamber and I don't think any of the AI developers actually want it to become one. Makes sense. Makes sense team. So perfect. So I think it'll be a interesting u interesting move to see how publishers go and monetize their content and like of course u build more avenues for their own growth. Okay, moving on. we had some interesting news from OpenAI. So, we did know that OpenAI launched ads few weeks back and ads are doing pretty well.

21:34They already hit 1 billion AR on the ads alone. But the update that we got recently was sponsored agents. So how sponsored agent works is so if you if you're seeing an ad within open AAI and then you click the ad you will get to an interface where you can chat with that agent of the sponsored ad and then it could be the business behind it and they could be you know you can they can answer queries about your your queries about the product about something that you want to know from that business. So it's an interesting way to then build an experience around monetization around ads and gives u the business also or a publisher who's advertising gives all them also an opportunity to know their user a lot better. So just wanted to you know hear the thoughts of the panel on that what are we seeing with that this evolving into are there more manifestations of this experience across the web that we will see it nurturing as well. Perfect. So, Lenin, do you want to go first?

22:38Yeah, I [clears throat] think see I would say this is a this is still an experimental UX, right? but what we have seen as a search behavior is that when people are looking for options, right? They they tend to, you know, write more let's say investigative queries on the options that were recommended earlier, right? So for example, if someone is looking for you know let's say u B2B webinar solutions right so if Airmeet is amongst the options then our own data tells us that you know the subsequent queries are about knowing more about the you know about air as an option right maybe comparison queries or pricing plans etc and u giving the advertiser an ability to sort of you know be present in that conversation directly not indirectly via let's say third party websites or third party publishers who whose content might be outdated, right? so I think this is a great opportunity for advertisers to sort of you know get into the conversation which is high intent and I'm pretty sure it will improve the conversions whoever gets into that chat experience. It also at least is alluding to the larger direction in which chat

23:55GPT might actually support businesses, right? Imagine every advertiser for them to support a chatbot experience, right? OpenAI has to know lot about them as accurately as possible. Right? So I would not be surprised if every business starts maintaining their knowledge base on chat GPT or you know a specific set of facts which which in a way are exclusive or maybe some you know specialized access to to their web MCP or back end MCP. So I would say it's a good direction from a user experience point of view and also advertisers should get a you know like a jump in conversions. Perfect. I completely agree. I think it it does add a new layer of for especially for the advertisers to be able to engage with the users and of course in a much more interactive manner too. Vinn where do you stand on this like yeah

24:56Yeah more than the opinion I have actually very open question maybe Lalit can you know answer so much is happening in this you know space right where traffic is coming from and like where the searches are happening. So in this aspect like are businesses ready to measure the traffic or attribute the traffic to certain you know source in this case the chatbot for example I mean discussing in terms of technical readiness is same chart bot which everybody has on the website is is it going to be same on the chart GPT brand like you know ads sectors or earlier we used to have a direct traffic attribution and all right. So how do we attribute that to this searches like I know everything about say airmit brand on chart GPT and then I Google airmit directly to go into you know website and all. So but we are still attributing that as a direct traffic lot of search already a lot of filtering already happened before. So as a business like you know where are we in that aspect like

26:12Can I yeah go ahead please. So if we look at the press release by OpenAI very closely, the only merchants that it has enabled specifically in the US is Shopify, right? You cannot have your own. so it will of course go ahead with people like Hilton and and and Marriott in the future, but Shopify merchants and the same is happening with news if we we'll dive into that further. But the reason that specific niche is getting targeted with all these systems is because of the protocol UC called UCP universal commerce protocol and the traffic will mostly be attributed to that protocol. All the data is being trans transferred through the UCP protocol from the conversations through the payments and this is how it is impacting most of the attribution for the end user over here. UCP. Yeah, for for short conversions, I think whenever if conversions are actually happening then definitely UCP. But I think when I had a larger questions around like u the challenges

27:31Around attribution where people spending too much time on an AI answering experience and then directly landing on the website through a Google search and businesses think of it as a direct direct conversion lead u versus u you know and not like an AI AI influence lead. So I think when I see this is this has always been a challenge since social media is advent. I think as more and more information sources become mainstream, I think attribution as a as a problem statement for for marketing teams will become even more challenging. Right? So I don't have a very specific opinion in it. see for for it to be impeccable attribution I think all these desperate let's say surface areas have to be sharing something which is common right earlier there was browser cookies and things like that which also is now very difficult right so attribution as a challenge is going to be big but but the good thing is that it is it's it's an internal problem right which is which is not like an urgent thing businesses can continue to let's say work with the advertiser provided

28:47Metrics and maybe have their own internal model to sort of over time figure out what is the right attribution but it's an it's an opportunity someone should build up you know like a dedicated AI model for this

28:59Yeah and attribution absolutely is going to become even more difficult to track so just taking a cue from what happened in in the past so of course browser cookies were there but when mobile apps came and then everything a lot of things were happening on mobile labs for Sometime the attribution was not solved but then the platforms took the ownership to solve attribution because that's only then marketing dollars can move into it only then a marketer will be comfortable to spend a million dollars to drive more traffic to their app and then eventually understand where the user is coming from. So I think if a platform so couple of things can happen here which I feel if the platform itself becomes big enough they will solve attribution at a platform level if not if there are multiple platforms then probably they'll all agree on a on a standard which all of them can adhere to because till attribution gets solved the marketing dollars may not flow so easily because they need to see some ROI there. So yeah, I think it will be a problem as we see more of this experience propagating, we'll have a better solutions come up here for the attribution piece. Perfect. great discussion team. Just continuing with our docket now.

30:16We had good news, interesting news from the AI visibility space as we've been discussing every month and great we have Harsh as well with us. So we heard we got some great updates on AI visibility. we are getting more and more closer to you know now AI visibility platforms are integrating u search console data as well and able to provide a unified picture to to brands and as to how their visibility is going and then tying that with the Google data as well. So I'll let Harsh why don't we hear something from you about you know what are the latest things happening in the AI visibility space and how does a brand a small publisher a big publisher how could they benefit from some of these updates?

31:04Sure. So I guess one of the biggest updates that we saw over the last few days is the decrease in the use of chat GPT in prompts. I guess com score has this data and increase in increase in other apps that people are using for search, right? So people used to like if if previously you were a marketer who were optimizing your platform for for AI visibility. you were just focusing on chart GPT and you were focusing on how can you make your website more comfortable for that and but now because let's say even even this week Muse is the most downloaded AI app as compared to Chart GPT in the United States. So people are now understanding that users are using multiple platforms for AI search not just charg. So you need to optimize for all the agents which is Claude, Muse, Instinct, Google AI overviews, Gemini, Perplexity, fast search is also available now. And because of these things, companies like

32:22Profound have now released an agent ready mode for the websites as well where people can convert their website into a markdown file or a txt file which is much easier for an agent to navigate through. But Ral, do you think that or V you can answer that an agent which is very heavily trained on one command which is grip which is basically finding a bunch of text in a haststack of text require a new interface or a new language with simpler text to actually search the answer that it is looking for. when would you like to jump in? Yeah.

33:19Trying to understand maybe can go ahead. can you repeat the question? like

33:27So so all the AI agents if you look at their training data specifically are trained on Linux commands which are grabb commands find commands and all that and those commands are very good in finding a bunch of text in an entire clutter of text right so let's say my name is h you have to find it it would be able to very easily find it across a 5,000line HTML file right do We still need a markdown file which is specifically telling the agent my name is Harsh. Like all these new services from profound or scrunch or even we are doing research on top of that suggest that every website have should have two two methods of delivery. One for human which is the HTML delivery and another for agents which is the markdown or the txt delivery. Is that actually required specifically when the website is very easily rendered on the front end for the end user?

34:33Yeah, I think it is needed at least at the moment, right? the way humans consume information or multimodel like you know they watch the YouTube video which is on that page or text or images. it is a better way to serve as a MD file for a agent visiting the website. even before profound and all cloudflare has this switch if bot is visiting it serves with a certain http header text/ this thing as I forgot the header name but it will serve the parallel MD file to the visitor if it is a like you can request special MD file I think it is optimal to serve a parallel system to the agent visiting the website. So my take is that it's all about the compute budget right u for for example just if let's say when you we are dealing with let's say any other human right imagine a team member or a service provider right so you would prefer a service provider which which makes you do less work right let's say you are working with a HR consultant who is hiring for an APM role

35:50Right and if The HR consultant makes you do a lot of job in terms of make you know even after a phone call if it is making you to write a job description and then you know making you to filter the profiles which they should have filtered out right so you would obviously prefer it lesser and lesser right so same thing compute budget token budgets these are real concerns and you know at scale I think these things will start to matter and that's what makes me bullish personally for agent to agent you know economy where today let's say today since the agent economy is at the very very nent stage we are all talking about our personal instincts right but imagine let's say hush is looking for a holiday over the long weekend and I am let's say someone who is who can provide a holiday service right holiday planning service right so can my instinct or muse be exposed as let's say a service provider right and harsh can discover that and rather than his agent doing the hard work of let's say you know searching reasoning and then figuring out whether this is good for let's say hush or not can it not let's say give an RF RS RFP kind of a thing where my agent

37:09Can get to know and you know much of the let's say token and compute utilization could be saved because I have those packages ready made and I can give that to the to hers's agent or anybody else who is looking for it and it can be just provided. So I think larger point I'm trying to say is that since we are very early so it appears that these are all things doable. but like all intelligent entities our agents will also start to prefer the least path of least resistance.

37:45Yes. On that note, I think instinct recently announced that you can your instinct app can talk to another's instinct app. They have released a protocol like some trusted network or something like that where you can add certain

38:01That's mostly for that's mostly for family members trusted trusted person.

38:06Yeah.

38:07I use it to set up dates with my girlfriend. Just kidding. Great. Fantastic use cases. We love to hear someday we'll love to hear more about that. I'm sure we'll have big audience asking about that. Perfect. team this has been a great episode. I think we've covered most of the items that we had in our docket. would love to you know open it up for the panel to you know just any thoughts any shares anything very interesting that you've seen come across your desk over the last week that you think can benefit our listeners open for everyone to jump in I have a question though like in this panel like you know we use multiple charts in our day-to-day and is there Any preference that which chatbot do you use for certain types of queries or is it random or does it depend on the form factor you are using? On mobile you go to one chart but if you are on laptop how how do you actually segregate?

39:21I I'll just I'll quickly go. Yeah. So I have u so I I have a cloud subscription. I have a perplexity subscription. I have obviously chart GPT and then instinct very soon muse and grogbot right so almost all possible popular ones I'm using so I generally since instinct is free right earlier a lot of my let's say research workload used to go to perplexity right so I have this habit of let's say you know for anything that I want to understand either I get a notebook element made or I ask perplexity to basically create a detailed presentation for it. Right now, all of that is going to instinct because in instinct is actually free, right? So, I'm not really paying for tokens there. So, I'm making instinct do a lot of heavy heavy duty research for me, right? Saving tokens. but in general, for more let's say I prefer chat GPT for GPT live experience. The voice chat is actually far ahead of anything else, right? My Gemini usage has actually come down quite a lot mostly because of the the GPT live experience Gemini still because of the chrome integration I see a ask

40:38Gemini right there on the chrome so I end up chatting with u Gemini as well in in the voice mode when let's say there is a let's say a web page which is quite long and then I have to you know interactively get to know more about things and then for charge charg also has a lot of these a bouquet of tools which are generally useful for getting the job done like an image generator and also you know by default a lot of connectors that I regularly use. So a lot of my let's say getting stuff done also happens to charge GPT cloud mostly and very exclusively for only programming and chargex for programming. Perplexity has actually usage has come down but still the perplexity computer and especially the orchestrator pattern that they have where they can work across all these models and they are really good in terms of judging what to use for what kind of a task I for many of my sort of detailed planning exercises right for example if I want to do a competitive analysis on a on an industry right or or I know that there are extensive you know different websites where information

41:54Will be there right u extensive jobs I go to perplexity and then the in the computer mode it actually surprises me with the quality of the output

42:07Okay harsh you

42:12I am quite I mean heterogeneous in this. So I will be I will end up using a lot. So yeah like I'm researching on charge GPT because you don't end up consuming tokens on charge GPT. So that is the reason I use chat GPT web for research. Then work happens a lot in codeex and cloud code. But because I have taken the cursor ultra subscription for grogbot I end up using cursor crowd agents to you know like I'm going to sleep but this is an idea. So I will just use that to run a crowd agent and solve these things. one interesting thing that I use is called town. So town helps me specifically deals with my emails. It is very good in email labeling. It is very good in telling me how the how how is my schedule and everything. I just like the interface town has with emails as compared to anything else. Grogbot has everything connected to my Google apps. So anything on Google

43:30Sheets, anything on Google Docs is through Grogbot and yeah like instinct for every personal thing from tracking my calories. So I track my calories on Instinct and then in at the end of the day it gives me a dashboard of hey you ate 200 calories more than your daily requirements. So stuff like that. Fantastic. what about you? What's your stack looking like? Yeah, I think u more or less majorly work related stuff happens on chart GPT and cloud code but chartpot experience I have this perception of you know some personal related queries happens on Google and perplexity work related chatbot interactions happen on this is like a two different accounts also like you know my Gmail personal versus workreated so that segregation is there and somehow that developed in my mind like you know for example perplexity I have very specific use case like if I have to start some TV series season 4 recap of all the three seasons perplexity does very well

44:46Like basically summarization of things right I whenever there is a summarization my perplexity comes to mind as a first choice and surprisingly yesterday I wanted to create a small meme and charg gave up saying it might be a copyright It was actually very popular meme and enough charted two three times. So it actually rejected generating with my own text on the famous template then went into another model another bot and happily got the output. So yeah I mean these habits are developing like segregations are developing right now. It is a bit of cognitive load. sometimes I query and charge then immediately close the tab chest will answer better for this and without looking at the answer of charge I move to what about you yeah very similar mine is is very simple straightforward so charge GBD for everything personal offwork thing and claude for everything work so I've tried to keep it very limited all my projects I've built there only so that at least it's focused it has the context and then all those things stay there. I've used a little bit of started using instinct but yeah muse

46:05Is the new thing that u got access over the weekend so now now trying to use that but still not a lot of workflows have not moved to muse yet but yeah I want to start doing a lot more with that.

46:18Got it. Perfect. My personal opinion is that perplexity is going to be a dark horse and this the the orchestrator pattern right and where you are not tied to one model and you can you can sort of route the work to any model which is cost efficient for the job that I think is a is an advantage which I think the the the bundled you know experience that you get with say any lab provider right the products that GPT ships obviously they have the frontier and maybe their harness is very much let's say like perfectly suited for their model and they might they may they may not expose the quality output when let's say someone is accessing over API but in general I think as models become let's say that layer commoditizes and there are freer and maybe you know cheaper open source models which are equally good for task which we right now currently pay quite a lot on right over time I think perplexity will have a very strong reason to win in generic knowledge work use case right so we will not be surprised if all of us are you know using perplexity for knowledge work

47:35Just like we use cowwork cloud cow work or charg work we will rather you know switch to perplexity and then for only specialized use cases we use either charg like or include. Yeah, that's my personal take.

47:55Fantastic. Yep. I think in some ways in some ways I think that's where things will eventually migrate to. but yeah, team another u great session, great episode. Love the insights, love the thoughts everyone shared. and for

48:14I have one one other thing, one other thing before we wrap up. I think Amazon has blocked Instinct and Muse

48:20From scraping the website. What's what's the opinion here?

48:24Harsh would like to share.

48:28Yeah, I mean Amazon does not want to adhere to UCP. Amazon is one of the few people who have said that they would not want that happening. But I'm unsure what is happening like what Amazon is up to. I'm pretty sure they I have had enough friends in Amazon to tell me that the internal models are really bad. They are unable to code. They are unable to work with it. we have seen glimpse of it. The Q model on AWS back end. we have used roofers on the Amazon app. So I'm not sure what Amazon is up to right now.

49:12Yeah. Ai, what do you think being in the US? geography. I think well there are from a business point of view I am there are few things which are happening I think Amazon now number one is number one is of course for Amazon security checking every transaction validation of the transaction is very important returns is a big big important thing for them so I think first of all at the onset I think that may be there and so that they don't want to just open up the platform for complete like full commerce as it is because the transaction volume and then because it's not just one supplier it's a platform so there is suppliers who they are responsible for so they don't want that their whole machinery which they've built over the years just getting without getting tested in a proper sandbox environment it goes away so number one I would think that's the reason why they don't want to open it up immediately first maybe dog food it to a on a smaller scale and then maybe open it up number two is u again I think we've talked touched upon this. So it has a direct business implication as well for Amazon. So when people go to Amazon discover products so for example a prime user who's going to Amazon they just don't shop one product which they've come for there's a lot of

50:30Products they buy along with that. So when the search when the shopping experience significantly changes to a specific intent, number one is ACVs will go down, their average order values will go down, their ad monetization goes down significantly, a newer brand discovery significantly decreases. So there are many things which happen. So I think Amazon again of course Amazon is usually is very ahead in some of these things but in this case maybe they just want to understand the landscape a bit more try it out a bit more maybe in a smaller geography in a smaller segment before opening it up for a wider across the board audience that's my thought why Amazon wants to be slightly more tepid in in launching this aggressively

51:17Yeah V any thoughts Yeah, on same lines as Ashwani said, I think they have put a blanket ban on you know muse using Amazon and all just to figure out things and pretty soon I think they will open up they will dictate the terms for these agents I feel. Yeah, I think mostly the revenue loss, right? Ad revenue loss and you know the average order value also reduces because their recommendation engine which they are very proud of right generally ends up you know at least bumping up the basket size by at least 20%. Right? So but I think this is inevitable right? So if either Amazon quickly acquires an instinct or any other platform and you know becomes like you know one of one of the alternative for commerce or or a personal assistant or they will they will become obsolete right because this is like as already a lot of merchants are on Shopify the direct share of D2C like the transaction share of D2C on the direct website is anyway jumping up right so I UCP is UCP in my opinion the universal commerce protocol I think this is a a snowball right as it be gets wider adoption

52:34It will become inevitable like you know unbeatable so Amazon should should as we discussed earlier also right Amazon is and all these aggregator platforms are at a loss because the direct merchant can have a direct relationship through the AI right so rightly so Amazon has now opens open their entire logistics as a AWS kind of an offering so that they still are they still have a fair share of the value chain. yeah but let's see how things happen but interesting interesting times to be seeing how things are shaping up. I would like to add a clarification that Amazon has joined UCP now along with Meta. And if we go to the news properly, yes, Amazon has a blanket ban on AI agents going to the agent to the through the website because it violates their terms of use. But specifically to Muse, they are like you did not tell us that they will be they will be doing this. we don't know how to handle this kind of traffic and yet maintain a good ACV. So I think they are just figuring out how to open it for a small audience, test it out and then make it for a larger audience because they definitely

53:52Don't want all the AI agents everywhere doing what you know Muse did on Facebook marketplace which we can take up probably in the next episode but they don't want that happening on Amazon right now. Absolutely. Yeah. And that's a wrap.

54:15Perfect. Thank you so much, Harsh for joining in. would love to see you more often. Vin Lit, always a pleasure to have you with us and sharing your insights. Thank you so much everyone for being here and listeners would love to thank you. Yeah. And listeners, we would love to hear your thoughts on this episode and and please do send in your questions if you have any. Thank you. And don't forget to subscribe. If you are on YouTube, the button is over here. And if you are on Instagram, the button over for follow is over here. Thanks, Ash.