AI Search Becomes the Top B2B Buying Signal, per HubSpot Data

0:01Hey everyone, welcome to our next episode of marketing to agents. and I have with me Lalit as always. Hey Lit, great to have you here.

0:14Great to great to be here Ashwani. Yeah, always a pleasure.

0:17Fantastic. Lovely. So Lith, a lot of things happened in the market in the whole agent space in the industry over the last 5 to seven days. So today we have a full set of great updates that are coming and would love to hear your thoughts. Would love to hear would love to discuss about what some of those things are and how are they impacting the industry as we are looking at. Okay to start with then the first key news piece that we came across which you did cover a little bit in our previous episode was that Google has launched an always on information agents in AI mode. So what that would mean is that you can keep me updated mode on your prompts. So the agents are running 24/7. They're scanning the news, social media, finance, shopping data, and they can keep updating you, synthesizing reports, synthesizing results, and keeping you updated on those on those agents, on those prompts. So would love to hear your thoughts on it. How does that impact commerce? How does that impact B2B sales? How does that impact the whole marketing piece overall? So would love to hear what your thoughts are around this new development.

1:31Absolutely. I think this is the agentic evolution of a very famous feature that Google always had that was Google News alerts, right? Right. And I do remember you know back in the 2010s Google News alert was very important for us where we would you know just maybe whatever was our concern for example let's say property in in Bangalore or you know some region right you would basically set a prop Google news alert and whenever the Google crawler comes across new information or which matches or or rather indexes for that search keyword it would basically trigger a news trigger an email alert and not just for that one link it could be a bunch of links also and that was a fantastic way in fact one of the best ways to keep on you know keep one up to date about how let's say a certain domain or a certain topic is moving now what what we what we just witnessed Google release is the agentic evolution of that and it it can be much more evolved than just a pro just a set of keywords you could be very creative about what you want to know about and the for for example it could be something that you know u any advancements in using AI agents to improve codebased security right that is a very evolved prompt right and then the

2:51Google AI mode always on agent would then make the judgment of any new information that they come across which sort of fits this fits as an answer for this prompt and it would alert you and it it can it could be on the app with a notification it could also be with an email you could also maybe set frequency of let's say get a get a daily briefing, get a real-time briefing. So I think this is fantastic for especially the research that organizations do because in in our organization we have had lot of interns who you know who keep doing let's say competitive research. I think this is a great way to put that on autopilot. So this would really drive lot of lot of efficiencies. The reaction time to information for organizations will be drastically shortened. Right. That's way and I think u one more thing which impacts with changing of this is that now because the agents are running the prompts 24/7 [clears throat] anything which needs constant updating it could be like you said it could be like in the wild time it was the alerts or any kind of a research I think that really becomes super good and information updating. So for a brand how do they use this is that of course it

4:12Does not now it becomes lesser important that okay how frequently you've updated your site because the agents are checking all the time so and the if there are prompts which are open they will keep checking it so if you're keeping your information updated or you are pushing updates regular updates if there are price changes which are happening on your site I think getting those updates very quick and very relevant would be very very crucial because I think those will then get picked up in the prompt and get summarized in those relevant answers. So it's not like you know that you're losing the opportunity. I think it provides more opportunities to you know keep sending your updated content more often changes because then those can still reflect in better Google answers or better prompt answers.

4:55Absolutely. And just imagine this from a shopper's point of view let's say if someone is interested in a Dyson air purifier right and it's it's very pricey. So you want to you want to basically you know buy it when the price is in the in a certain range right you could now then just let Google know and it would realtime alert you whenever there is a price change so I think this the the whole the intelligence layer right it's this is one of the most one of the most I would say v the lowest common denominator right people people keep keep people keep searching for information but in a way they want to be updated automatically right so this is like the the widest surface area which Google has innovated on to to bring in the agentic benefit.

5:43Absolutely. Absolutely. Yep. And then so then to summarize like what would you say is how would a brand other than like regular updates providing the right information are there more things which you think brands can do to make this more to make the most of this opportunity or this new paradigm of always on prompt searches other than updating contents. Are there things that brands can do more? I think see it opens up newer opportunities for brands to get discovered because you know many people would would rather you know set alerts for not just a specific category but could could also maybe start let's say they would be a lot more keen on let's say improving an aspect of their life let's take a B2C example let's say someone is on a weight loss journey right and they would want to basically look at latest advancements in weight loss using let's say ketogenic diet or or let's say you know what are the implications of vegan diet on someone's vitals right so that is a very high level concern that people I can very very well imagine you know that this being such a important

7:01Information category that people would want to set a real-time alert and they might also specify that only get me you know trusted sources or verified sources right And so this is this is let's say a few steps above the actual category query where a brand would want to get you know a visible in right so they're competing for a higher order let's say information query [clears throat] and if so marketers and brand owners right they it is a dimension for them to invest in real good research real fundamental information you know let's say extraction through research or through let's say expert collaboration and then publish it. So this would open up new let's say avenues for discovery and trust building.

7:46Perfect agree. Yep. Okay. Now let's move on to the next item. there's an interesting update that happened on the agentic stack and I know we touched briefly upon it and of course there was a lot of chatter about it a few months back. so this time what we've heard is that Visa and Mastercard have built out or shipped out a full agentic payment stack. So what this would mean is that they made their systems or they're making their systems ready for accepting agents as a key persona or as a key user of internet and if there are payment gateways or there are payment processes that should be able to manage how agents can interact with those. So what again this is a fundamentally new fundamentally important step because any commerce any transactions or anything which happens gets today gets stopped at the end of a processing the payments or something else and now we are seeing that okay now the stacks are also getting ready for accepting agents driven payments or agent-driven usage of the payment processes. what so first of all like any thoughts around that and then I would dive into another thing that we had

9:03Heard a few months back and then just see if those things relate to this as well.

9:09Yep. I think u some developments have have been happening in the in the way people expect agents to make payment. but I think what we have seen recently with Visa and Mastercard right I think they have they have addressed the key risks at least they've attempted to address the key risks around unauthorized payments or fraudulent payments right so I think they have touched upon the aspect of coming up with an agent directory or an agent identity right and also attaching score to those agents when whenever they are let's say attempt to make a payment on the network Right? So agents with high trust score would would basically be more successful in getting the payments through. So I think this was like the fundamental enabler and developments around anything related to trust in payments is like a fundamental enabler for the larger commerce to then get unleashed. Right? So this is I this is how I see this and I would still wait you know wait out to see a few iterations for this you know let's say stack and the the protocol development and my projection is that in another 6 to 9 months is when I think the the agent commerce will be full-fledged for common people to to sort start

10:28Experiencing it. you're on mute.

10:37Sorry. Yeah. So, one of the reasons one of the use cases which I was thinking is that for example, let's say if there is an agent and I create a prompt that hey I want to buy sports shoes and then I just want the agent to go ahead not just do the research but eventually even go ahead and do the transactions. Let's say they give me an option I say okay this one I like just go ahead and buy it for me. Now for that to happen for that full transaction to happen completely through the agents I think at and these could be like small retailers it could be a small brand D2C brand which who's which had the which finally gets my sale so I go so my agent goes to their side and then actually tries to do a transaction and all of those things which are needed for the transaction to happen so it's interesting that yes the infrastructure required will be quite different and the fact that websites might of course prioritize or at least understand try to build systems that can help this journey to be fulfilled could be very interesting to see and and could give brands another opportunity to be able to you know show something that they've done better.

11:52Absolutely. So shi one of the things which which actually happened very recently like just a couple of days ago and our brief does not cover that that is agented resource discovery ARD so ARD is is basically a new standard which has been shared by the Google cloud team the team behind deep mind as well and there there is a there's a good set of let's say industry leaders who are who are listed as collaborators on that standard. What this does is it first of all AR stands for agentic resource discovery and what this enables is that agents can in real time discover new resources. So today when someone is creating an agent right they are they are basically hard coding or predetermining what are the skills that it has access to what are the resources it has access to like MCP servers or tools it is all baked in in the in the let's say the construct of the agent at the development time and when it runs the agent cannot make any real-time decision of including any new service or any new skill. So that has limitations for sure right as the as the as the as the world is evolving especially on the long tail right imagine let's say someone is trying to plan a trip right and the longtail would

13:10Consist of let's say last mile activities that you can do at a destination right [snorts] now a developer who is building this travel concage agent may not have the information of let's say the last mile but independently just like Google maps flourished in every nook and corner of the world I think small small let's say agent agent supported websites or agentic websites where you know an agent can interact with a service will pop up organically. So the agent has to a travel concier agent agent should be able to realtime discover [snorts] a service and then interact with it and also include payments right that that had to be enabled. So ARD actually opens the gate for that and and it it has got a very beautiful architecture which sort of resembles like a DNS discovery right so a website can host its own set of let's say capabilities which they call as agent catalog AI catalog like all the services or all the APIs that a certain service has then there are registries there could be like so the discovery is federated through a set of registries there could be official big registries They could also be distributed registries locally maintained by let's say some associations or independent people and an agent running in let's say in US can

14:27Discover large mile activities in let's say Philippines in a small city in Philippines through this ARD protocol and standard. So that I think is the ultimate unlock right it it includes trust as well because the agent catalog is published on a website it is not some third party where the catalog is published because third party could be hacked they could be impersonators so the trust of the domain right you know for example let's say if there's a there's a really good resort chain in in Kerala right so they their domain is the is the guaranter that the catalog which is hosted on the domain is curated by their team right so no fake information no fake queries right and then local registries would make it in make make the discovery enabled to travel concage of let's say people in the US who are searching for let's say last mile activities in Kerala so this is fantastic I would I'm extremely confident and all of these are as I said right they are enablers right so they're all like groundwork foundational layer stuff that is happening in the in the world of AI which would ultimately unleash high trust discovery and high trust transactions.

15:44Fantastic use case example. I think that helps clear up the whole process very simply. So yeah, I think there's a lot of new things which will new use cases which will come up with all of this development now and looking forward to some of that. Okay. now with moving forward to our next item and I think next couple of two next couple of items are very very specific good developments for AI search that we've been talking about. So the first I wanted to cover is there's some data coming out of HubSpot u which says that AI search is now number one predictor of CRM purchase intent. So even much higher than demos, sales demos which happen or when a buyer goes and takes a demo of the product or they attend any of the you know training sessions or meetings with the with salespeople or product people of that product. More than that there are research done on AI on LLMs is number one predictor of CRM purchase intent. So that means what we are saying CRM is a core category. CRM is the most comp we

17:02Can say one of the most complicated categories as well because there are so many bells and whistles to it that you have to decide for your organization. Other products are typically you know one one use case products which are much simpler. So that means what this means is that for B2B SAS AI search actually has displaced product demos instead as the top intent qualification mechanism and you know getting the discovery budgets getting more and more ensuring that getting more qualification leads but at the same time just justifying buyer intent. So this seems to be a massive shift in how buyers are now deciding on which SAS software to procure and I think this will of course this is of course seeing going across categories as well but this is the first big data point which has come across for specifically B2B SAS now this having done this it seems that it change it has many ramifications number one of course is on the marketing budgets at the same times on how now the sales pipelines are going to be built at the same time now what does the attribution models look like now so I think there are ramifications across the full funnel of a sales journey a marketing funnel so so yeah very

18:21Interesting development any thoughts around this and and what are your views on this yeah

18:27Yeah so so Ashwani I wouldn't be wrong if I say that you know B2B GTM was was a very expensive human-led way to educate the prospect Right. So all of B2B GTM boiled down to just educating the prospects and what we earlier had was a human-led way to you know sort of engage in video calls or let's say inerson calls events and then then you know after the buyer is educated they would then be you know let's say then they would have the intent to buy right now what AI search and AI agents have done is that they have basically put the buyer education on autopilot for the for the prospect itself, right? The prospect themselves can basically spend as much time as they need on their [snorts] search agent of their choice, right? and can get educated and and hence when they are appearing in a sales demo or when they appear on the on the vendor website generating a lead or a high intent let's say signal that that buyer has then you know has has sort of traversed the entire education journey on its own you know on their own pace right they come at the door or or rather they enter the store fully aware fully sure about that they that this is the right choice for them right And hence I think we are seeing the the data also

19:45Which says that the the that these leads actually convert at four times you know four times better. So this is no surprise to me and and you're absolutely right. I think the B2B marketing the whole B2B marketing which was essentially buyer education masqueraded you know in let's say flowery terms it all has to now has to be significantly driven through AI search and AI agents. Absolutely right. And just to add to that, so off late not off late but with products with a lot of SAS also testing out the PLG funnel where education where information on the website was used to then educate the users and at the same time there was enough and more video marketing which was done. What this also means is that going back to the core crux which I think the whole inbound marketing started with which is you know you build a content foundation I think that has become again important especially with content which your which agents can understand you know so getting a getting your agent getting all agents to understand your content so for example you know at Airmeet there are we have

21:02About 50 or 1,000 features in the product. But of course, all of this may be not explained. Everything is not explained on the website. Everything is not provided on the website because some of these things are questions which come and then there are certain tricky use cases which come which is when our product specialists go on a call and answer for the customers and then customers say, "Oh yeah, this works now and we would like to buy your product." That means all of those tricky workflows have to be also provided to your agent. They have to also understand that okay at this tricky workflow if they have to run 10 events simultaneously with thousand attendees each with ticketing with 100 other parameters all of those scenarios have to be somehow answered available somewhere so that the systems can pick it up so that system knows that okay these can be handled across so I think if you're moving away from or if the decision-m process puts human at much later like education through human much later in the process. I think a lot of content or a lot of content has made available for these agents to really understand products and their use cases each of those use cases.

22:13Yeah. And I think that again one again one recent thing that Google Google cloud team and deep mind team release which is the the open knowledge framework OKF right that OKF is is again a very very nice way to interlink all the concepts that a company deals with like for for let's say for a company like me right I think ultimately all our features are conceptually linked to each other right they all represent a unique let's say a concept that that someone should understand and then there are related concepts club together for everything to work together right so OKF is is one of the industry u let's say accepted ways to start publishing right so llm.txt txt was something that that that was never really a like a published let's say accepted framework. It was just like a best practice, right?

23:10Yeah.

23:10But I think OKF is being pushed as the way in which the company knowledge, marketing information or any corpus of knowledge can be represented in the in the best way that agents can understand. So that is again I would encourage our readers to sorry our viewers to basically go and explore OKF and and that again you know sort of tells me that Google is is punching like big punches on on let's say the standards which are which are very much required for overall community and like ecosystem development. Absolutely. Like but then a question is that while Google is laying out these frameworks but then would other LLMs or would other agents pick up on these? So for example if I use OKF and then I'm I'm able to you know follow all of those guidelines but then does chat GPT still give me that importance or so Google must be trying to do it on both sides. So one is of course on the LLM sides as well and of course on the publisher side as well right that's how they would then build it as a standard right protocol across the whole agent commerce.

24:21Yeah. So the the genesis of OKF is is is basically it was to it was innovated on or rather it was sort of conceptualized to solve a problem for large enterprises right so large enterprises who have let's say lot of documentation spread across different formats and all of the knowledge has to be accessible for internal agentic work and there's a very loose term called company brain or company knowledge graph that people have been using on in the in the ecosystem, right? So there was no standard for for for that company information or company company brain, right? and the challenge was that you know a department uses like department one is using let's say a different kind of a let's say technology stack and their company brand is in in in different format a subdivision two is using a different format. So the interoperability was always the problem right. So Google started to solve the enterprise problem by publishing this new framework where the knowledge is interoperable right so and portable rather so so imagine someone who is a tax consultant right he could basically create its own their own OKF for let's say taxation in taxation

25:41For freelancers who are operating out of Bali and who are US residents right US citizens so that entire corpus of knowledge as can be can be structured in a in in in an OKF and can then be sort of sold to let's say other agents who want to to access that set of let's say knowledge. So it is a one level above skills. Skills was a again a very similar way to give let's say expertise to agents but OKF is is a combination of skills and the context that the company has. It it has procedural knowledge. It it also has factual knowledge. when I say it has it has scope for that it has scope for adding procedural knowledge it it also has scope for adding factual knowledge that constitutes the company corpus of knowledge internally right and that makes it complete for agents to work on and the and the standard then ensures that this is portable across different let's say within the organization as well as across organizations now your question around the acceptability of this across let's say other agent tech AI labs or agent systems. I think Google has this very strong dimension of search and discovery, right? That is that is in in their favor, right? So nobody can ignore that and every

27:00Organization would have to would would obviously be always compliant with Google way of discovery and that would make the job of other agents who are looking for information, correct information in correct format, right? They would rather rely on something which is already out there, right? And Google is also doing it as an open framework, right? is I think the AI foundation the Linux foundation is is already part of it or I think the AI foundation is part of it I'm not able to recollect that but they're doing it in in the right way with bringing ecosystem players in it and it is in the best interest of other AI agents also to start looking for this right this format and you know because the information is best presented in this format so the end user benefit is is in just accepting this format Okay. Okay. Understood. L. Yep. Interesting. building now on top of this is as agents are as you know buying intent is changing on one side. We saw another news come out this week which was that Adobe actually Adobe is Adobe has a great product for enterprises which is Adobe Analytics and they actually joined hands with SEM Rush to provide enterprises with AI search data. So

28:20Basically an AEO product but then providing it for enterprises. So as we are seeing the importance of AI search growing we are seeing companies either collaborating together or building products jointly to be able to address that need especially for specific markets. So Adobe plays in the enterprise space and that's what they have they're trying to use that to be able to provide that data. So what we are seeing again with of course air pulses behavior we are seeing a lot of mid-market firms reaching out now for them for whom organic was a mainstay of their traffic they were a known brand and now with search moving to AI a lot of these companies are now looking to actively monitor how their searches are happening or how their brands are getting represented in AI searches. So I think the need for constant monitoring and then leveraging that process to then grow their brand or or you know ensure that their brand stays on top of AI searches in prompts in you know consumer behavior. I think this would be a key marketing effort going forward. Again something similar to what happened as organic traffic or with Google search becoming a

29:40Mainstay for lot of brands. I think this would also become a key parameter and also will be part of key brand building activities as well. So if I'm trying to build my brand, I have to cover of course I have to ensure that I do stay relevant and I do stay on top in AI searches as well.

30:00Yeah,

30:02Absolutely. U I think visibility and optimization go hand in hand, right? it is it is actually super important for organizations to to now ensure that they are discoverable by AI and also recommended by AI. Right? So as we have been discussing in this podcast I think the one of the highest potency lever that that that a company has is their own website their own content right so if they can be you know let's say highly descriptive about who their customer is and what is the unique value that they deliver to that customer right I think they will make the job of the AI engines lot more easy right but they cannot be they cannot be relying on let's say second or second hand or third hand information right so they have to take control of their own optimization and you know their own visibility right so I think what we are seeing with Adobe and Simrush is is is again like a a strong interest from enterprises to start optimizing their AI visibility and I think specialized players like airpulse.ai I I think we we we have we have an edge in terms of we have our proprietary simulation engine which sort of simulates the the ICP behavior the persona the the

31:21Prospect behavior in in ways in which are going to be very important for you to monitor the right set of queries right because when you are monitoring and if that visibility number is giving giving you a satisfaction right so you have to first ensure that you're monitoring the right prompts right so specialized players will obviously you know end up doing a better job in terms of you know let's say covering the last mile or making it highly actionable than let's say old generation players who are still rooted in the SEO mindset. Yep. Well said and yep it will it is becoming quite important and again just from some of the data that we are seeing at AirPulse the demand is coming from across the board companies which are very very new and want to establish a brand they're also looking to start off it gives them a new opportunity because for them to crack organic traffic organic search took a long time they had to build a lot of content but this is absolutely now there's a new so it's newer companies have actually a great way segue to be able to of course rank higher or ina ensure that their brand shows up more often in AI searches. So yeah, I think it's a paradigm which all

32:39These companies are using and just from companies like Adobe investing or you know partnering with key players in the market is a testament to the fact that even enterprise at enterprise level this also becoming a key problem that they're monitoring and trying to solve for. Okay, perfect. I think one of the last comments or last topics for our discussion this week is that in general we are seeing u of course we are seeing basically more and more AI visibility platform specific and non-transferable information coming through. So for example being cited in chart GPT does not predict Gemini citations. So going deeper into this whole AI search we are trying to we of course as brands we are all trying to understand how each of these engines are picking up our content or picking up our URLs and citing them or sending traffic our way. It seems like the logic for each one of these LLMs right now seems to be quite different. for example in one of our earlier podcast we discussed chart GPD anchors very heavily on social content or let's say authority from Reddit or

33:58Things like that. Gemini of course brings in a lot of YouTube for example. so how does one as a brand as a publisher how does one focus on this and how does one solve for this? Yeah, I think the so we have to start looking at these AI agents as personalities, right? And these just like personalities are driven by their motivations, inherent motivations, I I think these they are also driven by motivations of the business model, right? So a claude or a charg in in the avatars of their let's say a codex or a cowork versus let's say a gemini or a googlei mode they have very different motivations very different business models right y

34:47I think they would they would then behave in different ways right so what what we need to do as brands is to basically w sorry cast a wider net in terms of let's say our monitoring monitoring activity and u the good thing is that it is not a zero something right like you you can be recommended by charg as well as gemini right so they're not anti- each other right they are in some sense complimentary right because ultimately one of the core motivation is to be at the service for the user right so if the brand is doing all the right things with respect to being more helpful in the in their marketing content for the prospect to get educated by their agent. Right? I think the as long as that fundamental is taken care of, I think brands don't have to worry. But there will be some sources where your competition can be can be let's say acquiring an edge, right? Let's say for Gemini, a certain pocket of sources matter more. A certain competition of yours is is sort of having a lead on that space. So regular monitoring would give you action items which your team

36:04Can can basically do. So at Apul we have this we have on a regular basis we monitor all the surface areas and we have this datadriven way to give you the recommendation on like what kind of citations what kind of publishers matter more for your domain and week by week which competition is figuring you know sort of figuring out what position right and what did they do last last week or what new thing that they have done that have sort of given them the edge right so so that your teams can sort of incorporate that in their set of activities. So this will be necessary you know fortunately or unfortunately the era of AI search is deeply fragmented. the internet search era which was let's say from 1995 or let's early 2000s to let's say 2023 that was still dominated by one single player Google and the the the task of the teams which are optimizing for discovery was fairly simple in the era going forward it will be fractured fragmented and we should be ready for newer and newer surface areas to emerge right as we last in one of Our last discussion we discussed about open source AI models right or open AI models

37:23Which are which are going to gain a lot more popularity because they will be underneath the let's say an application that you're using right so I think the this the domain is going to be lot more challenging lot more fragmented for high confidence on discoverability I think teams have to you know let's They as I said cast a wider net of optimization and be very steadfast and rapid in terms of optimization activities. Understood. actually a follow up on that and actually you did allude to it and I had this question before in my mind. The idea is that do you think that at some point these there will not be so many LLMs and it'll just like there will be specialized LM for different tasks. So for example we have claude code which does coding we have cloud design which does designing design help and then let's say Gemini for all our general queries or chat GPD specializing in certain type of queries or there will be a Gemini deep research mode there's already that mode so but do you think that this also this could also end up

38:41Being like a two-horse race where right now of course there are many players but then eventually each of this becomes a big enough category and there are just two dominant players in each of those categories. So coding there is just two dominant players. general queries there may be three dominant players and then you know then there are specialized queries and then there could also be very specific. So if there is let's say someone is doing is in R&D or they are into research on nanotechnologies or material sciences research then there is a specialized LLM just for that which is like which does not then is much better than a generic one or which is at least gives more hallucinates much lesser because it's designed it's specially customized and built for that. So do you think that could be in a future that we're looking at and hence this problem that we are facing today which is like optimizing for multiple engines actually boils down to eventually just you know there are two players in the category in which you are service servicing and then eventually just optimize for those two. You think that could be a future that we could see maybe next 3 years or 5 years?

39:44Yeah. So there is a fair possibility of that to be the future especially the especially let's say consumer touch points are heavily concentrated in Google and Apple right u enterprise touch points are heavily sort of u enterp so so enterprise touch points have been fragmented but generally I think if you look at the the large enterprises like the fortune thousand right so they they've always had let's say a a fingerprint of Microsoft or a Salesforce right so the whoever owns the highest number of let's say customer base right they they have the opportunity to sort of consolidate the optimization responsibility or headache in some sense but I would say the future can also be highly fragmented right especially given the roles in let's say the diversity of activities that humans end up doing. Right? So for knowledge work there could be let's say a bunch of let's say a couple of players which are dominating. For artistic work there could be a couple of players which are dominating. For let's say more heavy

41:02Industrial work there could be a couple of players players which are dominating. Right? one one for sure user behavior that I am very confident in is that the the the AI agent that does my work is going to be the agent that also does my search right so it will it will not be it will not be two agents that I am I'm basically going to search separately and then for work separately right because skuomorphism at play I think will have a very high degree of reliance and connection with one agent that does a lot of our work, right? So imagine a teacher, right? If they are working with a specialized application or agent which helps them come up with the course plan, come up with assets, come up with let's say you know let's say evaluation let's questions and it also helps evaluate the responses and also monitors their teaching as an activity. right [snorts] now for anything that the teacher should need the agent is the best you know place to discover that. So if that behavior is very strong like in future if let's say if we see more strong u expression of that behavior then I think the service will continue

42:22To be fragmented and as as models like how do models specialize they specialize with reinforcement learning right the whole post- training thing that we talk about right and what essentially reinforcement learning is basically compressing years of experience into let's say the post- training phase right with a lot of these journeys and like how some real task has been achieved. You map the entire journey and then you sort of you know let the agent sort of in in a way consume that or or experience that go through that what what took let's say a human 20 years you just have to make that AI agent go through all of it in let's say you know like 3 months. So if you look at the variety of job roles that humans have today, right, there is bound to be a variety of agents which specialize in those roles because there's economic value in that, right? And with with people who are so dependent on let's say their own agent for their own specific job role, they might also end up searching or relying on that very agent for discovery of new services, new things, right? So I do think there might be that an impression of that in the future. But

43:40For some very high-risisk categories like health right u I continue to believe that there might be some specialized agents right specialized surface areas where you want for health you want to go there right or let's say or there could be a channel through which the specialized AI is the health spe specialist AI is sort of getting distribution for example

44:06And then

44:07Yeah through through hospitals the hospital applications can have an interface where people are you know engaging with the AI agent and then that is powered by a specialized health agent because those domains will be highly regulated and I think with regulation there is some bit of a defensibility and mode right so it would automatically sort of make it difficult for the you know the more flowers to bloom in that gardens so to say yeah

44:36Yeah That's true. and yeah, you you mentioned a great point about health. I think yeah, health would need something very very specialized. It could also be yeah there anything which is a very which has enormous amount of data and it's quite sensitive information and the use cases also. So another thing which again just off the top of my head could be some part in banking could be something which is specialized especially the risk assessment and things like those but yeah it's remains to be seen how this future unfolds but interesting times. Well, with that would like to end our discussion today and great to have great to have you here and everyone it was a great good session. Thank you for listening through the whole thing and yeah please do let us know in your comments how this was and if there's a specific topic question that you have feel free to send it to us as well. Thank you.

45:40Thanks everyone. Thanks Ash. Bye.