Building AI Agents That Work Arjun Pillai Docket AI

So the reasoning models kind of came up with this inference time compute thing. So before that what was happening is you would pre-train the model quite a bit and then the model knows a lot of things. You ask the model to do something. It is one shot. You ask it once and the model is going to give the answer. It is one shot. That's it. With the reasoning models, what happened is you can ask for something and these models started breaking down your asks into subtasks and then kind of do the task again and again and again and again to a point where during the inference time during the computation time is taking more and more cycles generating more and more tokens to kind of generate the right answer. >> Hi, my name is Dmitri Bonichi and I'm a content creator, agency owner and AI

enthusiast. You're listening to the AI agents podcast brought to you by Jot Form and featuring our very own CEO and founder Idkin Tank. This is the show where artificial intelligence meets innovation, productivity, and the tools shaping the future of work. Enjoy the show. Hello and welcome back to another episode of the AI Agents podcast. I'm here with Arjun Pilai, the co-founder of docket.io. How you doing today, Arjun? >> I'm great Dimitry. Thanks for having me. Excited to do this. Yeah, thanks for being on the call. It's uh it's been a few weeks since we did the the pre-hat, but I'm really excited to get into it. We're going to be talking today about how, you know, you can crush your critical go to market targets with powerful AI agents as your tagline says. Um no, but in all seriousness, uh give us a little bit

of insight as to just a short, um story behind how you got into AI AI agents and then obviously focusing on uh what you guys are doing over there at docket.io. Yeah, I I've been in the sales tech mc space for a little over 13 years now. Good or bad. Wow. Um yeah, docket is actually my third company. Uh my first company was a sales intelligence company back in 2012. Uh I was 23 at at that time. Um yeah, so I I graduated worked for like just about a year resigned and started company. So when you are super young and stupid, you don't know what you are getting into if you get to start a company, right? So if if I really knew what a startup is, I probably wouldn't have, you know, it's so tough. But uh anyway, so that's how I got in.

Sales intelligence was not a space. I was in India in Kerala. That's the state southwestern uh state of India. >> Sure. Y >> um then then yeah, I built that company. We grew to 72 people. I sold it to a company in Denver in 2016. they moved me over here um and I was already into data sales tech and martekch you know a little bit of predictive AI some bit of predictive analytics and sentiment anal analytics all the you know prelim AI and machine learning stuff so I had a little bit of an exposure then I did another company that was into conversational sales and marketing um that had a little bit of AI again then I sold that company to Zoom Info and many of you might know it it's a NASDAQ listed public company uh they are the the biggest when it comes

to sales tech uh martekch data tech primarily I mean sales tech maybe salesforce is the biggest um so I was there I I built the accountbased marketing platform for them that had a lot of AI because bidding optimization everything is machine learning not LLM AI but the more predictive AI then um I was the chief data officer of Zoom info again a lot of machine learning AI stuff things like that the data science team reported up to me um then started docket um at it was the time in GPT3 was just 3 3.5 was just starting to happen. Um AI was happening thick and fast. Uh so that's the kind of rough background right. I've always been in this domain >> and then um I I've been uh uh fortunate to kind of play around with some of the AI things in the past and

when AI was happening since I was in the right place at the right time I saw when it was coming. That's pretty much the story. >> Very cool. No, I that's a great start to the the conversation, right? And you know, now just getting into it a little bit more. What was the um you know catalyst that started your current company with your co-founder? >> Yeah, I mean I'm a startup guy. Um I always like starting up. I always like solving a problem, building a team. You know, the the journey is the destination when it comes to a startup. There is a learning that happens every single day, every single hour. And I've been always a lover of that that journey in a startup. it's very peaceful for me. Um so that's the reason for starting up but then specifically dockit the problem statement that

I saw um while at zoom info and also pretty much all the companies prior to that and also with the revenue leaders that I've been talking to if you take any company roughly actually less than 10% of the go to market knowledge is ever documented in a company you know you talk about the document document of sales and marketing knowledge in a company it's less than 10% the other 90% lives in unstructured data like a call recording or a slack chat or something like that or lives in somebody's head. You know, this is your product people, product marketing people, sales engineers, engineers. Uh and and when I was looking at this, you need this go to market knowledge for the sellers to perform and the buyers to buy, right? So, it's just a no-brainer to have this thing together. But until language models came along,

this was a super hard problem. you couldn't solve this problem of this largely unstructured data and then generating knowledge out of the unstructured data was not solvable. So when GPD 3.5 was happening, I was like this is potentially the first time in the history where this problem can be solved and you can build out a knowledge graph that is grounded on a particular enterprises knowledge and that in that enterprise you can have a salesperson get access to the information and the buyers can get access to the same information powered by the same today we call it sales knowledge lake but basically the fundamental same data lake. So that's the uh rationale behind docket and that's what we have been executing over the last two years. >> Yeah. So I guess kind of explain a little bit more I I think maybe people could understand how

and why it would require um large language models to exist for for this to kind of take place but explain a little bit more like what they what they ended up solving problem-wise. >> Absolutely. So take sales right when a new buyer is coming um the typical things that they want to know let's say at the top of the funnel at the awareness consideration stages of the buying journey what we have seen is there is about 20% of the questions that get asked 95% of the time what is your pricing how do you price what are your how are you different from this competition do you integrate with Salesforce do you have sock 2 you know like you take a set of the top you know 100 questions Those questions are getting asked like 95% of the time at the top of the funnel and

then as you get into the details the details kind of increases a little bit but there's still a clear parto principle right there's 20% of the question that get asked 80% of the time so if you look at all the call recordings that are happening in the company if you're looking at the right Slack channels happening in the company if you look at the sales enablement collaterals if you look at the RFPs if you look at the website I explained like six sources There's probably four more sources like Jira or Confluence or Zenesk or Salesforce. You know, look at these 10 or 12 sources in any given enterprise. You have the knowledge there. The tribal knowledge in that company is all there. But the challenge is how do you take this super noisy data some some is super high signal right? For example, if I'm

preparing a PDF, a product marketer prepares a PDF that is high signal to noise. It's very good data. But if you look at slack, it is super low in signals. It's mostly noise. So now you have to take all these different kinds of inputs, do a particular data processing and then bring it into the lake. And then you have to figure out okay now I have got signal which signal is accurate and which signal is inaccurate. A good example to that is imagine a CEO says an answer to a question and then there is a new account executive who joined last week answers the same question. I would argue the CEO's answer is probably going to be true. Right? So if you know who are in the company and if you weigh their answers slightly higher than the rest of the tribal knowledge, then you

can get into a higher accuracy. If you consistently purge out the outdated and conflicted data out of the lake, then you have better accuracy. And how do you do all of these language related things on top of the knowledge? You need a language model to do that. >> That's where you needed a GPT4. to be honest until GPD4 landed it you couldn't solve this problem at the enterprise grade quality after GPD4 GPD 40 GPD 4.5 GPD5 and all the other like cloud and all these additional models today we are at a point where yes the models are good enough for us to build for you know any kind of enterprise in the world >> and would you say that you know as uh things kind of change and and improve even on the models I know you said that we're at a good enough level

for enterprise but what maybe new unlocks have you found have come as models have improved right like yesterday to my knowledge uh claude 4.5 sonnet just basically threw like a wrench into everyone's idea as to how good an a model can be agentically and and how long it can self um I mean the previous record was five hours of autonomous work right for coding and then they said huh 30 which um it was supposed to it was supposed to double not 6x X. So yeah, >> I think AI is going exponential all the time, right? The exponential curve has probably slowed down a little bit, but it's not it's still exponential. It's not linear yet. >> At some point, maybe it will. >> Um, so what is the biggest change that happened? I think the biggest change happened probably in September of last year when

the O series of models from OpenAI came through models. >> Yeah. So the reasoning models kind of came up with this inference time compute uh thing. So before that what was happening is you would pre-train uh the model quite a bit and then the model knows a lot of things. You ask the model to do something. It is one shot. You ask it once and the model is going to give the answer. It is one shot. That's it. With the reasoning models, what happened is you can ask for something and these the right answer. So that that's where the reasoning started to come through you know to to whatever extent we call it reasoning right. So I think that has been the biggest change that happened end of last year. So the biggest one was 01 followed by deepseek's model kind of kind of changed

the whole paradigm right in November I think deepseek launched and that everybody was like wow okay this is awesome and and the cost is not much and things like that anyway so that uh changed so that enabled us and a lot of the other companies to start going into more of an agentic path wherein instead of do using oneot answering we all started using agentic architecture for everything that we are doing. What that essentially means is rather than just asking the LLM to do one task, you are now starting to give a responsibility to the agent and the agent is also responsible for taking uh like understanding the responsibility, deciding which are the tasks, executing the tasks, calling the right tools and getting the whole outcome delivered. Right? That's the big shift. And I don't think any company has completely internalized and built on top

of it yet. Cloud cord is obviously amazing. Uh coding is the biggest jump that happened for GPT 4.5 sorry cloud 4.5. Um but yeah so that's the world that we live in including us. Everybody is kind of taking that agentic architecture rebuilding some parts of it rearchitecting some parts of our platforms. >> Yeah. No it's very cool and very exciting. Um, I feel like every month I every week I have a freakout moment at this point. Like I saw that uh earlier today. It's funny. I was mentioning this to a video um AI agent company. They you know like I don't know if you saw Sora uh 2 just was um released two hours ago or three sorry three at this point is when I found that out. Um and it's it's crazy. It's it's it's constantly uh developing. It's constantly improving. And as somebody

who's in a company that is built foundationally to help people go to market and I'm sure using a lot of these models in order to do so, how do you as a founder um and as a company kind of continue to iterate and improve upon your product and stay up to date with all the the newest and latest tech? >> It's difficult. Um the source I would say is Twitter. You know, as much as Twitter is noisy, I mean I still call it Twitter X. Okay. No, no, no. It's okay. You I don't call it X. I still won't I won't do it. >> Yeah. So, as much as the AI Twitter is like um very very noisy, I still think that Twitter is the place where good people are putting good content. There is LinkedIn. LinkedIn is a lot of hype. You know, fundamentally

like people take things from Twitter and put it into chat GBD rewrite and post on LinkedIn. Honestly, not everybody, but >> that's fair. No, that's a common thing. It's a lot of AI slop right now. Yeah. >> Yeah. Yeah. Yeah. versus Twitter there's originality right people are actually coming up with things there is um pretty good discussions specifically in the AI topic when it comes to marketing and there are some topics where I would say LinkedIn is the place to be but um at least in in AI it is uh Twitter RX so that's where the the learning starts um but then the what are the things there are two things one is something has launched and then how do you use that how do you build better on top of that and things like that but then there are also the architectural change. How

is claude cord being developed? How is claude developing cloud cord? Or how is open AI improving chat GPD? So those are more architectural discussions, right? So as an entrepreneur or as founders building in this, we need to understand both. One is how is this latest thing going to help me? How do I build on top of it? How do I use that service? Then how did they build it? So that we can start learning that build process from them. So for example when we are architecting our our platform kind of in a more agentic fashion we are taking a lot of inspiration from how open AI is architecting charg how they are using the tool calling what are the what are the things that they call as tools and what where do we think they are going to and then we kind of decide okay

this is how we should take the path um and and there is a third layer too the third layer is the product product direction right one of the things that's happening with the foundation models is they are all coming up the layer Um a foundation model is not just a foundation model today. There is an app layer to it. Chad GPT has the charge GPT app layer that is uh that has got enterprise connectors and document generation and so like there are a bunch of things happening on chart GP at the application layer that is competing with a lot of startups. Similarly, cloud has a cloud code layer that is competing with the cursors and all the other coding agents layer. Right? So the foundation models are also coming up the stack and as they come up they are killing a bunch of startups. So

the third layer to all of this change. >> Yeah, absolutely. It's happening right like um video generation to your Sora example or Google launched their video generation. I'm sure a lot of the video generation companies are like where do we go from here because these guys are coming up the stack. Um ad generation video companies right they're generating ads but if Google can generate the ads out of the box with their foundation models what do we do? We'll add workflows and all but anyway so that's the point. So those are the three layers that we look for. How do we use them? How are they building that so that we can build accordingly? Uh what are the product directions that they are not taking so that we don't become a roadkill to some of these companies. Those are the three layers I'd say with >>

roadkill. Sorry. It's true. It's a good way to actually put it. It's just kind of like crazy that that is the it's true. Sorry. It's just a funny term. >> Yeah, for sure. >> Yeah. Um, so I I guess you know that does that does actually give it give me good context as to as to where to go with the rest of this, right? And and when you are building out a product as it were and attempting to not become that kind of quote roadkill, right? Um, what can you do and what do you believe that you do at such a really great level um that compares uh to your competitors and and you know obviously you're working in both marketing and sales. Um, I think some of the the more interesting stuff that I've seen probably is from the autonomous AI seller uh stuff

that I was checking out you use case- wise. Could you could you kind of speak to that a little bit? I think it it's pretty cool. >> Yeah. Yeah, absolutely. I believe that there today all the sales is I'm a human selling to you as a human, right? H to human to human selling is what is happening in the world today. >> But I believe that in the short term, it's already happening. There will be an iteration of the selling process where humans will become super reps because humans are getting augmented with AI. AI is enabling reps to do more. So there will be a bunch of good sales people who are really using AI to better themselves and be very good at what they do. And then there is a second set where there is an autonomous selling. I truly believe that a lot

of the sales process will be autonomous in nature. Right? Think about a $20 thing that you are buying. You don't want to talk to a salesperson and companies don't cannot put a person to talk to you for $20. That'll be autonomous for sure. You will still have an agent talk to the agent just buy. You go to uh you go to Dyson's website to buy a Dyson vacuum cleaner. Today it is just a static page. I believe that that page will have a concarch agent. It's an autonomous seller who is talking to you, right? So that's what we have built at docket. So if you go today to docket.io, IO, there's a marketing agent, an agent that pops up on the website. Click on it, it's going to just start talking to you with voice. So, it's a truly multimodel experience where it talks to

you with a combination of voice and text and images and videos and it can it can kind of ask for your email. It can qualify you. It can book a meeting for you. Can take actions on your behalf. It can write back to Salesforce. Bless you. Uh so, so that is the autonomous selling that we are talking about, right? So we at docket we believe that there will be a lot of the selling that will go autonomous and we are one of the earliest movers into their paradigm where we want to really get companies uh kind of understand what does autonomous selling look like? What are some of the use cases where they can deploy it without a lot of risk and how do you kind of increase the efficiency of the go to market and experience not only for the sellers but also for

the buyers. >> Okay. Well, you know, you brought up something pretty interesting here, uh, that I I think between us probably is understood to be a part of the AI spectrum, so to speak. But a lot of companies and even business owners are kind of at a point where I don't think they consider the multimodality of of AI, right? I think we're we're early enough in AI where it's fair to say that most think about it from the chat, you know, level still to some and the text to some respect. Can you speak a little bit to the progress of voice and in that part of multimodal multimodality because I I I found it to be incredibly big in the last few months. I mean I remember when the alpha of 11 labs 3 dropped I went okay here we are it's here you know

>> we are here yeah >> yeah yeah >> yeah exactly yeah we we also use 11 labs for our our um texttospech layer we believe 11labs is the best out there at least right now um see voice has been a promise for a really really long time back when Steve Jobs acquired Siri in 2011 I believe back then itself voice was like thing. Obviously, Steve Jobs saw it back back in the day and then Google introduced the Google Home. Um, Amazon introduced Alexa. >> The amount of investment that Amazon or Google, all these guys did on these voice enabled systems that they thought is going to rule our homes, automate everything, you know. Um, today you can't even see enough Alexexas and Google Homes anywhere and and Amazon has like really reduced their investment in Alexa and and whatnot. But so voice has been a

promise for a really long time and people were all excited about voice. So it's not like people don't think voice is a good modality. I think we communicate very normally in voice. And in a world where translation from one language to the other language is super easy. I think voice is going to be like way more better used. like my parents can speak in malayalam which is our regional language and charge GPD can respond in malayalam or English right so it's that that easy today so I think voice modality will kick in today it is still in the early adopter stage but if you look at people who are using the voice mode of chat GPT or Gemini it is increasing right and I think the first place where voice modality is we are going to be comfortable with the voice modality of AI is

everywhere we have an IVR Today IVR is like call in and say if you're this press this button book that is the IVR thing that all of us know and we wait for 20 minutes for an agent to speak with because this IVR is not understanding whatever we are telling it. Uh so that is going to be the first place where we are going to see IVR getting replaced by amazing AI voice. Um and then we see contact center automation right a lot of the contact centers where humans are responding today that is going to start getting automated with AI because AI can speak AI can understand AI can take actions you want a password reset you don't have to talk to a human AI can do that for you so those are the places where the voice AI modality is going to kick in

and then people will start getting comfortable with it and over a period of time you will start seeing that going into different parts a restaurant booking front office of a uh sales and marketing processes, all of it will start getting voice modality. >> Yeah, that makes sense. I I think it's um it's going to be more universal than we think moving forward. And um I mean maybe we don't have to spend too much time on it, but why do you think that it kind of fell out of practice maybe on the on the home front? Cuz I do have uh two of the um mini HomePods at my house. Um, to be quite frank, I kind of use it only because of the ecosystem and being able to use them as speakers and stuff like that. I I quite frankly think Siri is a steaming

pile of garbage at the moment in comparison to everything else that exists. Like just to be frank, but um like if you compare it to the other tools, but why do Yeah, it you have one. And what's really frustrating is >> it doesn't even use the cheapet like capability that released on all the other iOS devices. And I'm like I love this for Yeah. >> Yeah. It's not even plugged in, right? It doesn't work, right? Why would I It's just when I call somebody Shri, it'll just light up. I was calling somebody else, but Siri would just Yeah. >> I think the reason why it was not working is because the voice modality was not working and the actionability of like the real usage of that was not exactly. So I can ask anybody what are you using Siri for? I'm sure they are going

to say two things to set alarm to set reminders. >> Set alarm. Set reminder. Yeah. Ask what the weather is. Maybe play. And the playing music for me is the only thing that I'm like, "All right, that's kind of nice." And when I'm cooking and stuff, but like it's the worst by far from my understanding versus Alexa or uh even Google Home. But um yeah, hopefully they'll hopefully they'll improve that moving forward because uh that's the that's the rumor mill. The rumor mill is claiming that they're going to do a complete overhaul. Um, but they've been saying that since 2017. So, >> yeah. Yeah. Apple has a lot to catch up on Apple intelligence in general. But overall, the voice paradigm, I think, did not pan out because the ability of the agent wasn't there to understand what a human is saying in a very

very natural way and then being able to take actions very organic way. It wasn't that easy, you know, but I think it'll all get settled. the voice paradigm will get better. A lot of the things that we are using will have voice um kind of going in deeper. Um it's just taking its own time. Um because we have gotten used to if you look at it when did we start using a keyboard and mouse? It is like decades ago, you know. So everything takes its own time to kind of come into our world when it comes to change management. >> Yeah. And I guess a fair follow-up question to that would be when when do you think kind of the change will will come universally in regards to some of these sales and marketing practices being AI um AI AI identified however you say this

completely across the spectrum. So it's almost like asking uh you know it's it's difficult to say because you know I saw you drinking water right my question would be which drop did you drink to curb your thirst right it is not in it's not probably in the first drop it's probably in the last drop it's somewhere in the middle >> somewhere in there yeah >> right so it's almost like that so when would a company be AI uh using AI for sales and marketing I would argue that probably most of them are already using AI in some form or fashion But it's for either low value tasks or is it it is just for content generation or it is just for Salesforce data entry or is it just for conversational intelligence uh deal review whatever like some small thing it is a um absolutely being

used for back office things today more than the front office things today. So there are a bunch of these things but I would say that in the next two to four years we are going to see significant penetration of AI happening into the sales and marketing organizations across enterprise and SMB. Obviously SMB will adopt first mid-market will adopt second then enterprises will will adopt more um but I think it is already happening in a pretty significant way and we'll this this journey is for the next 10 years or 20 years. So it's got to be regularly long road and whatever we are seeing right now is still very very early. >> Yeah, that makes sense. Yeah, that's a good explanation. I think the somewhere somewhere in the sip Yeah. of water was I was I satiated but I don't I don't know where it is.

Yeah, and it's a fair it's a fair point. um that I I think you know and do you think it's be how much do you think is just like time and adoption and how much do you think it's like a lack and this could be a part of it a lack of trust by companies that it can do what it can do and how and also capability because you said enterprise capability already probably there right so do you think it's just a matter of trust and time >> um that's a very good question I think it's a combination so the enterprise change management is is a real problem, right? Like we have a way in which we have been doing business for many many years and then just because you have this external agent or tool that comes through, you're not going to be able

to completely change it. And in the agentic paradigm, the biggest problem is um I'll give this as an example. We have people dying on our roads every single day because of human errors. But if Tesla kills somebody, you know, we are going to learn about it. Every single media house is going to report it, right? Because in the world, we have built this world to be okay with human errors, right? We have checks and balances. Whether it is within a company or out there on the road, there are checks and balances for human mistakes. But if the the agentic policy of mistake, we are yet to figure out in an enterprise. We know that sales people are saying all the wrong things on sales calls all the time many sales people. But can an AI say one wrong thing? Can AI hallucinate even once? The

answer in most cases is no. We cannot have AI hallucinate. So if you are waiting for AI to be 100% accurate, you are never going to uh get AI in. Right? So that is the journey that companies are on. Just like how um we are okay with humans making mistakes, how we have checks and balances. The policy shouldn't be like AI shouldn't be making a mistake, the policy should be like when AI makes a mistake, here is what's going to happen, right? And and enterprises need to kind of internalize this thought process. So that will take time, right? A company like a Microsoft or a Google or some of these bigger companies, they cannot simply accept that AI is going to just make a mistake and that's that. Uh that's why an SMB is going to adopt it first because it's okay. Nobody's going to

sue a a 40 people company try to get $2 million out of them. Um so now >> again I know I know that it's not a direct answer that I gave but that is the challenge that the world is grappling with and and Demetry honestly on the other side right unlike any other technology revolution this is the speed of the buildout the capability buildout is happening way too fast for the adoption to catch up how much have we built out from a capex standpoint like a trillion 1.1 trillion I don't know like last I checked it 900 billion. I'm sure it has gone crossed a trillion by now. And then um OpenAI is saying that they are going to give 300 million to uh uh Nvidia and Oracle is saying that they got like 300 uh billion from OpenAI whatever right so some some crazy

number. So we are talking about 1.1 trillion of buildout that has happened in a capex and to back up on the actual AI revenue numbers how much do we have? I mean 15 billion by open AAI 10 billion by anthropic u and cursor is making some money lovable is making money half of that directly goes to anthropic the other half is their revenue so like you know let's say 40 50 billion is the active revenue that we have and people who are listening to this if we are going wrong here call us out and tell that it's not 50 it is 70 that's fine but there is a big gap between where the revenue is and where the capability buildout is there's like a trillion dollar difference between them right so the trillion dollar difference is the adoption curve. So people have to adopt it

to a point where the revenue matches the capex buildout that is happening at least at some level, right? So um that's going to take time. The reason why we are kind of impatient about AI adoption is because the capability built has been going way too fast for the adoption to catch up. >> Yeah. You know, I think this is a fair that's a very fair point. like the the problem with a lot of um companies uh right now and and and where they're at, I think, is said this many times on the show, so hope maybe people don't think we're belaboring the point, but tell me the last time an associate was good at their job. >> Yeah. >> Like not not to say not to say it too rude, but tell me the last time. And and the joke there is like, well, they're

not going to be an associate that much longer if they're good at their job, right? They're going to be promoted. So, you know that there's this level of like inadequacy in new hires, whether it be because they're new to the industry, new to the workforce, um, or just they struggle at jobs consistently and whatever it is. And um, that tells me that you're willing to accept 60% of the time is a good, right? >> And it's like across the board in every industry. And they're like, "Oh, it's only right 90% of the time." I'm like, >> "Do you hear yourself? Like, you're not you're not okay with this person and you've been paying them like between insurance and their associate." I'm like, "I know what the median salary is in this state. I know it's probably like $60,000. Plus, you have to pay for tax

salary tax, plus you have to play their social security." I'm like, "So, I'm like, so you're spending $7,800 a month practically >> and you're okay with 60%. You're not okay with like $300 a month for like 90. That's kind of wild, don't you think? >> Yeah. Yeah. Yeah. It is the policy to to my to my earlier point, right? We have the policy when a human underperforms. What do you don't have a policy? >> Yeah. Like you know, so you just have to prep the world there and it's a change management problem. It'll take time and we should all just be patient about it. There are some markets in AI that are at PMF uh which is like coding agent. Clearly that market is at PMF. As our cursor is going to make it, who knows? Probably yes, probably no. Like we thought Vince surf

was going to make it. It did not. It got dismantled in the middle. A good outcome for for at least the investors and people involved. May not be for the employees, but overall a good outcome. But um you know like we don't know which companies are going to win, but like the the application building space like the lovables of the world space is at PMF. Coding agent space is at PMF. But some of the other spaces in AI, they are not at PMF yet. Uh go to market AI is clearly not at PMF. There's no go to market AI company that is like inflecting like crazy. And and again listeners, if you know a company that is inflicting like cursor, let us know. I'm happy to be considered wrong. I actually would love to be wrong here. Uh the >> I don't think I agree

with you. I don't think there is. >> Yeah. Yeah. The other things that seem to be inflecting pretty well is the vertical AI stuff, right? the industry specific vertical AI companies they seem to be doing pretty well there's a significant amount of services also that are going along with the AI so they are literally doing services FD is forward deployed engineers and delivering that outcome to the uh the vertical uh industry companies and there the growth is pretty good actually so those are the places where AI is starting to see the PMF but you know everything will get PMF it's just a matter of time >> yeah no I think that's fair um It's uh it's something where maybe like with we acknowledged with Steve Jobs earlier, it was early, you know, >> right? And um it with Siri. Oh, I shouldn't have said their

name. >> And see, I got I got stuff. I got everything reacting though. Nope, I wasn't talking to you. All right. Shouldn't have said the name, but with with the S with the S person. And um you know it's very interesting and and I guess from your perspective then you're in a position where you don't feel like there there is that big company right now. What do you like focus on on a day-to-day basis as a as a co-founder in a company like yours to make sure that you're you know providing value to current uh clients and and customers and trying to bring in more customers and you know improving the product in order to kind of get that um product market fit. Our job is to help the customers get to that paradigm, right? That's simply it >> for us. Uh, see, we are

a boat and for us to win the tide has to rise. >> That's it. So, our job is to rise the tide, right? And the bot will automatically float. >> So, our job is to kind of get um the the message out to the world. Here is how you can bring AI to the go to market AI strategy that you have, right? Here is how you can improve the go to market. Here is the step-by-step process that you can do. We recently rolled out a a a white paper on AEO uh answer engine optimization. U we put in a ton of work into that white paper. We got so many people thanking us you know uh for for that white paper because it really added a ton of value because people are confused beyond doubt on what this answer engine optimization is and there are

a lot of vendors who are also confusing people and then there are some vendors who are playing it in the right way. It's a AI is generally confusing to buyers. So our job is to help the buyers see through um some of these these noise that they have and help set the expectations that AI is not the magic wand that Harry Potter got. Right? That's a different wand. AI is not that one, right? So don't don't think that the Elder Wand is going to come wave and your enterprise is going to become Hogwarts, right? Not not happening. So set the expectations, not promise the world. Help them see through the noise. Yesterday Chad GPT introduced this uh instant checkout feature, right? You can buy from Chad GPT in Etsy and a couple of other places. Stripe launched a bunch of things alongside. So I wrote

out an email and we sent that out to a bunch of prospects and customers saying here is the launch. Here is what that means for your business and here is what you can do about it to get ahead of it. Right? And people responded back saying that thank you. I did not even know that that launch happened because it is not their job to know that Sora 2 launched. It is our job and once we know that we inform our customers that is our job. So help them see through the noise. Inform them what's happening what it means to their world and setting the right expectation. This is what we do. We consistently do it enough and then customers when they are adopting it, they'll automatically choose docket. >> Yeah, that makes sense. Yeah, that's sure. you know, and I guess just from an internal

standpoint at your company, right? Um, it's got to be interesting, you know, getting more into general AI questions for a moment here. I think we we it's been really good to learn about what you guys are doing and your thoughts and what do you think is going to be the general adjustment for daytoday tasks and agent well day-to-day tasks being impacted by AI agents um on a consistent basis and what do you think the path is for that? because I've interviewed some people. I just interviewed the the head of a AI department at a company, which is actually kind of crazy that we have those now. And he's essentially in his personal time going to each department and trying to improve what they're doing, right? With uh >> the Yeah. What do you What do you think kind of the method is for people to

get their tasks uh lessened across an entire company through AI in general? because obviously you have your own specific like thing you're doing and maybe at your company right now you're trying to see on the on the book ends like email this that and the other thing like how do you feel like this is going to happen internally at your company and maybe how it'll happen at other companies too AI and agenticly decrease the workload across the board >> let me start with the northstar of what companies should aim for and then I'll try to pave a way to get there right the northstar is um back in the day when people are starting to become leaders or managers, you would teach them how to be a delegator, right? Here is how you delegate and the thumb rule is every time you get a task,

your thought process is who on my team will do this task better and should I keep it or not? So these are the two questions a good delegator asks, right? So give find the right person and hand off the task or responsibility to that person. Keep only the things that you can add value to yourself, right? This is how you become a good delegator. Now in the AI world, the northstar of your company should be every time a human is getting a task, their first thought process is how do I get AI to do this? If you can get all of your team members to think that way, you're an AI native company and you will automatically find the right tools and services because people are smart. They will figure it out. So this is the northstar. Now, uh how do you do that? How

do you do that is by showing a bunch of examples, right? So for example, whenever I get something after every single sales call that I do, I I have a project that is set up in chart GPD that automatically takes the transcript that automatically uh you know creates a an email for me that creates a call summary for me that creates a business justification for me. It automatically updates my Salesforce. It sets up a follow-up task for me in Salesforce. The the whole thing is done by AI, right? That's one thing that I do that AI is kind of giving me a lot of leverage. When I did that, I took that and I put it into into my Slack and said, "Hey guys, here is what how I am doing it." So share examples whether it is your example, whether it is the example

that you found to be amazing on Twitter. Just take them, keep sharing it with people and encourage the top people in your company, the executives in the company to go AI native first. And if you need an external consultant to help uh them out, get that external consultant to make sure your executive is becoming AI first. If your leaders are AI first, your team is starting to become AI AI native. The last thing that I'd say to not make the answer too big, humans are controlled by incentives strictly. That's it. So if you cannot incentivize, then the behavior will not change. You know, if you want a salesperson to sell a particular product, you include a spiff for that product for that quarter, they'll sell the out of that product. That's that quarter, right? That's the incentive. So, if you look at um you know

a company like a Zap year or many of the AI companies, their hiring strategy now starts with EI. If you're not AI native, don't apply. Right? So, the incentives have to be aligned. So, when you have your bonus structures, look at how much AI they are using. Look at the analytics of the AI tools that they are using. If they are not leveraging it enough, that should impact their bonus. If they are using it enough, give them more bonus because they are actually saving money for your company. So find the right incentives to make sure that you are giving the pat on the back for the right people to going AI native, right? And then and then showcase people like in your monthly all hands, just showcase a person how they went AI native to the entire company. The company will automatically go AI native.

So these are the fundamental things. This is regular change management that you have to handle, but this is how you should handle it. >> I think this is a very good point. I I think you're you're you're spot on here because, you know, incentive structures are what make the world go around and what especially what make business go around. But I don't I don't necessarily think a lot of people uh are doing a great job of handling incentive structures uh for AI at their company. They're like, "Oh, you guys should really try to do this. You should use it." >> Right. there's not like a necessary playbook or, you know, incentive structure. So, I think I, you know, totally totally on board with it here. Uh, I think you're I think you're on to something. So, yeah. No, I that's that's a great point. I

had considered the incentives. That's the first time we've gotten that answer. So, thank you for that. >> Perfect. I I believe that executives are in the right uh task or at the right vantage point to decide what are some of the top problems that should be solved using AI. See the best way to bring in the right AI to your company is to start either from a problem statement or from a KPI that you want to impact or from like a um uh like a solve a pain for a particular segment of people that you have. It goes back to the first one kind of. So problem and KPI. So in most cases executives know the pain top pain points in the company. You ask a hands-on CEO, hands-on COO, hands-on CRO, hands-on CIO. Operational leaders in companies, they know where the bottlenecks in the

companies are. Right? So mistake that the leaders are doing is they give very open-ended direction to their team saying go implement AI. That is too open-ended. Don't do that. You know the pain points, right? So you spend the time to figure out those pain points and hand it off to the right person. If it is in revenue, give it to RevOps people and say our our conversion on our website today is too low. It is 67%. I want it to be at 0 92%. Go figure out the AI that can help me get there. That is the right direction. But executors shying away from that gets everybody going into 100 different uh directions. That is when you have to now implement a head of AI committee an AI committee that is trying to pull back reel back all these different initiatives happening. But if you

have been more um you know directional directive giving more convicted at the top about these are the problems I need to fix then people will run into lesser set of directions and it's easier to reel it back in. >> Yeah, that's a fair point. Yeah, I didn't consider I didn't consider that either. I I guess you know we are kind of coming close to to time here. So just want to give you the opportunity to speak to anything last but not least kind of that you'd like to share about AI in general and obviously what you're doing at docket.io >> with AI. I think here is my parting thought I guess. Um fantastic discussion by the way you know I I enjoyed the whole thing. >> Uh I think what you need is a quick win. So please don't take projects that are going to

take 9 months to deliver. Who knows? You might not be in your seat by then. You know, I I'm being honest. I see during our deal cycles, you wouldn't believe how many people um who we are talking to is not there by the time the deal cycle is ending, right? It's crazy. People are letting go of a lot of I mean companies are letting off a lot of people these days. So what you need is a quick AI win where you can say this was the problem this was the solution this was the impact this was the lift that's it right and the moment you are in a position to go and present it to your executive and at your all hands at your board meeting that is when you are getting valued so don't go for like the big swanky transformation project that you

have done in the digitization world or like the the the SAS world that was a different world In AI, you need to pick something small, not a point solution. You need a vendor that can do more than just a point solution for you, but get a quick win under the radar. Do a pilot 3 months KPI. Here is a success metric. Get it done and and show that and that will get you the confidence and people are going to let you do more after that. So that would be my u my two cents with docket for example, right? This is what we tell our customers. Hey, with docket you're going to sign in 20 days. We are going to make you go live. 2 weeks after that you're going to get the first set of additional leads and in about 3 to four months you

are going to get to, you know, uh 2x ROI, right? And we will put the report together for you to present to your leaders. Easy enough, right? So, if you're a marketer, this is what you need to do. It's a it's a it's a lowhanging fruit. So, focus there. Get a quick win under your belt. That's what will enable you to do a lot more. >> Wow. Yeah. I think that makes a lot of sense and I by the way I enjoyed this conversation um as well and you know I I want everybody to go check out the website dockit.io that's their domain docket.io for Arjun and everything him and his co-founder are doing over there. I think we're we're really happy and glad to have had you on the show and just everyone listening as well as Arjun. Make sure to leave a review

on the podcast uh so we can uh help out those ratings for Arjun and get the good word out about um what he's doing over there at doc.io. For those of you who are also not subscribed on our YouTube channel, please make sure to do that. We primarily get a lot of viewers over there, so we'd appreciate that as well. Thank you so much for watching this episode, and we'll see you in the next one. Bye, guys. [Music]