Why Most AI Projects Don’t Show ROI | Sid Bharath

Vit Lyoshin (00:00.994)
Welcome, Sid. Thank you for your time today.

Sid (00:03.893)
Hey Vitt, thank you so much and thanks for having me on.

Vit Lyoshin (00:07.916)
Yeah, of course. so let's start with conversation this conversation with you're working with many companies, talking to many leaders, and when people implement AI projects or initiatives in their organizations, most of them don't see ROI, at least from like my research and statistics I looked at. Why do you think that happens?

Sid (00:34.443)
Yeah, I think like the number one thing, you know, when I'm I've looked at these numbers as well in all of these studies. for starters, they're looking at kind of bigger projects that some of the bigger companies have done and projects that have taken maybe a year or two to implement over the last couple of years. So there's a couple of factors that are happening here. One is some of these projects are using older models that are not the current generation really capable agentic models. And beyond that, I think a lot of them were very high profile.

large scale projects. So they were trying to like do like a complete AI mo makeover, you know, go big or go home kind of thing. And those kind of projects, like even forget about AI. Like if you do any any of those kind of large projects, there's you're never really gonna see the ROI because it's, you know, maybe your people are not bought in yet, or the tech is not there yet, or you just like have, you know, got it wrong and and so on. So I think a l a lot of the studies are taking in

Vit Lyoshin (01:12.205)
Mm-hmm.

Vit Lyoshin (01:20.226)
Right.

Sid (01:32.811)
the data from those kind of projects. But if you want to do it right, AI can absolutely give you ROI if you start on a small level and you really like focus on what is it that you're trying to solve and get clear on like what is the smallest piece of work you can do to implement it, to test it, to see if it works and to see if it's if it gives you ROI before you scale it.

Vit Lyoshin (01:56.682)
Mm-hmm. I see. Yeah, that makes sense. Okay. And then another one I have for you is when people talking about AI native and becoming AI native organization, there are some misunderstandings I think. Like what do you think are some of the like misconceptions about that, how people think about it?

Sid (02:17.003)
Yeah, I I think one of the biggest misconceptions is that like AI native means no humans or large pieces of work are untouched by humans. And I think there's a there's a lot of work that is not if you become AI native, you are using AI a lot in your workflows, but you still have to validate the output, make sure it's working, and decide what direction to take your work.

So AI native doesn't mean just like handing everything over to AI and letting it figure out everything for you. It means very consciously including AI in your processes, figuring out what is worth automating with AI and what is still valuable to be done by humans.

Vit Lyoshin (03:00.679)
Mm-hmm. Okay, I see. w when people implement AI projects, sometimes people not sometimes, but many times people blame it's like not the right model, or maybe it's because we have bad data or something like that. Those two I hear the most. what's your take on this? Is there anything else in in this puzzle?

Sid (03:26.517)
Yeah, I mean I I kinda also just blame the the models a little earlier, but those were I I would say that A, if you're if you're looking at projects that were done up say a year ago, those models are just not as capable as today's models. So we saw, for example, Opus V was launched yesterday, like an hour. And you know, Fable Five is is a completely different model and completely different beast than compared to say your Sonnet four point five, which which was released six months ago, right? So I think like

Vit Lyoshin (03:42.829)
Mm-hmm.

Sid (03:55.146)
Every six months you're seeing sort of this this leap in in in in capability of the model. But beyond that, yes, bad data or incomplete data. If if you don't give the agent enough data or the right data, then it's not going to do the things you want it to do. Right. So there's a lot of plumbing work that goes into making sure these these models work well. They can't I I I feel like people, no matter how good the model is, people are like, it's just gonna like intuit what.

what I need or it's just gonna figure out on its own. But you need to give it all the the data. I think at the end of the day it's still it's still it it's an AI model and it's not it's still not like human doesn't have human intuition. at least my perspective. But I I so so those are the two kind of things. I think the data one being the most pertinent one. So if you're doing something today with the latest and greatest models, you could still have bad data in complete data and it won't give you the right answers. And I mean I I see this in my work as well. Personally I use

Vit Lyoshin (04:49.324)
Mm-hmm.

Sid (04:53.195)
I use it all day long. I'm talking to a cloud code every day, every minute of the day. And if I don't include some piece of data, it's just gonna get something wrong, right?

Vit Lyoshin (05:03.797)
Yeah, yeah. Okay, yeah, th that's that's very typical, I guess, with everything, right? If you don't give your employee enough information, they're not gonna get you the right output either. So it's the same thing here. So yeah, it make makes sense if you think about it like d yes, we call it AI, but if you think about any other process or another person, that's the same thing. Yeah.

Sid (05:16.503)
Exactly. Yeah. Yeah.

Sid (05:28.715)
Mm-hmm. Yeah.

Vit Lyoshin (05:30.411)
And then when you go in with your working with your clients and going through their projects and implementations, or maybe they had some failed implementations, what are some of the reasons for those failures that you like collected over over the time?

Sid (05:47.436)
Yeah, I think like some in some cases it's a maybe not being aligned on what you want the the project to do, right? What what is the outcome of it? So if we haven't like decided what success metrics are, then we're not gonna be able to agree when it does something whether that is a successful thing or not, right? another one I think is a project. I I'll give you an example of an Azure project. I'm not gonna name the client, but basically like

Vit Lyoshin (05:59.811)
Mm-hmm.

Sid (06:16.031)
We were trying to figure out this massive complex workflow that they have. They're an e-commerce company, but they take orders by like email and stuff like that. So they don't have like a full e-commerce setup, mostly because their customers are like old school customers, right? And so it's kind of very handshake kind of business and they send, you know, invoices and and make orders via email. So a lot of that is just like email work and various workflows. They have a very complex workflow where they have to get the

Get a logo from a customer and custom designs and send it to different manufacturers and then ship and then like get a quote and ship and all of that kind of stuff, right? there's a lot of moving pieces there, and they didn't give us the full workflow when we start working with them. So we couldn't design for every edge case, right? So they again, like back to the data thing, like the AI can do a lot of the stuff, but every time there's a new edge case where, you know, like we haven't designed for it. Now we could design.

an AI that does anything and everything, but then, you know, if you do that, you sort of lose control over the output in a way, right? And lose control over your token usage or other stuff. So there's there's different variables at play that you have to balance out.

Vit Lyoshin (07:26.121)
Mm-hmm. Mm-hmm. Okay, so yeah, it looks like you have to still go back to basics and document and standardize all your procedures before you're actually gonna implement any AI projects, if you will. Okay. Do you ever tell your clients not to build AI yet, at least? Or for shortened things?

Sid (07:34.848)
Yeah.

Sid (07:48.316)
I tell them to experiment with it. I think it's important that if you that you, as say a company that we're working with, especially in the leadership team, you have to have experienced some of the benefits of AI yourself for you to have the conviction that this is going to help the rest of your team. Because if you don't, then you know, like again, you're not gonna really be able to create something that is that works for your company. Like if you don't have that conviction, then

Vit Lyoshin (07:51.5)
Mm-hmm.

Sid (08:16.021)
You don't want to measure it. You don't want to see whether it's a success or a failure. But if you do have the conviction and you know what it can do, then you're pushing it, you're you're evangelizing it inside your company. Everyone's getting excited, they're ready to adopt it and people start adopting it, right? If people don't adopt it, it doesn't matter.

Vit Lyoshin (08:19.692)
Mm-hmm.

Vit Lyoshin (08:34.177)
Yeah, yeah. Okay. So let me shift gears a little bit about and talk about your company and what you guys do. Can you tell us a few words what's refound is about and e exactly what you do and and things like that.

Sid (08:47.883)
Yeah, of course. So refound is, I mean, the whole concept is to re you know how you found the company, but you have to refound it now in the AI age with AI at the center and and become, you know, AI native. And so that's what we help companies do. And our work is mostly work with non-technical companies. So think e-commerce companies, retailers, manufacturers, even financial services, right? Or or hospitality. And so we are

Vit Lyoshin (09:12.459)
Mm-hmm. Mm-hmm.

Sid (09:16.897)
helping them identify where the bottlenecks are in their business. Where are all the processes that are holding back the growth or costing them a lot or just friction, you know? And every business has that, right? There's like breakdowns in communication. There's like these long manual workflows that people are doing, which eat up a lot of time, that kind of stuff. And we we we figure out how to automate that by building agents that can do that work. So essentially we're basically building, we're building your AI agent, AI employees

Vit Lyoshin (09:37.602)
Mm-hmm.

Sid (09:46.711)
for your company that your current employees will work side by side with and manage over time.

Vit Lyoshin (09:53.834)
I see. Okay. So let's say if I give you like a sales or customer support department or something like that, where would you start with your like audit or investigation?

Sid (10:07.797)
Yeah, let's let's talk about the sales team. So sales sales processes are kind of similar across the board across different types of companies, right? So you have your leads that come in and then you need to qualify them, or sometimes you do cold outreach and then you book like a a an initial call, right? So there's an exploration call or a discovery call from the discovery call. You are during the call you're pitching. Then after that, maybe it goes into proposal. You have to create a proposal, send that to them, update the CRM.

And then once a proposal accepted, then you create a contract, you sign the contract, there's a deposit, and then it goes into your delivery team or your product or whatever, right? So that process is kind of similar, but along that way, there's a lot of manual work happening of making sure the CRM is up to date, making sure, like I mean, creating your contracts and proposals and other documents, making edits to it, multiple touch points and following up and so on. And and so

and then also maybe researching or prepping for for calls, right? So all of that manual work takes away from the actual work of the salesperson, which is closing sales, right? And they don't really don't many salespeople don't actually want to be the ones creating the proposals and contracts or data entry and input into Salesforce, right? So that is the kind of work that you can hand off to an AI.

Like an AI should be sitting on your calls, and the moment it you you hang up the call, it should say, Hey, here's the outcomes. I'm gonna update the CRM with so so these points, and then I'm gonna generate a proposal for this particular client. And the AI is doing that, and you're just reviewing the work and going, Yeah, that looks good. Okay, make that change, so on, right? So that's that's kind of how you work side by side with an AI where you're doing the real sales work if you're a salesperson and the AI is doing all of the rest of it. Think of it as a capable.

Vit Lyoshin (11:39.789)
Mm-hmm.

Sid (11:54.495)
assistant if you were you know, if you had one as a salesperson.

Vit Lyoshin (11:58.956)
Mm-hmm. Okay, understand. And then are there like how do you assess if something is like ready for for automation? Do you need some sort of like prerequisite work to do, like documentation kind of like we talked about, the data or anything like that?

Sid (12:16.725)
Yeah, if there is good documentation that really helps. But what we do when we start working with a company is we start with an audit. And the audit allows us to interview the people at the company and dig into the way they work and figure out what that process is. So again, most companies they don't have these documented. Some do. But most of them don't actually have anything documented. But when we talk to people and be hey, what exactly do you do? Show me step by step.

Vit Lyoshin (12:18.709)
Okay.

Vit Lyoshin (12:24.747)
Mm-hmm.

Sid (12:44.789)
That's where like the a lot of the goal comes out. They sometimes they'll share the screen or they'll walk through it. And then they'll we'll ask them, okay, how long does this take? You know, what what are you doing here? You'll find many times people are saying, Well, I take I copy this data, I paste it into the spreadsheet, then I paste it into this other app, then I download a CSV from there, then I put it here. And I'm like, all of that, all that is it's not even AI at this point. It's my sometimes it's just like automation. But but most of that work can be done by AI as well.

Vit Lyoshin (13:12.075)
Yeah. Yeah, I'm smiling because I I heard this so many times also. Even like people doing the same function, essentially, everybody's doing it slightly different. And when you get to this to trying to solve this problem, it becomes very interesting how to like change people's behaviors also at the same time. Yeah, yeah. Are there any steps that you recommend like not automate and keep the human in the loop? Can you talk about those cases a little bit?

Sid (13:21.26)
Yeah.

Sid (13:31.849)
Exactly. Yeah. Yeah. I think yeah.

Sid (13:42.018)
Yeah, for sure. I I'm a hundred percent think that there is for if we're talking about businesses, a lot of businesses are very relationship based. and so that that aspect of of the business of like face-to-face sales or customer support, you know, that kind of stuff is irreplaceable. I I think humans should be doing that. there's also like the I think the I think the cr there's there's still a lot of

creative stuff that humans can do that I I don't think AI is AI is creative. It it generates stuff, but I think like truly, truly out of the box thinking is so is still mostly done by humans. to be able to see some sort of thing that no one else is seeing, I think that's that's still something a human can do. the strategy of a company, you know, reviewing work, all of this is is still very much human work, right?

So you know, when people say AI is going to replace human workers, I don't I don't think so. I think it's going to replace people who don't do any other stuff. If you don't think strategically, if you don't, if you're not creative, if you're not, you know, maintaining relationships, then yeah, maybe then. But then again, if if you're not doing any of that, what are you doing? Like if you're just entering data, then you know, a computer can do it for you. I think you should learn those skills, those soft skills. A lot of those soft skills cannot be done by AI.

Vit Lyoshin (14:55.423)
Yeah. Okay.

Vit Lyoshin (15:00.961)
Mm-hmm. Yeah, yeah. Okay. And then also what companies are learning from your audits? Like, for example, maybe like from I'm talking from my experience trying to optimize certain processes. That's what I do as project manager for development teams. And sometimes people like leadership specifically don't realize that the team does something and they're like, why are you guys doing this? I didn't know it's happening. Are are there any like stories about figuring out some

Interesting things that they didn't know about.

Sid (15:34.008)
Absolutely. I think like as in smaller companies, not really, but in larger companies where there's sort of like a couple of layers of management between the leadership and like the frontline workers, there's often times where the leaders just don't even know what's happening on the ground or with people like the ICs, right? And so when we uncover stuff, the work that's happening, where, you know, hey, this person is entering data like multiple times in different systems or copying and pasting and you know, all of that kind of stuff, that's that's surprising to them. That's like, okay, I I didn't realize why.

Vit Lyoshin (15:47.191)
Mm-hmm.

Sid (16:02.901)
This process took so long until now. And when we uncover it for them, it's it's a new learning for them. Yeah.

Vit Lyoshin (16:08.785)
Mm-hmm. And and do you have like like maybe simple checklist so people don't have to go through your whole audit, but maybe like follow three, four, five simple steps that absolutely essential before they even decide to go to AI automation or implementation, like the basic stuff.

Sid (16:25.685)
Yeah.

Sid (16:29.579)
I think like the most basic thing right now if you're listening to this and you're at work is just to document your day-to-day. So as you do work, just like note down somewhere in your phone or computer or or a notebook, even just like, okay, right now I'm on a podcast. Now I'm writing an email. Okay, now I'm doing this. Now I'm doing that. Right. Like just everything you you did there and like what steps you took. And then at that point you can look at it and go, okay, what are the things that took too long, right? Or or were maybe

didn't require my my time, right? Like it doesn't require my time to be able to like put some data into a into HubSpot, right? So why am I doing that? And then then you can identify areas where you should be giving that to an AI.

Vit Lyoshin (17:10.551)
Mm-hmm.

Vit Lyoshin (17:14.813)
Mm-hmm. Okay, yeah. Makes sense. Document inventory and then figure out which ones can be easily automated or not. Yeah. Okay. Sure.

Sid (17:22.519)
Well, I'll give you an example. Okay. So I'm I was creating an email course for our website. So you know, like a lead magnet basically, you put your email in, you'll get an email sequence, right? Now the core of this is the content of that email sequence, which is something that I came up with. And I came up with a thing and I worked with AI to shape the content. Now, to actually you know execute the email sequence, I have to go into my email service provider and put all the copy in there in the in the emails and then

Vit Lyoshin (17:33.121)
Mm-hmm.

Sid (17:51.532)
You know, do the the like the the sub subject lines and then how many days in between and create the sequence and then connect it out to the website. All that stuff, I'm like, I don't want to do that. Because it doesn't require intelligence, right? What required creativity and intelligence was creating the actual email content. The rest of it, I gave it to my agent and I said, Hey, connect to the app that we use, which is resend in this case. They have an API, they have a CLI, connect to it, figure out.

Vit Lyoshin (17:57.665)
Mm-hmm.

Vit Lyoshin (18:03.861)
Mm-hmm.

Vit Lyoshin (18:09.089)
Mm-hmm.

Vit Lyoshin (18:14.242)
Mm-hmm.

Sid (18:19.265)
how to create the full email sequence and then figure out how to connect it to our website. That's it. The agent went and did it. Right. And so all of that stuff that probably would have taken me like it's manual work, but sure, I could have done it in maybe fifteen minutes, twenty minutes if I was really focused. I don't really want to be doing it. I gave it to the AI, the AI did it in twenty seconds.

Vit Lyoshin (18:24.011)
Mm-hmm.

Yeah.

Vit Lyoshin (18:38.421)
Yeah, exact yeah. Sounds very familiar. Yeah. Okay. so let's talk about AI native organizations and like can we define it and also how does decision making work in AI native versus normal organization?

Sid (18:41.611)
Yeah. Yeah.

Sid (18:56.341)
Yeah. I'd say i the definition, I don't know if I can give you like a very succinct, repeatable one, but I would say like an alienative organization is one where

Sid (19:10.955)
AI is doing all the manual and repetitive works work that your employees would normally be doing. So the example I just gave you, you know, the the the work of creating an email course and launching it, there's two pieces to it. One is the creative and strategic thinking of what is the content of the course. The second is the operations work of setting it up and connecting systems and the automations. And that second part of your work.

should be automated by AI. And so everyone's role has those two pieces to it the strategic, creative, high level work, and then the low level operations work. And I think in an AI native organization, every employee has figured out, figured that out and is able to, and AI agents are automating that part of it. The other thing is there's between within companies itself, there are communication breakdowns, communication gaps.

Product team isn't informing the engineering team about what's coming up, engineering team isn't telling the sales team what the capabilities are, all of that kind of stuff, right? And in an AI native organization, all of that is is handled because the AI has full visibility into what's happening across the board and is keeping everyone in this in the loop. Right. And so if you are not sure about something,

Vit Lyoshin (20:16.301)
Mm-hmm.

Sid (20:37.653)
And you have to go to human to ask them or or wait a few days to find out, that's not an AI native organization. But if you can ask your AI agent and can tell you exactly what is happening across the company, you know, keeping your your security permissions in in mind, then you're an AI native organization.

Vit Lyoshin (20:47.562)
Mm-hmm.

Vit Lyoshin (20:54.509)
Okay, so how does exactly this change in like people's decision making and their roles and day-to-day jobs, like should should they spend more time on just making decisions, creativity, and then have some AI agents to to to do the execution or not execution but rather like configuration stuff and all of that. So how does it look really in reality?

Sid (21:23.831)
That's exactly how it would look, right? It like for for for me, for example, if I the way I worked a year ago has changed completely to to the way I'm working today. Because the way I work today is mostly me coming up with ideas and you know, giving it some shape and then the agent executing it for me. And so all I'm doing is just like, okay, I have a new idea for an email campaign. Here's what I want, here are what the emails should look like, and I tell it to the agent and the agent sets it up, right?

Vit Lyoshin (21:26.433)
Okay.

Vit Lyoshin (21:41.314)
Mm-hmm.

Vit Lyoshin (21:53.034)
Uh-huh.

Sid (21:53.29)
I want to create a new sales outreach motion. I want to you know, create this new agent for this client, right? And so all of that kind of work is me thinking strategically and creatively and then having the agent execute. And I think that's how other employees should be thinking about it as well. What part of their job requires the creative stri strategy, decision making or or human you know, relationship building work?

That is something that they should be doing and then everything else around that is AI.

Vit Lyoshin (22:26.769)
Mm-hmm. Yeah, I I think I heard this in in your other conversation with somebody that where you said like we used to do we used to ask AI to give us some instructions what to do, but now we just tell it basically, this is what I want, and you go do it.

Sid (22:43.585)
Well, exactly. Yeah. I think like and and and you know, people still use it that way, right? So when you chat when you talk to Chat GPT and you Hey, what should I do? If you think about it, the AI is telling you as a human what you should be doing and you're doing the work. Whereas it should be the other way around. You should just tell the AI, Hey, I want you to do this, this and this, go do it for me, right? And then it goes and figures out how to do it. Yeah.

Vit Lyoshin (22:48.044)
Yeah.

Yeah.

Vit Lyoshin (23:01.248)
Yeah.

Yeah yeah. How you go how you go about like training people, upskilling people in organizations so they can actually operate alongside with these agents and and actually be productive.

Sid (23:19.701)
Yeah, I think a the way we train and upskill is a little different. It's it's not pure content stuff, right? It's kind of more like show and tell. Cause I can I can explain to you theoretically what an agent is and how it works and stuff. And that's not gonna like it's such a new concept for so many people that it just doesn't really matter. Like you can read about this all day long or watch videos. But if I show you on my computer, if I open up Cloud Core or Cloud Cowork or whatever, right? And I say, okay, like here

Vit Lyoshin (23:41.325)
Mm-hmm.

Sid (23:48.278)
I'm connecting it to this app. Now I'm telling it what to do and it's going and doing it. That's when that clicks for you, right? Then you're like, I get it now. And so what we'll do is when we train people and upskill them is we'll we'll make it very custom to the work that they do. So based on whatever you do as a work, as a marketer, as a salesperson, accountant, whatever, we we'll come and sit next to you and we'll go, okay, here's the accounting work you're doing. You're like downloading data from Shopify and from Amazon, then you're like combining it into a spreadsheet, then you're doing this.

Let's have Cloud connect to those APIs and we're going to set it up for you here. And now if you tell Cloud what to do, it'll do it for you. Right. And then they see it happen within seconds and they they see the full final result within seconds, all the reconciled the inventory or whatever, right? The sales. Then that that blows your mind. Then we say, okay, now let's turn this into a skill so that you don't have to go through the same instructions over and over again. You could just type the skill out and call the skill and it just runs the whole thing. And then then they go, okay, now like.

It just all makes sense. So now your go your day goes from you doing all the work to you coming in, typing out a bunch of skills or calling a bunch of skills. The AI is showing you the work, you're just reviewing it and you're good. Right? And so then you maybe, you know, you've you finish work early or you just you take on more work. I don't know. Up to you.

Vit Lyoshin (24:58.679)
Uh-huh.

Vit Lyoshin (25:05.069)
Yeah, probably more work because that's really the productivity change. I find myself doing the same thing, like I'm having like a couple of different open windows and chatting and telling them what to do. Do this here, go update there. and and those things just go and do all of that stuff. And I sometimes find myself like I can't even think fast enough for them to to like tell them.

Sid (25:28.533)
I I Yeah, a hundred percent. I I I I I I feel like I've been I'm doing way more work now than than previously.

Vit Lyoshin (25:38.764)
Yeah, exactly. That's what it feels like. And actually in the evenings I'm more tired because the brain is always like thinking and giving instructions, giving more instructions. so yeah, it's like no time for break. Yeah, yeah, okay. so are there any sort of problems or solutions that agents are like they not fit for, you don't recommend using it for?

Sid (25:51.701)
No no time for breaks, man. Yeah, yeah.

Vit Lyoshin (26:08.359)
and and people like should stay away

Sid (26:17.099)
I I again I I I think I think maybe like just still at you know the creative and strategic work, right? Like I number one is I just I don't think people should like hand over completely. If you're doing marketing, for example, let the AI just write your blog post for you or your emails for you. I still think that humans should be involved in that. And it maybe it's just like my my kind of I don't know, me just like my standards or something, but I when I

Vit Lyoshin (26:33.185)
Uh-huh.

Sid (26:44.821)
I can tell. I can tell so easily when it's AI generated content. Right. And then it just to me it it feels like you haven't put any thought into what you're trying to do or say. Right. And that's just like one example. Like, you know, like if you are if you are working at a company or building a company or managing people, I still I I think that that work is still important for you to do, right? In terms of people, people relationships, managing people, talking to them, your sales work, talking to customers, or as a customer support person, understanding what the customers' problems are.

As a marketer understanding what your who your target market is, as a leader understanding like or like deciding the strategy and direction of the company. That is not stuff that you should be handing over to AI. I maybe AI can do it, sure, in some cases, but then at that point, what is your purpose, right? So I I I truly think that like you should maintain that because that's what gives you your edge as a business or or an employee versus

Vit Lyoshin (27:42.124)
Uh-huh.

Sid (27:42.711)
you know, just handing everything over to AI. Just don't don't let the AI think for you, basically, you know?

Vit Lyoshin (27:47.276)
Yeah, okay, okay. And I I have some people who one way or another involved with startups and in startups usually it's not enough time to do everything and people struggling and things like that. So a question for you, like if you were starting a startup tomorrow, what sort of processes would you let agents run or help with rather than like humans focusing on most important things and AI handles

Like more operational, busy work, day-to-day stuff. What would be some examples?

Sid (28:22.987)
Yeah. It's a good question. I'll give you examples of like our own company because I guess at in some way we're a startup too, right? Like we we we started, although I have like many, many years in of experience in AI, like maybe eight or nine years, I think and I've I've done worked at a lot of different companies. The we started refound fairly recently, like in the last year or so. And we've grown a lot, but basically I think when I was starting off, I was

Vit Lyoshin (28:30.273)
Okay.

Sid (28:52.981)
The first thing I did was like build my own AI assistant. So I created my own version of Jarvis, which I I actually call it Jarvis, but I created my own version of Jarvis, which is essentially it it helps me plan my day, right? So it helps me keep me focused and organized. And I I was sort of like, okay, here's here's the business we're starting, here's what we're trying to do, like help me work through our ICP and our our our business model, right?

Then I was like, okay, now let's connect to my email. Now I needed to start managing some of my email and my client communications via email. Then I gave access to my calendar. Now start managing my calendar, help me book meetings or you know, help me you know, prepare for meetings. And then I connected my Google Drive and say, now, okay, now let's manage all my documents in Google Drive, all our contracts and proposals and stuff. All of that work slowly over time was became, you know, me doing that kind of stuff, just like handing it over to the AI agent to say.

All right. I've got a call coming up with a potential client. Prepare me for it. So it prepares me. I get another call with a client. I do the call and then I pitch the client. At the end of the call, it takes a transcript, turns that into proposal, and then it sends it from my email. I review it before it sends. Then the client comes back and says says, Yes, this is good. It then creates a contract and sends it to them. I review before it sends. And then it creates an invoice in our accounting software and sends them the invoice as a deposit. And yeah.

Vit Lyoshin (30:05.655)
Mm-hmm.

Sid (30:18.591)
And so then it's tracking everything. And over time, I add more and more tools and more and more skills. And it starts to take over a lot of that work. So my work is prep for five minutes, get on a call, and then that's it. I'm done. I prep the next thing while the agent does the rest of it and review the work that the agent does. And then that that's just like the one piece of it, right? So I love that's the sales and then some marketing and the accounting. And then you know,

Vit Lyoshin (30:35.308)
Uh-huh.

Yeah. I

Sid (30:46.017)
Hiring people is like helping me create employee records and managing employee salaries and then the same even building building work for for client, right? Okay, based on our call, can you generate a spec for this? Let's okay, then I review the spec. Okay, now turn the spec into tickets so that our engineers can start to build it out and then set up our infrastructure and connect our database and you know, so on and so forth and test it and and and so on. So

A lot of that work again gets done by AI agents.

Vit Lyoshin (31:16.971)
Yeah. Yeah. And I'm listening and smiling here because it it we used to have the whole departments to do this stuff and like the whole CI C D and and and how engineers work and how they pro that's what I'm most familiar with. That's what I deal with every day. And now everybody essentially can have an agent who runs this and figures it out and you you just tell it, go read the documentation and figure it out and set it up and you wake up and it's ready to go. it's amazing. Yeah.

Sid (31:47.287)
There you go.

Vit Lyoshin (31:47.822)
Yeah. so let's talk a little bit in the future, like five, ten years from now. which of some of the business processes that you think will disappear? or build not disappear, but like replaced by AI, I should say.

Sid (32:04.189)
I'm gonna get flamed this. If okay. I so again, I'm I'm still of the opinion that it doesn't fully replace humans and especially if there's a lot of your work is is human bas relationship based strategy work, creator work. That being said, there's still a lot of like roles that will disappear or or departments that go away are fully automated because it doesn't require a lot of that. And so I think maybe these departments will shrink in size.

Vit Lyoshin (32:07.531)
Ha ha ha.

Vit Lyoshin (32:12.823)
Sure.

Vit Lyoshin (32:26.252)
Mm-hmm.

Sid (32:30.839)
First of all, we're already seeing a lot of automation in engineering itself. Like software development is nearly completely automated. In a lot of the most tech forward companies and AI native companies, the entire engineering and coding work is fully automated. But the engineers are still very important piece of that work because they're the ones reviewing it, they're the ones deciding the direction it should go. They're the the architecture, all of that, right? So you know, like I i that that that's not

Vit Lyoshin (32:51.148)
Uh-huh.

Sid (32:58.261)
That's fully automated, but it it hasn't gone away as a department. It is it is transformed. On the other hand, say accounting, where it's a lot of just like collecting data and reconciling data. That's just like I mean, there's not even AI at that point, is mostly just like scripts, right? You can create Python scripts to handle the data. The AI can generate the Python scripts to do that based on the shape of your work. So I think that is those that kind of work where you are pulling together a lot of data and trying to reconcile them or match them up and match up codes.

Vit Lyoshin (33:18.573)
Mm-hmm.

Sid (33:27.915)
your skew codes with s with something else, gone. Yeah, I can do it. Right. I think any kind of like operations work of

you know, working with different systems and and and and pulling data from one system and putting in another system. Or you know, like the admin kind of work of like emails and calendars and calendar management, document management, also gone. AI agents will be doing all of that, if not most of it. yeah, that that's that's kinda I I I basically, you know, the operations heavy businesses that that drowning in a lot of repetitive work for them, I think

It's there's a lot of scope for AI to help and I it's gonna benefit them in general.

Vit Lyoshin (34:14.049)
Mm-hmm.

So what I take from this is mostly operations heavy businesses will benefit the most most of it and they can invest more in other things or I think I I read this recently about IKEA, I think, that they actually didn't lay off anybody because it this is a crazy time right now. Everybody's laying off people for one reason or another. Most of it still is from over hiring over the pandemic period.

But everybody linking not the most people linking it to AI also, right? And this is a scary part for more for many people. but I think there are companies, and I think it was IKEA, maybe somebody else, I don't remember, that actually did a lot of AI projects, a lot of transm transformation in this area, but they keep kept everybody employed and they actually like shifted their roles or maybe gave them some other responsibilities or whatnot.

And it it works. People just I guess will have to wait a little bit and give it time and companies will start hiring again. There's not gonna be there's gonna be more work than less with AI because it's gonna produce so much more and somebody will have to still help with it, right?

Sid (35:35.128)
Yeah, a hundred percent. I think like 'cause it's it's doing so much that you have to review all of this stuff. And so it it ends up being more work. And also, like, you know, just because I created one campaign, let's say as a marketer, for example, I used to be a marketer, so this is a good example for me to give. I can relate where maybe it would take me, say, a day to set up a marketing or an ad campaign or a new ad, but now I can do more, right? So instead of a day, it takes me half a day, l like a or a couple of hours.

Vit Lyoshin (35:40.918)
Yeah.

Vit Lyoshin (35:51.084)
Uh-huh.

Sid (36:02.485)
Now I have the rest of my time to do other stuff, maybe multiple campaigns. I can scale the number of campaigns. I'm still working the full eight hours of the day, but now I've like put out four marketing campaigns instead of one, right?

Vit Lyoshin (36:12.735)
Right, yeah, yeah, right. Or you can put more thinking into this and do more experiments with the same one and figure out. Yeah, yeah, absolutely. Okay. do you have any sort of predictions about AI agents in terms of like business processes or adoption and things like that? A anything that you personally think.

Sid (36:38.357)
Yeah, I think I think, you know, one of my predictions is that it an AI agent is gonna become a very core integral part of a business's arg chart. There's gonna be like maybe I I you know, I've seen people try to define this and give give the agent name. I don't know if they'll it'll go that far, but there's definitely gonna be like, hey, this is the agent that handles all of these things in the business, right? So this is like an AI coworker and it's part of the arg chart.

And so if you need anything, you go to this agent. Or if you need anything here, you go to that agent, right? Like there's gonna be maybe at a smaller business, possibly just the one central agent, but like as a larger business, there's gonna be multiple different agents communicating with each other and communicating with the humans. And those are gonna be part of that art jar.

Vit Lyoshin (37:24.733)
Mm-hmm. Interesting. Yeah, I actually give names too, but I stole this idea from somebody. So it's just somehow easier to to remember who is who, I I guess. Yeah, yeah. Okay. so another question I have for you is what's your advice, recommendation for somebody who's listening and maybe their organization is moving towards AI or they wanna like experiment themselves.

Sid (37:34.431)
Yeah. Yeah, yeah, exactly. Yeah, yeah.

Vit Lyoshin (37:53.762)
What's your advice like what to learn? Maybe some skills, maybe some technologies, some resources to go to and learn more about it?

Sid (38:02.252)
Yeah.

I think the number one thing is the best way to learn about this is to just get your hands dirty and play with it. And so you know, use OpenAI's co work, open codecs work, chat GPT work, whatever they call it now. It's like the naming is so confusing. And then Claude's version is called cowork. yeah. So try try that. That is a very easy, beginner friendly way to get into agents. And once you're familiar, we you can get into like say Cloud Code or Codex in the terminal.

Vit Lyoshin (38:10.061)
Uh-huh.

Vit Lyoshin (38:20.235)
Yeah, yeah, work, yeah.

Sid (38:34.933)
Which is what I use on a daily basis. And just don't don't be afraid of like break things, right? Just experiment, try, giving, give it access to some tools and just ask it to do stuff, just like crazy things, right? Like see what it can do, see where what it can't do, right? And figure out where where things are breaking, and then you know what is the boundary between the work you need to do and what what the AI agent does and how it works. And then once you start to like figure out what it is it can do, then

Vit Lyoshin (38:58.829)
Mm-hmm.

Sid (39:02.891)
You know, start documenting your own work and and look at say, okay, fine, I think the AI agent can do this piece of my work. You know, what is it that you don't like doing in in your day-to-day life? See if the AI agent can do that. And then, you know, if you want to automate that, maybe turn it into a repeatable process, create a skill or create some sort of like remote agent that runs automatically, maybe every eight AM every morning, you know? That's that's kind of how you like step by step start to

Play with it, experiment with it, learn how it works, and then actually get benefit from it. And if you want some, if you want resources, I mean I have a I've written a blog, it's a very popular blog on my website, sidbarath.com. And that's my personal website and my personal blog where I just have written a lot about I haven't written in the last couple of months because I've been busy with work, but prior to that was mostly me when I was experimenting with things and trying things out as new models came. I was playing with them and and testing it and and writing my opinion on it. so go check that out.

Vit Lyoshin (39:40.823)
Mm-hmm.

Sid (39:59.64)
And yeah, just the best way is just to like learn on your own. Hands on.

Vit Lyoshin (40:05.271)
Hands-on, yes, hands-on. Okay, great. Thank you, Sid, very much for information. It's great to hear your opinions and your expertise. you've tried this yourself in the field, so that's good to learn from you. but at the end I usually ask three of my QA innovation QA questions. So you if you don't mind. the first one is can you define innovation in a few words?

Sid (40:35.115)
I think innovation is

Sid (40:41.131)
Finding new solutions to problems that didn't have solutions or had

unoptimized are you know incomplete solutions before. So basically solving in a a a problem in a way that made made the old way look look silly in hindsight.

Vit Lyoshin (41:05.841)
Mm-hmm. Okay, good. which innovation you think in human history changed the world the most?

Sid (41:16.123)
it's so it it's you know that this is one is like you could go all the way back to say the wheel or fire, which you know, 'cause like one little thing just changes the trajectory of a of but I I think like in the modern world, it's it's the the transistor, right? Compute. Because everything is downstream of compute. Everything today is compute, right? You're like how do you how do AI agents run or AI language models inference? It's all it's all compute, right? So being able to turn

Vit Lyoshin (41:22.72)
Yeah.

Vit Lyoshin (41:30.605)
Mm-hmm.

Vit Lyoshin (41:40.802)
Mm-hmm.

Sid (41:44.969)
silicon or sand into something that can compute and think. That is, I think, something that's changed the world in the most recent years.

Vit Lyoshin (41:54.135)
Yeah, okay. And the final is which tool or technology or software or something that we use today you think we will be laughing at ten years from now?

Sid (42:06.455)
I mean, I think just the entire notion of having to open up an app or a software, especially in in the context of business where op you have to open up your HubSpot and Typhon and MailChimp or whatever it is, you know, like these different user interfaces, which are really just UIs on top of data. Like that would be I think that would be crazy because now I don't actually open any apps anymore. So and I think that that's obviously I'm at the very forefront of what's happening.

Vit Lyoshin (42:26.465)
Mm-hmm.

Sid (42:34.815)
Most people haven't caught up. But when they do catch catch catch up in a couple of years, you know, we'll look back in say five, ten years and go like, I can't believe I opened up multiple softwares at the same time because now your agent is just doing all of it, right?

Vit Lyoshin (42:38.954)
Mm-hmm.

Vit Lyoshin (42:47.637)
Okay, interesting. looking forward to that. Yeah. All right. Thank you very much, Sid, for your time. It's been a pleasure. Hopefully we can stay in touch. And we'll talk again soon.

Sid (42:51.243)
Yeah, yeah.

Sid (42:59.969)
For sure, V. Thank you so much for having me.

Vit Lyoshin (43:02.37)
Yeah, absolutely. Bye.

Sid (43:04.684)
Bye.