Demystifying GenAI: A Founder's Primer with Vijay Rayapati

‍

In our latest episode of Founder Moments, listen in from Vijay Rayapati, Co-founder and CEO at Atomicwork, on why every founder should be thinking of GenAI, how AI can be leveraged within organisations and his predictions for the future.

‍

Vijay:

if you really want to build AI first AI lead team, AI core kind of software product company most of your employees should use AI so that they can actually realize the power, acknowledge it and also adopt it.

Pranay:

Hey, everyone. Welcome to our quick Matrix Moments podcast.
I have Vijay of Atomicwork with me here, so Vijay is the founder and CEO of Atomic which is a company which is reinventing HR and IT workflows and creating a better employee experience with AI at the core. I think Vijay just made a trip to the US and came back so we thought we’ll chat about demystifying AI, what’s hype, what’s real and how founders and business builders should be thinking about implementing AI either in internal operations or in their product. And so I think with that, Vijay, first hype or real, how’s it feeling?

Vijay:

So, I believe it’s real. The reason I say it looks like a real trend because lot of us when we use these new tools including ChatGPT are finding immediate value. So the way at least I use the judgment is if something can provide immediate value there is a potential for it to provide durable value. And unlike Clubhouse which most of us have signed up, used once or twice then we never went back but in last few months lot of us are going back to ChatGPT again and again and again and getting super excited about LLMs. So I believe it’s a real and durable trend.

Pranay:

Awesome. And can you talk about – I know we discussed this before but about the adoption of AI tooling within Atomicwork and what’s been the experience with that and across engineering, product, marketing.

Vijay:

So I think one of the things that internally we find if we really want to build a product with AI core first we need to become power users of AI ourselves across the team, across our product, engineering, marketing, even like the one member go to market team. So the first thing we adopted was Co-pilot for our engineering team. As expected there was a resistance so this is like what we started adopting I think early Feb. You know, saw a video of Satya talking about Andrej Karpathy using Co-pilot when he did like programming for Tesla AI and afterwards as well so that really inspired me to push the team saying why don’t we try it now. There was some resistance but as people found value, the utility value, and I think there is a durable value and everybody in the team including the CTO and those of you who know our CTO, Kiran, he generally has a very high bar for adopting technology. So that was the first change.

Then there was a ChatGPT moment, you know, I was trying to use it then one of our PMs were trying to use it. I wrote a first investor e-mail, I think when we just came out that first weekend after it came out and I sent that e-mail I said this e-mail was formatted by ChatGPT and that actually encouraged like bunch of other people in our team also because the team gets to copy down that e-mail to explore it. Of course some people like didn’t go back to it but a whole bunch of us including myself and our PMs like Shankar, Ramesh and marketing, Sai and all we started using it more and more. Then most of us also bought pro subscriptions, I told everyone if you're in a PM team or marketing team you can buy the pro, the company will pay for it. I think there is a real value in it, it can actually make you super productive. We are now using tools, you know, to essentially find some prospect intelligence, the core thesis is anyway not about how we used but the philosophy that if you really want to build AI first AI lead team, AI core kind of software product company most of your employees should use AI so that they can actually realize the power, acknowledge it and also adopt it. So that's essential in the thesis.

Pranay:

Got it. And I think two schools of thought, right, one is people saying that AI as an incremental Co-pilot and I guess over there it’s like GitHub has a Co-pilot, Notion has a AI writer so everyone’s adding this on, obviously it makes people more productive but incumbents will still capture most of that value because it’s AWS for compute and then you have Notions and Microsofts of the world who are minting money from it. The other school of thought which is maximalist is probably that hey, LLM is the new backend and Chat is the new frontend which also seems like a,  I don’t know how true that is. The answer probably lies in the middle. So just wondering again from your visit like how do you see founders thinking about it in India, founders and sort of experts thinking about things in the US and how should founders building in domain think about it.

Vijay:

I think great question, so I'm not obviously an expert so I just become prompt engineers like most of you, in last few months but I think I learnt two things. So when I went to US in March most of March I was there and I met like lot of people. Everybody was asking how are you thinking your product and product evolution with it’s all the excitement around AI. So these are like not just startup people, people who are CEO of a company that is ready to go IPO to – even like my friend Akshay of Notion he asked like how much of your roadmap has changed because of AI. And then I went and met like veterans like Bill Moran who was a partner at Sequoia and has seen things from programming languages to early web to mobile to cloud to now AI. And so this is a joke I often tell about my founder friends, when young people pitch a trend maybe you can ignore it but if old people pitch it then you truly have to give attention especially because these people have seen like decades of changes. And ML has been like a pipe dream for most of us for like last 10-15 even 20 years. Yeah, we did like some cool things here and there like the classifications, clustering, etc. But with AI suddenly it feels like the value addition has gone 10x. Even when you use like a ChatGPT or AgentGPT or AutoGPT or like any of these cool Chrome plug ins that you can use for generating or what not I think two things I felt. I basically felt before going to US I was also thinking LLM is just a layer, we plug it into our product. But after I came back, I basically said no, it has to be at the core. The reason why I say it has to be in the core so what are the core things for a product if you look at any product. You have most products have this cold start problem which is there is a configuration, integration, setup, you know, where we either put lot of software automation on top of our product or through people to solve around these problems. I see we can actually use this assistance to help people with a human layer to make that super productive and reduce our friction. And second thing is most enterprise software products have very clunky configurations, very clunky admin management things which we can actually make very delightful using AI.

Third thing I believe if you're building content generations whether you're building forms like you can do like not just lot of intelligent defaults you can actually see lot of seeding work for your end user in using AI. So at least my perception has changed, our team’s perception has changed, now we’re saying, okay, AI has to be become core. If we think that way at least we’ll make AI as a strong layer or a strong component of our product. If we think of AI as a layer then you might only think okay I can use this for as a virtual assistant or maybe I can use this for some summarization or some answering kind of thing. One of the reasons I asked everybody in the company to start adopting and move and also forcing on them is because I want people to realize. When they use a lot of these tools sometimes as a end user of these tools you realize how powerful this can get and how we can use in that product. That’s the reason why I said first you have to become if you're a startup of five people, ten people, three people, it doesn’t matter, you have to be the power users so you can actually realize, acknowledge or write off. You say okay, I have used, I don’t believe, that's my conviction, that's not for you.

I think in US whenever I met people everybody is deeply thinking about making it a strong part or a core of the company and as I said even my own thinking before I was there for 4-5 weeks was think of it like a layer. Because you add it as a layer everybody can add it as a layer, then there is no differentiation for you, right. And but I also believe this can't become a layer because of SaaS experience because all entre folks thought if I just partition my database, add APIs I can become a SaaS software, it didn’t happen. So all like this mainframe kind of folks thought if I put a server client then I’ll become – all these server client folks thought if I just start like a player I’ll become like a web album. All the web guys thought if I integrate I can become a mobile app company, we all have seen this at least some of us who have been there for like last 20 years or so in that. The desktop companies did become web companies the web companies are not the big winners in mobile, the mobile companies were not the big winners in cloud and now we’re seeing with open AI the big cloud vendors are not the big winners in AI. My suggestion is experiment, first use it for personal use cases now like lot of tools and then decide for use. But if you ask me personally would I bet if I have a 100 bucks how much I would bet on AI I would bet probably a 50-60 bucks on it.

Pranay:

Got it. And obviously right now a lot of reapplications are consumer facing where hits and misses are okay. Right, but in a B2B workflow like yours you cannot afford to have those misses, they can be expensive and very costly for you, so what’s been the experience of implementing some of these LLMs in the product itself?

Vijay:

So I think great question, I think LLMs I think all of us know they do a lot of hallucination, they can make up things which are not real which is great even if you're writing a book or storytelling which is not great if you're giving customer support response. Right, because you can't make up things. So I think that's why I use this word seeding, you know, you can use AI for seeding work lets still your end users do the refinement before they deploy. Maybe it will take some time for us to truly allow them to do the full work, right, so the way I'm still thinking is this year and even next year we can actually use it to seed a lot of work and also like large language models the reason they’re so good and also the reason they’re so bad is because they’re so large. You also have to build a specific language model for your use case so you can reduce this.

Pranay:

Small language.

Vijay:

Small language models, domain language models, specific language models, SLM TLM and whatever you want to call it can reduce this hallucination. Now but seeding the work with human skill acts – when I say human your end user or your admin or your operator can review, either discard, start from scratch or start on top of it, itself could be a huge value. So that's why I say like if I had a 100 bucks I won't put 100 on it, the reason is we’re still not there. Right, so we’re still in like a huge hype cycle and on Crypto also like I mean personally I try to use using bought tokens and what not but very hard for you to find immediate value. With cloud you put large infrastructure, deploy your sale code you found value. So now with AI you're seeing that, you can ask it to summarize something, format something, create something. You know, it’s early days but tech can get so better and the joke I tell people is when mobile revolution started it took a year for mobile companies to update stuff. When cloud started, you know, AWS, Google, Microsoft, all these folks were shaping stuff every quarter. Now in the AI if you just notice what happened in last four weeks or eight weeks and twelve weeks these people have reduced that from a quarter to weeks, right, which obviously shows the progression has been like superfast. My first advice again is go use bunch of tools -- if you're a designer use Midjourney, Adobe, Fivefly, Canva, Figma is when people are writing plug ins ChatGPT plug ins right now on Figma. If you're in product manager you must use ChatGPT, AgentGPT kind of things. If you're a CEO there is no reason that you can avoid skipping using GPT for your PMs and founding teams.

Pranay:

Can you share examples of how PMs are using it?

Vijay:

So very simple, and I shared this screen shot even with the personnel. We simply wanted to find taxonomy for how companies create department names. What is the traditional thing you go look at bunch of competitors, bunch of companies, then you come up with whatever like ten departments. Right, these departments think of it like engineering, product, customer support, G&A, legal etc. For ChatGPT this is just one prompt, you can do this whole damn thing in five minutes then you spend half an hour refining it so which can reduce the two hours of search work. The beautiful thing about transformers is they can reduce search work, some aggregation work, you still need to do refinement work. In some companies it could be called engineering, some companies it could be called R&D. In very big companies the department is called R&D.

Pranay:

Or finally IT.

Vijay:

IT. You know, IT will come under G&A. So your IT and your legal like lot of these things come under G&A. But in a few thousand people company they might call it engineering, some companies they might call it product engineering. Now you decide should you allow customization for department or not but the seeding work could be leveraged. Two, we’re using it to write internal e-mails, external e-mails, format them. Three, we’re using it for SEO refinement of our content stuff because and I'm sharing this on my Twitter as well, ChatGPT prompts for a SEO. You know, the engineering team using Co-pilot will write lot of spaghetti code, plumbing code, skeleton code which makes them super productive. I think there are like lot of use cases which have emerged within very short time. I think nobody touched Co-pilot until ChatGPT. It is like I heard about Co-pilot more than a year ago.

Pranay:

But ChatGPT was the marketing moment.

Vijay:

ChatGPT was that moment where you really felt and even when you talk to other founders, other CEOs of state companies if they’re using it it’s a crime if you don’t use it. You know, because productivity for you is like matters a lot more than like a super big company with a super big customer base.

Pranay:

And you said you’ll bet 50 or 60 dollars out of 100 on this.

Vijay:

Yeah.

Pranay:

So prediction time,, so what do you expect that comes up in the next one or two years. Obviously the pace of shaping is phenomenal and the pace of innovation is crazy but where do you think this gets us to in terms of product?

Vijay:

Okay, so I'm not an expert as I said I just became a prompt engineer but I believe three things that could potentially happen. This is also like lot of learning and reading that I have done and heard from people and debated with people etc I think today we’re still seeing AI mostly for human elimination. As a PM I'm using, as a designer I'm using, as an engineering I'm using so I'm getting more productive. Hopefully in few years we can get to an operating leverage. You know, your 10 engineers probably will be able to do 15 engineers’ work so there’s a big operating leverage.

Pranay:

And what’s the difference between the two?

Vijay:

So the difference is today we’re still looking at it from an individual lens saying how AI can help my end user and how AI can help my engineer, tomorrow you’ll look at it from a lens of how AI can help my company, that's the operating leverage. Then you can even deliver business leverage as I said. Now you take a SaaS company, you know, SaaS is a high tech. We spend enormous amount of capital on product engineering, customer support, customer success. I think in these three areas I genuinely think that it can deliver even a business leverage for companies that adopt it can deliver a business leverage. Why I say it can deliver business leverage is as I said like companies have cold start problem, activation problem, adoption problem which can make it better and better using less people which essentially means probably -- this is my dream, I don’t know whether it is going to happen or not. We all had this wow moment when Whatsapp acquisition happened. Very small team built such a great product, scaled product, but we didn’t see like hundreds of examples like that.

So recently when Figma acquisition happened we again all felt wow moment because the scale of acquisition but look at the team size. This is not like a 10,000 people company, you know, it is like less than 1000 people company built such a phenomenal product and go to market and revenue I now believe like last decade you would have probably needed 10,000 people even if you're a tech company to get to billion dollar AI. Hopefully, you know, this is a lot of hypothesis, ifs and buts, but I genuinely believe this thing is we all found immediate and intermediate value, we’re seeing some durable value. If it can deliver scalable value, right, AI as in tech hopefully lot of us can build companies with less than 1000 people for like thousands of enterprise customers, make hundreds of millions of dollars of revenue without applying lot - See this doesn’t mean now we reduce employment, right, or like okay, people don’t have jobs. People can go do lot more work and I still remember my school -- this is a joke I told in one of the panel. When I was in school bank employees in India went on a strike when the government decided that they’ll deploy computers because the bank employees thought the computers were taking their jobs away. I mean it’s actually a real thing, you can go search. But what people didn’t realize is a bank which was operating with 1000 branches then can get to like 20,000 branches by using technology. That's what helped, right, that's why banks have like now you go to like lot of modern banks and if you go to even Kotak in a tier 2 town there are less than 10 people in the branch, they’re using and serving probably few lakh customers. So I genuinely believe humans will always have jobs whether that job is seeding job, refinement job, editor job, procure job of decision maker job we don’t know but I think we’ll create tremendous opportunity. And that's why I said I’ll probably bet like 50-60 bucks, learn, don’t ignore it and trends are always shaped by humans, they’re always dropped by humans. It’s great if you drop as humans you know okay, this is not the trend. There are like a lot of ethical concerns, safety concerns. Those are all right things but again I give an example of when telephone came lot of companies said we can't put this shit in our company because what if our employees call and tell our competitors all the secrets of how we’re building things, doing R&D works etc. Same thing people had with e-mails, right, even in tech when our services companies didn’t even used to give laptops to people because they thought that can actually cause problems but now taking work laptop home is such a common thing.

Pranay:

And maybe just a flip or a contra view but obviously so much of the hype is going on, what advice would you have founders to that's maybe the opposite end of the argument.

Vijay:

I think we have seen whatever people call it learning curve effect so right now we’re in the peak of effect. We will all have some disillusionment then start from there. So that disillusionment can happen because look at the way we’re giving benefit of doubt to ChatGPT versus we didn’t give benefit of doubt to Google or when Ping came up in their previous model. Because when big companies launch these things we expect them to be perfect, when startups or small companies launch these things we’re okay even with the imperfections, that's how things get better. So that's another reason why I also said I will not put 100 percent, I mean like if I have 100 bucks I wouldn’t put 100 on AI, I would still encourage  to think deeply about it. Use it, then think about how you want to shape your road map or your product and you don’t ignore it was my first feedback. Why I'm saying don’t ignore it is this is not a block chain where you don’t know what the use case is, you never had a immediate utility value, there is a utility value, we’re seeing some early signs of durable value. What we don’t know is whether it’s scalable and sustainable. But there is nothing wrong, you know, as a startup you can move fast and you can do a lot of good things in few weeks. So you should take some bet, no bet is probably a bad bet on this one.

Pranay:

Got it. Awesome. Thanks so much, Vijay.

Vijay:

Thank you, Pranay.

‍