Transcript
Surojit Chatterjee [00:00:00]:I think SaaS will be dead. I have been saying that...actually, I would say, before anybody else.
Daniel Darling [00:00:15]:Welcome to the Five Year Frontier Podcast, a preview of the future through the eyes of the innovators shaping our world. Through short insight-packed discussions, I seek to bring you a glimpse of what a key industry could look like five years out. I'm your host Daniel Darlingfx, Venture Capitalist at Focal, where I spend my days with founders at the very start of their journey to transform an industry. The best have a distinct vision of what's to come, a guiding North Star they're building towards, and that's what I'm here to share with you. Today's episode is about the future of AI employees. We cover training and managing an AI workforce, instant agentic collaboration, new AI economics, the death of SaaS, AI lifting up the developing world and the future of agents. Our guide will be Surojit Chatterjee, CEO of Ema. Short for Enterprise Machine Assistant.
Daniel Darling [00:01:00]:Ema is on a mission to reimagine how work gets done in large organizations by building universal AI employees. These aren't just standalone chatbots. They're sophisticated mesh light networks of specialized agents that can autonomously execute workflows across departments like HR, Customer Support, Sales and Compliance. What sets Ema apart is its no-code fully agentic platform allowing nontechnical users to configure onboard and manage AI employees using only natural language Instructions. With over 150 pre-built agents and a proprietary ensemble model called Ema Fusion that orchestrates over 100 large language models, the company really is pushing the edge of what's possible in enterprise AI. As a result, Ema raised a Series A of $50 million led by Accel and has become a rapid riser in the AI landscape. Surojit has one of the best product resumes in tech. He was most recently Chief Product Officer at Coinbase, helping scale one of the most important companies in the crypto economy.
Daniel Darling [00:02:00]:Before that, he led product teams at Google for nearly a decade, overseeing products like mobile ads, shopping and search, and before that served as Chief Products Officer at Flipkart, where he helped build India's leading e-commerce platform. In addition to building Ema, Surojit is also an active angel investor and a successful one too, backing startups like Udemy and Palantir. He holds a Master's in Computer Science from Sunny Buffalo and an MBA from MIT Sloan hi Surojit, nice to see you.
Thanks for coming to chat with me.
Surojit Chatterjee [00:02:29]:Nice to see you as well, Daniel.
Daniel Darling [00:02:30]:Ema, which is short for Enterprise Machine Assistant, really seeks to equip every single employee with an AI assistant that makes them 10 times more productive and it sounds fantastic. What is it about how you're building Ema that is different to how others are approaching building agents?
Surojit Chatterjee [00:02:50]:Absolutely few things. First, our assistants are not single agents. They're actually what we call AI employees. So I would love to use the word AI employee. They're actually agentic measures. So they are usually a sequence of agents or agentic workflow that comes together to automate business processes. Automate processes in customer support, HR, employee experience, sales and marketing, and variety of other areas, functional areas in medium to larger enterprises. What's unique is first we have built it in such a way that any business user can use it.
Surojit Chatterjee [00:03:28]:You don't need to code, you don't need to be a PhD in machine learning. With just a conversation you can create a new AI employee and you can then just configure it, attach your data sources. We already pre-connected to almost 200 plus enterprise applications, all types of cloud storage, databases, data warehouses, whatever it is. Ema can interpret any type of data, can reason about unstructured and structured data, can do tool use, can use all these applications and tools, can read from and write into them, has long-term memory and short-term memory, so can reason about this data, can take actions. So that's what it is. Which means it's very powerful platform where you can basically automate anything horizontally.
Daniel Darling [00:04:20]:Fantastic, thanks for that great overview. And so are we talking here, when you think about this concept that you introduced, the universal AI employee, these are still specialized employees in a specific role from there, or do we have a governing universal employee that then orchestrates all these other kind of specialized employees underneath them?
Surojit Chatterjee [00:04:42]:So when you talk about universal AI employee, basically it means it's really our platform that enables you to create or sometimes we already have pre-built AI employees, like 30 plus AI employees, and we keep building new ones for customers, right? Our customers can build it themselves. So this is a platform that can basically manifest any new AI employee very, very quickly. And that's the whole idea. So every AI employee is unique in some way because they have a different agentic architecture underneath it. So literally if you open the hood, you'll see how the agents are connected. And those connections are made initially by a model which will bring the agents together, but then human in the loop humans will give further kind of instructions, feedback, move them around to connect them in unique ways.
Daniel Darling [00:05:34]:I think a big part of this is getting that employee to have the right context about its role, about the business that it's operating in. You talk about having pre-built connectors into a lot of systems. But how are you teaching and training these employees to get to a level of sophistication where they can mimic an employee?
Surojit Chatterjee [00:05:55]:So our experience is every enterprise is unique. So for example, everybody has HR, but everybody's HR process is slightly different. It's pivoted completely on your data, internal data, your policies and your instructions so you can give natural language instruction. No code is needed ever. Natural language instructions to these agents that built these employees to behave in certain way. And that's all that's needed. Usually customers will set up their AI employee. Customers will try out different use cases, relevant use cases, give feedback like a subject matter expert on the customer side will give feedback based on the feedback, natural language feedback, these agents will learn and retune themselves and get to a point where it's better than what that team of humans are doing.
So onboarding session usually involves you'll upload like an onboarding document that you might give to a human. Or you will actually write down like you are saying conversation. You would write down instruction. Like when you hire a new employee, human employee, you know, you'll spend some time onboarding that employee. You'll give probably some documents. In our company, we create an onboarding like a getting started guide for a new employee, which is a long document talking about their role, giving them pointers to all kinds of documents, internal documents and so on. Same thing with an AI employee.
You'll give them onboarding instructions and then you'll give them feedback. It's very much like managing human employees, except it doesn't sleep and doesn't eat anything. It works 24/7.
Daniel Darling [00:07:40]:Every organization and department has its star employee that really sort of 10x type of employee that would be great to model this AI agent from. How is that being done by Ema to make this not just a standard employee, but the strongest employee it can be.
Surojit Chatterjee [00:07:59]:Great question. So if you look at customer support, which is one of the top sort of revenue generator today, a lot of customers using our customer support, AI employees, typically our AI employee will learn from the best human agents, human employees. In customer support, a big challenge is your FAQs and your documents are often dated because you are launching new products, new services and people forget to update the documents. Or sometimes the human agents don't even read those documents. So what ends up happening, if you look at how the top agents function, they not only rely on the documents that they could read, they also have like tribal knowledge in their head. They talk to other people. And not only talk like in customer support. In every function in a company, the best performing human employees are actually gathering information not only from what is written, but also from other humans, and other sources, and so on and digesting that so they have a lot of tribal knowledge.
Our AI employees, like in customer support we have an AI employee that automatically finds these gaps in the knowledge base and suggests to the customer, hey, your knowledge base is stale. You may want to add these additional things that I found your best human agents already are using. And customers are happier because of what they're doing.
Daniel Darling [00:09:25]:As part of that, there must be some orchestration with other departments or other AI employees. Are you facilitating that as well at Ema, where maybe an HR person is going to speak to the customer service people, who's going to speak to finance, et cetera.
Surojit Chatterjee [00:09:41]:So that's the beauty of our platform. We are a horizontal platform. We are not building one vertical agent. We have built hundreds of agents. And then those agents come together to build AI employees and those AI employees talk to each other. So our HR AI employee can talk to our customer support AI employee which can talk to our sales automation AI employee which makes it very easy for customers to first govern and manage this, what I call now, this agentic proliferation all in one place, like governing all of them. Who has access to what data, security, explainability, bias, mitigating or removing bias and so forth. You can manage all of them together and you can orchestrate them together. So our AI employees can call into other AI employees, they can say oh, I'm going to hand this over to my AI employee colleague over here for them to take this forward.
Daniel Darling [00:10:39]:Is that all happening almost in a real time format versus the traditional, I'm going to go schedule a meeting.
Surojit Chatterjee [00:10:45]:AI employees don't schedule meetings to talk to each other. They can just converse with each other in real time. Yes.
Daniel Darling [00:10:52]:But the productivity leap would be huge, yeah.
Surojit Chatterjee [00:10:55]:Absolutely. I think as organizations instrument more and more automation, agentic automation across the board there is almost like an exponential takeoff at some point because this agentic or these agents, these AI employees in our case talking to each other and they are more efficient of course. Right. So you suddenly get like you have one AI employee, you get some productivity boost, you get more and more get some productivity boost. But when they talk to each other, it's like one plus one equal 11 or something, not two. Because then they can suddenly bring in other synergies that you did not think about.
Daniel Darling [00:11:38]:In your position, what does an organization of the future and an org chart of the future start to look like in the coming years?
Surojit Chatterjee [00:11:47]:This is kind of a strong opinion. I believe the future will be every function in any organization will have both human and AI employees. Human managers will manage like bunch of AI employees and bunch of human employees together. You'll have like AI employees almost like a colleague. You may not perceive them that way, but it's literally that. Right? You'll also have, I mean we have use cases where like human is giving feedback to AI employee, but AI employee is also coaching humans like giving feedback like sales, for example. The AI employee is listening to the sales call and you could have made it better by this way. Or listening to a customer support call and pointing out where you could be better or where you made some mistakes.
In some cases these are actually regulatory issues. Also in fintech and regulated industries, what the customer support can and cannot say is highly regulated. And you may actually end up paying fines as an organization if your customer support violates any regulation. So there's compliance. Right. So typically what used to happen or still happens, I think, the manager or the head of customer support will have someone take a sample of those calls every week. And you will have some people just going through those samples, listening to the samples or checking the transcripts of the samples. Now you can 100% of those calls or those interactions, you can have some AI employee look over a human employee's shoulder and give feedback and so on.
Surojit Chatterjee [00:13:22]:So net net it will be a very different organization. It'll be much leaner in terms of number of probably people you need to the size of the business. Every business probably may be a lot bigger as well because of the productivity gains you'll be getting.
Daniel Darling [00:13:38]:And are there any departments that you think, though potentially, could be far more automated or completely automated first that you're seeing in your experience?
Surojit Chatterjee [00:13:48]:I mean it's quite obvious at this point, like customer support is clearly the first place where the automation will kick in very, very quickly. And you can think of it this way, a lot of organizations have already outsourced customer support. Now if you could outsource to another human, you could outsource to like AI. I think the next is we are seeing sales and marketing. There are lots of things that can be automated. Again, salespeople can be a lot more effective. Not that AI will overnight replace all your salespeople. AI will go and have dinner with your customers...Not yet, right?
Daniel Darling [00:14:24]:Is that the end of the steak dinner and the golf course then? Is that what's happening?
Surojit Chatterjee [00:14:28]:Maybe with Humanoids one day. But for now I think the sales team, any salesperson you talk to they'll tell you, I wish I could sell more and do less of paperwork, right?
Surojit Chatterjee [00:14:41]:Nobody is saying, oh, I'm so excited to do all this paperwork. So all of their paperwork they have to do, which is updating systems and so on, communicating, et cetera. Writing something, writing a proposal - can be automated, making every salesperson a lot more effective. Internal processes, a lot of back office. Right? What you do in finance, in HR. Again there are a lot of strategic things you need to do in HR, but then you spend 80-90% of time doing a lot of tactical stuff today. That's the reality of every organization, right? The tactical stuff takes so much more time. And if you talk to any executive, any senior level person, they're like, oh, I don't have time for doing strategic stuff because I barely keep my heads up out of and from all the tactical things. So a lot of that gets automated.
Daniel Darling [00:15:34]:I'd love to get your perspective on the business model and economics behind this. You talk about hiring Ema as your call to action there. What do you think is the business model for AI employees? Because there's two kind of schools of thought here. One is that it will just be a fraction of the salary costs of the humans that it's helping to automate, which is I would call the optimistic side. And the other is this kind of race to the bottom where it will mimic more like cloud computing and eventually drop down to a very, very low cost bucket. Where do you sit?
Surojit Chatterjee [00:16:06]:First, our business model and I think a lot of the agent AI companies business model is based on consumption, based on outcomes, rather than based on number of seats, which is the prevalent model today in all SaaS. Software, which is you buy 100,000 seats of some software and all those hundred thousand people rarely use that software maybe once in three months. I think there is a huge bloat of software spend today in every enterprise and it's actually useful, right, because every enterprise will be saving a ton of money by reducing the number of seats that their applications have. That's the first level. Second level is like you said, how many humans do you need? The price of compute is exponentially dropping. We have seen how the OpenAI API prices. So yes, I think just by virtue of that it will become easier and cheaper probably to use AI employees in future. That does not mean the companies cannot make money as there's just a lot more usage, right?
So it's technology. That's what always happens. As technology improves, you can make things cheaper, better, faster. I think that you will see that.
Daniel Darling [00:17:32]:And are you concerned for the traditional SaaS businesses overall in terms of the incumbents and their economic model? If, as you say, it's almost being collapsed, the whole notion of spending anything on software versus spending things on outcomes now.
Surojit Chatterjee [00:17:50]:Yeah, I am. By the way, you asked me the question like we put "Hire Ema" because we do want our customer to think about the budget for Ema should not only come from your technology budget, IT budget should actually come from your people budget as well. Because you are actually getting real work done in the enterprise. Anyhow, on the SaaS question, I do think, I think SaaS will be dead. I have been saying that actually I would say, before anybody else, before six months, before even Satya made that comment, we have published a blog post on Ema's website. I think SaaS days are numbered. That doesn't mean SaaS companies will not exist.
Surojit Chatterjee [00:18:34]:They may evolve to different business models. They will have to. So it's not the same as "all companies will be dead". No, not at all. But I think the SaaS business model will be challenged because do you need that many licenses anymore? The other thing that will be challenged is what is the actual value. Because end of the day a system of record is just a database, right? You have some UI flow on top of a database. Now if the UI flow can be manifested in real time by an agentic layer, which is how I think of Ema, we are an agentic layer. AI employee is an agentic layer over all of your data and applications and we can manifest which part of UI you need.
Surojit Chatterjee [00:19:22]:You don't need to see say all of Salesforce all the time. So you are at a given point only doing a certain set of things. And it's confusing to see a complex software. If that happens and that is happening very fast, you will question what is the real value that this software is adding. Now I think longer-term there will be some introspection about the entire enterprise stack. What stays and what goes.
Daniel Darling [00:19:53]:Do you think there's any defensibility to those systems of records that have the database and maybe leverage that position to build upon?
Surojit Chatterjee [00:20:01]:Look, a lot of them have massive distribution today. So that's the first layer of defensibility. Very obvious. But I think we have seen this movie before, right? At some point it seems like, okay, nobody will stop using, I don't know, IBM or if you think of any like PeopleSoft or whatever, right? And then people or companies realize, oh, there's a much better way. Cloud is a great example, right? Oh, why would anyone use cloud? Right. Or everybody needs data center. So there's lots of data center software companies that existed before.
Surojit Chatterjee [00:20:38]:You know, last 10 years we have seen this massive movement to Cloud because it's just better. I think the defensibility in the end for anything is long-term value. Is it really giving you the value, the ROI? I think the old trick of okay, keep upgrading your software, pay a little bit more per seat, will not work. Right. You have to rethink your model completely.
Daniel Darling [00:21:05]:You had the amazing position of being both Coinbase and at Flipkart and so have a really good appreciation of the flow of money in the system through both of those positions. Do we need to introduce a new kind of economic framework around this kind of agentic future if they're going to be interacting with each other all the time? Do we need to have things like micro payments or a micro economy for all of this activity that's going on behind the scenes?
Surojit Chatterjee [00:21:33]:I'm actually very excited about what I see as intersection of crypto and agentic AI. So there are already companies building like agents can pay each other in crypto, where micropayments are kind of easier to do. I do think that will happen with regular currency as well. Or maybe every currency may be tokenized one day. So everything is on some kind of a blockchain, which makes it easier to make these transactions. Right. Today there's a lot of friction to do payments. Right.
Surojit Chatterjee [00:22:09]:It needs human intervention a lot. And that's what makes me very excited about kind of the digitization of the entire payment infrastructure will happen. Like tokenization, even if it's a US dollar, maybe it's a stable coin or something. Right. Which makes it very easy to move around and so forth. So this is probably longer horizon. But I do see no reason for it not happening because agents can. Right. Companies may have agents negotiating, you know, prices and moving payments.
Daniel Darling [00:22:46]:Or even just like the interactions between companies through their agents could be very small requests and tasks that maybe have micro payments built into that. Would you see that happening?
Surojit Chatterjee [00:22:58]:I think so. See today it's, you know, procurement and payments. It's a challenge, right. Some human has to look at and approve and so on. Of course you have to get a lot more confidence that the agents are doing the right thing. They're not hallucinating and so on. But you can say okay, below a certain amount, maybe the agents look through and follow certain rules and make payments to your vendors, for example. Right.
Daniel Darling [00:23:23]:Do you see any other kind of economic shifts as a result of this agentic future that we're moving into?
Surojit Chatterjee [00:23:28]:Massive economic shift, Right. I think a big challenge will be which professions humans should take on and what AI should do. How do you upscale and reskill everyone? Because I think if an essential skill will be just like 30 years back, an essential skill for everyone was, oh, you have to be able to use computers, right?
Surojit Chatterjee [00:23:55]:Today we don't think constantly about it. Like kids learn how to use an iPad or phone or computer almost naturally. Almost like holding a pen. Right. They just know it and they just get it in schools. I think in future being able to work with AI will be an essential skill and being able to manage AI will be an essential skill. Maybe taught in kindergarten and like early school and so on. But that change has to happen pretty rapidly because the technology is moving really, really fast.
Surojit Chatterjee [00:24:27]:I think there'll be lots of other societal change like more transparency. Look, what AI agents are doing are all explainable, right? There's always a record and auditable, kind of unlike maybe humans sometimes, right? So humans may do a phone call and there is no record for it and so on. When you are talking about payments, I actually think, you know, there'll be more transparency if AI was negotiating and paying and so on, right. Because there is, there's no other interaction happening. You can actually go and see the entire record, an audit. Which will also bring in a lot of change in how governments function, institutions function. I'm very optimistic about it all.
Humans are extremely creative. We will create and innovate faster. Right? Build newer things. And a lot of what we are doing today as like repetitive, tedious job will be delegated to AI.
Daniel Darling [00:25:29]:Yeah. And how do you see it potentially manifesting differently as a future in the developing world versus the developed? And you started your journey in India. What would it start to look like rolling out in a nation like that with its economic and industries set up as they are?
Surojit Chatterjee [00:25:48]:Great question again. When mobile came, India completely leapfrogged, Right? And most businesses are mobile first or mobile only in India, Right. You can go to any city, order anything, anything at all. You can get it within 10 minutes or 20 minutes. It's just mind blowing, quick commerce, it's called. Mind blowing how efficient it has become. And it has actually instrumented hundreds of thousands, millions of jobs.
As an economy, India has done better than ever because it embraced technology. I think like today with AI, same thing, there is worry what will happen. I think just the opposite will happen. Developing markets may find actually, because the big challenge in developing markets is shortage of skilled labor. Right? Now you can wait for everyone in the labor force to get more skilled or you can take help of AI and accelerate your business. Imagine that.
Daniel Darling [00:26:50]:That's a great point. Yeah.
Surojit Chatterjee [00:26:52]:And that's what creates more wealth, that's what employs more people and so on. So in a place like India or any developing world, like if you think of Africa, if you think of South America, healthcare...it's not like there are too many doctors, it's too few doctors compared to the population. Right? People in villages, they wish they could actually get a consultation easily. They have to probably travel for three hours to meet a doctor. Can they at least first level talk to an AI doctor to get, you know, and then get a referral or whatever. Right. It's not that there are too many teachers in India, like too few teachers, too many kids to teach.
Surojit Chatterjee [00:27:38]:The class sizes are huge. What if every kid had their own teacher and that's possible now, had their own personal teacher who will personalize the curriculum for them, for their interest, their skill set, their abilities. That will grow the population, kind of the skill level of population in an exponential way. I think if you look at where the developing countries are challenged most, it's not that it's the skill of people, if everybody is skilled and also the health. Right? The kind of the fundamental is like literacy, skill level, healthcare. Right? If you can make significant improvement in service delivery, in education, you will see actually an explosion of growth and wealth.
Daniel Darling [00:28:31]:I think that's incredibly well put and does shine a really optimistic light of exporting that talent to the developing world. I'd love to end our conversation around startups. Again, you're building a startup, but also you're a fantastic spotter of startups. Having invested in companies personally like Udemy and Palantir. Are you a believer that we're moving towards a proliferation of startups and eventually to these kind of one-person billion-dollar companies that have fully automated, highly agentic or where do you see that evolving to?
Surojit Chatterjee [00:29:01]:Absolutely. Look, startups and small businesses are engine for growth in every country. Right? It's not like any country can become wealthy with just having three large companies and we are talking about developing world a little bit earlier and India, I mean, that's it. There's an explosion of startups in India because skill levels have improved, a lot more people who can start companies and regulations have decreased so people can start companies. So I think with AI, the help of AI, it'll be possible to start companies with like you know, five people, 10 people making. And it's already true. If you look at many of the AI startups you can create a much larger company with much fewer people.
Surojit Chatterjee [00:29:51]:I used to work at Oracle in my early days. There's a whole team just cutting the binary release and putting them on CDs. It was like, you know, and shipping those CDs and there's a whole process around it. You know, today people don't even know what that looks like. You just push your code and it works. I think the same will happen. It's already happening. And that's what makes me very optimistic about the future because you will see a proliferation of new ideas, people exploring things, building companies much faster than and much leaner than before.
Daniel Darling [00:30:30]:A great positive note to end on. I really enjoyed our conversation here. Thank you so much and congratulations to everything that you're building at Ema. Really look forward to following that progress.
Surojit Chatterjee [00:30:40]:Thank you so much.
Daniel Darling [00:30:41]:Amazing to see the future of agents through Surojit's seasoned eyes. A future where AI employees don't just assist, but collaborate, they coach, they negotiate, and they scale with us. Organizations will become exponentially more capable by orchestrating agentic systems. Agentic systems that reason, adapt and act all in real time. And maybe they even transact autonomously, powered by micropayments and tokenized currencies. As this agent-first world emerges, traditional software models will be rewritten, new economic systems will take shape, and developing nations may rise up by embracing this new workforce. Surojit's roadmap doesn't just hint at better tools, it sketches a transformation to how the world could work.
To follow Surojit, head over to his account on X @SUROJIT. That's S U R O J I T. I hope you enjoyed today's episode and please subscribe to the podcast to listen to more coming down the pipe. Until next time, thanks for listening. Have a great rest of your day.
