#HRTechChat: Darwinbox’s Samrat Krishna on Killing the Toggle Tax and Unleashing AI Agents at Work

Posted on by

AI is changing the game—but not in the way you think. In this episode of #HRTechChat, we’re sharing a special swapcast from the PeopleTech podcast hosted by Mark Feffer, featuring Samrat Krishna of Darwinbox.

Samrat unpacks how AI agents are killing the “toggle tax” and rethinking enterprise software altogether. Instead of juggling dozens of systems and interfaces, agents now act on your behalf—fetching info, taking action, and streamlining workflows across platforms.

They dig into the mechanics of MCP (Modular Component Protocol), how Darwinbox is building for agent-first environments, and what HR leaders should automate today to unlock strategic headspace tomorrow.

If you think AI is about replacing people, you’re asking the wrong question.

Tune in to learn how to shift from assistance to agency—and why that shift changes everything.

Our #HRTechChat Series is also available as a podcast on the following platforms:

See a service missing that you use? Let our team know by emailing research@3SixtyInsights.com.

Transcript:

Mark Feffer 00:00:00
Welcome it. It’s very nice to have you here. I’m glad to meet you and. I wanted to start. By asking you about AI in general. I think you said that you don’t think the next phase of AI is going to be about assisting people, but it’s going to be about agency. Can you tell me what you mean by that?

Samrat Krishna 00:00:24
Oh, absolutely, Mark, I think. What we’re experiencing right now is an extraordinary step jump in what technology promises, right? But to understand what agency means and how software can deliver, maybe we should take a step back and understand. What were we really relying on technology to solve for us right from the very first principles lengths? Any enterprise software? Promised an interaction between a user and a database mark, right? Let’s take an example. If I was. Wanting to punch my attendance in for the day, I was essentially telling the database that XYZ employee is present at a certain location on that date, correct? Here is what used to happen, right? And to understand this interaction between the user and the database, there are two constructs that we need to understand. 1 is a user interface right? For instance, in this particular example, I may pull up my mobile phone and a mobile app that my CM. Provider gives me. And punch in my attendance. And the second thing that we need to understand is logic business logic, right? Right. The software would have hard coded some rules against this particular employ. You know, this software would identify that this employee belongs to a certain department and their shift starts at 9:00 AM in the morning and closes at 5:00 PM in the in the evening. It identifies whether the Puncheon is late or. Early. And then goes ahead and tells the database that this is. What was the? Case right? If a particular exception had to be flagged that hard coded business logic would make sure that the exception was raised. Right. So 2 concepts to understand right between the interaction that is user and the database, there is a user interface and there is business logic correct? What had happened in the status quo in the? Last 10 years was. Many user interfaces popped up right, so here I’m punching in my attendance. But if I was stretching that logic and and and taking that one step further, if that attendance function is being used to compute my pay. And I wanted to know what my pay and I wanted to access my pay stub. I would have to go to another. Another user interface right and rely on another set of business logic. To be able to. Fetch the payslip that I was to access. Let’s extend that one step further, right. If I had a question about my pay stub about how my pay was computed, I would have had to go to another user interface to speak or register my request with the HRBP. To help me with this particular request. Now what we are suffering with is. Yeah. So many different user interfaces that we have to access. All of them. Come with the baggage of hard coded business logic, right? Now it seems very complex, right? You have to navigate so many different systems to achieve a simple task, which is I just wanted to look at my pay and understand how it was completed. Right. But something magical happened in the last one year Mark and and and this has shaken up pretty much every technology vendor out there and every organization should be paying attention to this step chart. Which is. The dream of a single user interface is now real. Right. All you have to. Do is go to a conversational interface with an agent. And tell it, hey, what’s my payslip and how is it computed? The agent or this piece of software is going to ask 2 questions. One is do I have the tools to achieve the goal that the user? Wants me to achieve. And 2nd. Is does the user have access to the tools they need to achieve the code right? I could be asking, for instance, my managers pay and that may not be information that I’m Privy to or I should have access to so the agent also needs to make sure that the user actually has permissions or access. To the action that they are asking the agent to do, and that’s all. That’s what happened. The agent goes to multiple user interfaces and negotiates on your behalf. Either fetches information or takes an action on your behalf and achieves the end goal as long as it has the tools and the permissions to achieve the goal. That’s the money that has happened. So all of the baggage we had with multiple user interfaces, all of the you know baggage we had with hard coded business logic across multiple systems, not talking to each other has now been simplified. Because this piece of software developed agency and is able to navigate across multiple systems to take actions on your behalf. And achieve the end code provided you have the permissions and the agent has access to quote UN quote tools, we can go into a little more detail on what tools are and how they articulated, but the step jump that we’re experiencing is just that software developed agency, if you. Right. That on your behalf.

Mark Feffer 00:06:01
And and some of the things you’re just talking about some very simple, but they really add up. I I saw a study somewhere that just the context switching between applications can eat up 20% of a person’s time over the course of a week. It’s huge.

Samrat Krishna 00:06:18
is crazy. It is crazy that so much time was wasted. I I call this the toggle tax mark, so the every, every time you have to toggle from one system to another system. You’re losing context. You’re losing flow, you’re losing. And and and. What seems like such a simple problem statement mark this dream of 1 user interface or this dream of unified business logic across the enterprise was very hard to achieve. If you think about it, there was an entire business software category that got built on this think mule soft think snap logic. These guys existed to connect systems. Right. The you know the you know the challenge is. A lot of employee productivity gets wasted when context gets lost and business business systems can’t. Talk to each. Other and that’s the promise that agents are delivering today, right? They’re able to save that time for you. They don’t need you to be there everywhere, right? They’re doing things on your behalf, right? All you need to make sure is every business software that you have in your organization or in your enterprise needs to make its tools available. For the agents to work with. Right. And that is happening through something. A new protocol called MCP right without without complicating this too much. I’ll just. Maybe a good analogy to think is. You know, you may have your CRM system in one corner that has all of your customer data. You may have your finance system in one corner that has all of your financial data, and you may have your HR system in one corner that has all of your employee data and you want to execute an action that touches all three of these system. All three of these systems will have to decide to join the same party. Correct. And that party is MCP. Right. So it’s really just a translation layer where all of the systems can make their tools available to your enterprise agent and that agent is able to go out there and take decisions on your behalf or make actions on your.

Mark Feffer 00:08:28
Behalf right now, Darwin Box just launched MCP Server. And so for some reason.

Samrat Krishna 00:08:39
Just. OK. That question I think that was.

Mark Feffer 00:08:43
Yeah. No, I’m just going to. I kept this minute of him or me. Darwin Box just launched MCP Server and I wondered for someone who’s not on the technology team. Ken, you tell me about that. What’s what? Benefits to us a break.

Samrat Krishna 00:09:04
Yeah, sure. So Let’s take the same logic that we. I’ll repeat that so. Let’s take the same example that we quoted in the beginning, which is a user wanting to understand. How they’re how there is a discrepancy in their pay stub. They had to engage with three different systems. All of them desperate and have different business logic. They had to again, like you said, swivel chair. Between 3 systems losing context and having to follow up on notifications from three different interfaces. What MCP makes possible Mark is it makes sure that all of these systems are able to come and speak with each other and make themselves available for the agent to take an action for you. OK. Let me attempt. To answer that question a little, a little simpler. What an agent would we discussed the idea of agent needing access to tools to achieve an end goal correct? Tools are literally just actions, for example. Applying a leave on your behalf is a tool. Fetching your leave balance is a tool. And. Fetching an information against an active sales deal that you are running is a tool which a CRM has to provide. Again. Fetching information about a support ticket that you’re solving for your customer. Is a tool. If you think about it, all software are just collection of tools, right? There is a set of tools that allows you to apply leaves that allows you to fetch your leave balance. That allows you to access your pace leads. That allows you to look at your benefits all of. These are tools. Right. So HR is a collection of about 200 tools. CRM is a collection of about 3. 100 tools. Finance is a collection of about. 100 tools, right? MCP is what makes all of these tools available. In one server. So now imagine. A simple use case which has to touch. The HR data, CRM data, finance data and ERP data is. Is is possible to execute on an MCP server because all of the tools are made available. On this PC. Right. That’s literally all that’s happening. This is a playground for the agent, right. And and in this playground for the agent, all of the systems are making their tools available on MCP. MCP is just a global standard, mark. Like, we had APIs back in the day, right? Application programming interfaces, which would connect two different systems. We now have MCP and MCP. Is this common playground where actions can be performed because vendors like Darwin Box go out there and make their tools available for these agents on an MCP. Server. Similarly, vendors like. Spot come to the playground and say hey, we will make these tools available for your agent, for your player. I hope I was able to articulate that in a way where. Yeah, makes sense.

Mark Feffer 00:12:22
Yeah. No, I I think it did. Let me just ask. You know, there’s so much noise around AI right now. There’s so much hype. But what? What does it really take? To make these agents work in the.

Samrat Krishna 00:12:39
Yeah, I think exciting question, Mark. And if I was a, if I was an HR leader or any function leader in a company, I would first of all ask. Some very basic questions right I would ask. What’s a task? That takes away time. That doesn’t need creativity. And and and has no. Exponential jump in the value that it provides, right? What’s a? What’s a task that’s easy to automate? And what’s the task that takes away a lot of the time? A lot of the productivity? Of my employee. On a day-to-day basis, I would ask. That question first. Right. And I would. Figure out ways to automate that. This is what we were doing. In fact, in what I would call the AI 1.0. Era, so to speak, right? But once agents have arrived, they are like I said now able to do a lot more of that because I 1.2 was very limited to individual systems. But agents can now work across multiple systems. And so now you can actually go back and ask questions around what’s the use case that I want to. Automate and even if it is across systems, I would prioritize and create an agent for that right. I would go to my vendors and ask them what agents can you provide. Me that he that that make my job a little easier that make my employees time worth more right. So I would ask those hard questions I would ask what can be automated and what is non creative that can be taken off of my advice please. 2nd is know that anything that can be articulated as a series of steps can be achieved. With an agent. As long as you can articulate a particular starting point and an end point and articulate that there is 3 steps to achieving that, you can automate that through an agent, no matter how many systems you have to touch, no matter how many. You know the workflows you have to build all of that can be automated, right? And to make agents work for you. Just figure out what is it that you want to take off of your employees plates and then ask your vendors these questions and ask your IT teams these questions on on how do we execute, what’s the path of least resistance to deploying agents to solve these problems right. I would perhaps give you an example Mark. We work with about 1000 plus customers across the world, all of them struggle with talent acquisition. You know, mounds and mounds of resumes spilling on and it’s so hard to run off, run through all of those resumes and figure out who who. Should I talk to right? Right. Job descriptions are dynamic. They are evolving so rapidly so. An agent can now go through all of those resumes. Understand my priorities. I could just ask the agent. Hey, I am looking for someone with Ivy League experience who works on a particular technology, but I also want them to have some sales expose. Either by having worked closely with sales teams or having a client facing you know, role in in, in. You know. In some of their expert. This query is extremely easy to execute today, right? What would have been 3 or 4 steps back in the day, right? Simple. I just want to talk to someone who has Ivy League experience and has market facing experience and has project management experience. This query is easy to execute hence making recruiters lifes a lot more easier. And spending time on whether the candidate is a good cultural fit instead of worrying about how to get to the resume that you know is a moderate 60% fit to. Buy job description.

Mark Feffer 00:16:33
So with MCP, what what can an agent now do inside Darwin Box that it couldn’t do before?

Samrat Krishna 00:16:41
Yeah. First of all, Darwin Box is an end to end HCMS suite which is a collection of multiple modules. So time and attendance and performance and pay role and talent management and so on and so forth. So if I was to ask you know very simply a query that takes you know historically would have taken us multiple hard coded. Business logic to achieve. Let’s take an example. Maybe mark. I think the best. Thing. To do is. Let me answer that question again, right, let’s take an example of an employee planning verification. Right. Here are my constraints. I want to take a 5 day vacation. But I want to make sure that I’m. I’m maximizing and I’m I’m I’m minimizing the number of leaves that I’m taking to plan that vacation right? It’s a simple ask. I want to plan a five day vacation. With the minimum number of leaves right now, this agent will go ahead and understand my holiday calendar. First of all, right. So clearly the user is trying to club their vacation with some public holidays. It will go ahead and understand my holiday calendar. It will go ahead and understand my shift that I’m assigned, right. I do a five day we. And and Saturdays and Sundays are free for me, right? Or you know, if you’re in a manufacturing floor, perhaps a Wednesdays and off for you. Right basis your shift and so on and so forth. So it understands my holiday calendar. It understands my shift. And then it goes ahead and tells me that there are four options in the next three months where you can club. Just a couple of leaves with three public holidays to make sure that it’s helping you in planning that vacation, but it doesn’t stop there. Remark right? It goes and gives you that option and then it asks you would you like me to apply for leave basis whatever option you choose. And then you can say, hey, yes, I would like to do that. And then it asks you a question, what kind of leave? Would you want to? Apply. Would you like to apply a casual leave? Would you like to apply an annual leave? And so on and so forth, which means the agent is also accessing my leave balance and telling me that these are the leaves that are available to me, that it can go ahead and book from it. And then finally I can just go ahead and say, hey, just book an annual leave. Right. So not only did it identify the best possible opportunities for me to take vacation. Clubbing with public holidays, it understood my shift. It understood my preferences and it gave me options and once it gave me options, it looked at my leave balance and told me, you know and and and went ahead and took an action on behalf.

Mark Feffer 00:19:24
No, we’re we’re talking. About tools here in a lot of sort. Of under the hood. Kind of things. Could you sort of? I don’t want to say bring it down. To. Me. But umm, how do you how do? You convey all of this to, you know, HR. People or or it leaders, not the developers. How do you get them excited about this or or at least understand its?

Samrat Krishna 00:19:52
Yeah. No, absolutely. I think it’s quite simple, anything. Any technology needs to be applied. To business in terms of the value that it can unlock for you, right? And according to me a I can unlock value in four key verticals, mark one is it can drastically improve. Operational efficiency for you, right? That’s the first one. Operational efficiency could mean your day-to-day activities. Can I automate them? Can I, can I make them smarter, right. Can I take the mundane out of my employees lives and you know and give them more creative work and more strategic. At the first lens, right? So creating a workflow on your system back in the day would consume. A lot of time today it’s just one prompt away, right. Type up. Hey, I want to create a workflow for a first time manager. Right. I want to give them an experience where they’re prepared for their job. Maybe this includes 4 different steps, right? That that orchestration of the journey an agent can execute for you, right? The second pillar that can help you know leaders evaluate the value of of these agents is can it drastically improve employee experience. For you. It is such a broken experience mark to have to go and ask questions to an HR business partner or a help desk portal that I have a problem. I need help and then wait for a 12 hour turn around time or a 24 hour turn around time for a resolution for my problem. Right an agent can reflect about 80 to 90% of those queries that are typically hitting your. HRPS you know simple things like I need an asset replacement or I need information about my benefits, right? All of these things can be taken. From you know your HR business partners or your or your employee service delivery teams, hence saving you. The third is can it make everyone in my org smarter right? Can it actually help me make better decisions. For instance, you and I have been in situations where we’ve looked at our. Can you? Teams and said hey, attrition is rising in my company. Right. But that’s such a broad strokes metric to work with. What do I take away from it, right. Can I understand it a little bit better? An agent can go out there, right, and understand. And and. And it can do this on MCP, right, mark, it is able to search compensation data. It is able to fetch engagement data. It is able to fetch compensation benchmarking data that some other vendor may be providing and that made that available. Time. And it is able to do that analysis and tell you specifically that hey, perhaps this particular location women in the team are feeling like they don’t have enough growth opportunities and and and and this may this feeling may be coming from you know their compensation being much below parity. Against the against the industry, right. So that’s the third pillar can can I be better with my analysis? Can I make better decisions? Thanks to my agents giving me a deeper dive of what’s working and what’s not working right. And the 4th is quite simple. Mark, can I can it drastically improve my talent luggage. Everyone struggling with this problem statement. We are in the most volatile talent ecosystem period ever, right? So how do you make sure that every employee has their? Aspirations registered with you has, you know, a curated learning path that is upskilling them regularly. Access to projects and gigs that will that will that will help them learn new skills and help them collaborate with teams beyond their regular teams that they work with on a day. To day basis. Right. So. Operational efficiency, how can you make your admins life easy, right? Second, employee service delivery or employee experience? How can you make your employees day-to-day experience faster and and and and simpler? 3rd is can you be smarter with your decision making? Can you rely on agents to give you the right information at the right time? The right nudge to be able to be better with making these decisions and 4th, can you drastically improve your team’s ability or your talents ability to upskill and learn more contextual skills that allows you to sort of future proof your workforce. Right. I would say if you’re an HR leader or an IT leader, ask hard questions on how agents and AI can help you solve. You know these four problem statements and that will definitely translate into value for your entry. Right.

Mark Feffer 00:24:35
And I think one of the big questions that everybody’s asking about tools like this is what, what’s the context or the restart. Again, I think one of the questions people have about tools like this is really about the relationship between the human user and the agent. Is is this about a partnership between the two of them? Or is this? Really, at the end of the day, about replacing people with machines.

Samrat Krishna 00:25:05
Yeah, that’s a great question. And and the anxiety is very valid, mark, but. I I want to maybe share a couple of things, right, anything that AI is doing today, it only learned from us, right? It it it. So what? What AI is doing today is it’s great at organizing world knowledge. And in the last 20 to 25 years of Internet, we kept giving Internet a lot of information about how we do things. It what AI is doing is really just utilizing a lot of that information and and helping us, you know, do things that we used to do back in the day with a lot of effort and do it simply. So it’s a great manager of knowledge. It’s not an inventor in itself, right? We’re far away from the reality where you know. Hey I can go out and why haven’t we solved intergalactic travel? Shouldn’t we just ask AI to solve that it it cannot. That’s the reality. We haven’t solved it as humans and that’s why I can’t solve it, right? So it is super important to understand that artificial intelligence is really just augmented intelligence in my head, right. I know we like to fantasize this and and there is a lot of, you know arguments pro and and. And and and and against the rise of technology. But I. Can tell you this one thing right. It is just a partner, right? It is something that is designed to help make your life a little easier so that you do things that matter. And finally, move away from things that drain you of energy that drain you of you know what you were supposed to do. You were supposed to think you were supposed to make decisions. You were supposed to question the status quo and improve the status quo at any point in time. That’s what humans were supposed to do, right? And once you find the status code, rely on your agents. To automate that stuff. Leave the grunt to technology. Ease the grunt to artificial intelligence. Yes, it’s becoming more and more intelligent and it’s able to do things for me that it wasn’t able to do before. And that’s a good thing for us because it frees up to do frees up time for us to do a lot more interesting things evolve our, you know expectations and and and. And job descriptions and so on and so forth. So we are firm believers that a AI serves humans and and and that’s the length that it should be looked at. I don’t think they’re they’re. If we are on that path, we’re far, far away from what what people call. Synchronicity or or or general intelligence.

Mark Feffer 00:27:35
I Love Actually the term augmented intelligence. I think that’s a a really great way to put it, to put it that have actually heard a lot of people. Use that so yeah. Well, listen, thank you so much for taking the time. I I hope someday you’ll come back and. We can talk. More about this and it was great to meet. You and thank you for being here.

Samrat Krishna 00:27:58
Absolutely, man, this was a lot of fun. I hope I was able to do justice to the questions that you asked and and. Ah. Yeah, super, super excited to do this and I’d love to do this again.

Share your comments: