Videos
Mit Majumdar on AI-Native Processes and the Execution Imperative
Insights
- The greatest AI gains will come from redesigning processes around business outcomes rather than adding AI to existing workflows.
- The primary barrier to AI adoption is shifting from technology to execution, requiring domain expertise, organizational change, and sustained workforce engagement.
- As AI automates more work, human roles will increasingly focus on process design, judgment, validation, and ensuring that AI-generated outputs deliver the intended outcomes.
Mit Majumdar, EVP & Global Head of Services at Infosys, discusses the shift from productivity and legacy modernization toward AI-native process redesign and the agentification of enterprise workflows. He explains why meaningful transformation requires organizations to rethink processes around desired outcomes, supported by domain knowledge, change management, and human oversight rather than technology alone. Drawing on examples including AI-enabled audit processes, the conversation examines how agents can expand the scale of work while increasing the need for human judgment, validation, and governance. For executives, the central challenge is increasingly execution: turning AI strategy into sustained adoption and measurable business impact.
Jeff Kavanaugh:
I'm Jeff Kavanaugh. We're here at the Infosys Connect Conference in Los Angeles. Very happy to be here with Mit, who leads our services group, and more importantly, some of our new great work going on in AI. Good to see you, Mit.
Mit Majumdar:
Good to see too Jeff.
Jeff Kavanaugh:
You bet. Let's jump right in. What is it that you're seeing right now with the next frontier of AI in trying to unlock value?
Mit Majumdar:
So when we started right with AI, the journey was about productivities, SDLC productivity.
Jeff Kavanaugh:
Saving money, yeah.
Mit Majumdar:
Saving money.
Jeff Kavanaugh:
Efficiency.
Mit Majumdar:
With efficiency, right? So productivity. Next came up is legacy modernization. SDLC productivity, PDLC productivity, legacy modernization. These were the things that were talked about. Now that is still there, but that is like everybody knows about it. Like code efficiency and things like that.
Jeff Kavanaugh:
Like code efficiency and things like that.
Mit Majumdar:
Yeah, yeah, yeah, yeah. And using that capability to do modernize your legacy ecosystem.
Jeff Kavanaugh:
Technical debt reduction.
Mit Majumdar:
Technical debt reduction. And then If you don't do that, then you will not be able to leverage AI in its best way.
Jeff Kavanaugh:
At the next level.
Mit Majumdar:
Yeah, exactly. Because if your tech landscape is legacy, it's difficult to use AI on top of it. You can, but you will not get the best out of it. But now the talk is about agentification of processes. Now, agentification of processes has two parts to it. One, you have the existing process and you slap AI on top of it. It's called AI augmented. And then you get some benefit for sure. But it is very incremental benefit you will get. There's another way of looking at it. AI first, AI native, meaning you redesign the processes with outcome in mind. You have an outcome in mind and you redesign the processes.
Jeff Kavanaugh:
It's like the Clayton Christensen from Harvard saying there's a job to be done and working backward from that. Yeah.
Mit Majumdar:
Same philosophy.
Jeff Kavanaugh:
Yeah.
Mit Majumdar:
Exactly same philosophy that you redesign the whole process, that's where you get the maximum benefit of using AI. Right? So that is where the world is moving. It's not there yet. I am having many conversations with some of my customers who are willing to completely redesign their processes.
Jeff Kavanaugh:
That's taking more than just a good tech expertise and depth. It's requiring good thinkers, a more consultative and actually more creative mindset.
Mit Majumdar:
It's not at all about technology. Technology is there, it's given. When you look at Nandan's presentation, he says that technology curve is this way. Adoption curve is this way. There's a big gap.
Jeff Kavanaugh:
Execution gap.
Mit Majumdar:
There's a big execution gap. So technology is not the issue. The issue what you just said, knowledge of domain, knowledge of the business, understanding of the outcome, and it's the people that you have to take along. It's the org change management that you have to do. You cannot just design the process and assume that agents will run everything.
Jeff Kavanaugh:
And that's contrarian thinking rather than saying people don't matter, agents will do it. It's saying people matter more and you make sure your agents and everything are designed to optimize what they're doing.
Mit Majumdar:
It is the people who will tell what the outcome is needed. It is the people who will tell you the business processes that can be designed based on that outcome. And then you will use agents to agentify it, right? There are many ways within Infosys, we have designed how to redesign the processes, the approach to it, and all that. But that has less to do with AI, more to do with that thinking around outcome-based process reengineering and then you use AI to implement it.
Jeff Kavanaugh:
You have a special role in the company where you're looking at some of these AI native and some of these partnerships and what are you seeing that's kind of on the edge now that people haven't seen yet?
Mit Majumdar:
Initially what I thought and I learned something new recently, right? So we were in this discussion with Anthropic, where we had this knowledge of Claude code, SDLC productivity, modernization and so on. And then we had this thought process about completely agentification of processes, right? There is something that is in between. That is this whole cowork concept.
Jeff Kavanaugh:
Yeah.
Mit Majumdar:
Claude Cowork, right? Where you apply that in different organizations, say finance or legal or marketing, right? They have ready templates for some of these processes already ready, right? So you use Claude Cowork, and some of your processes, small sub-processes, will automatically you can identify some of them, right? That is a very nice transition towards full agentification. You start with Claude Code, you move to Claude Cowork, and then you move to complete agentification of processes. So very nice transition, which I noticed was a very interesting new learning. Another thing that is being discussed a lot is security.
Jeff Kavanaugh:
Let's look at some examples. Given that you have a lot of responsibility in the area of services and companies. Can you share any examples of services firms where they're making some early progress here?
Mit Majumdar:
Oh, 100%. So we work with one of the professional services organizations where they completely agentified the audit process. So earlier when you used to do financial audits, right, it was based on outliers. And then you sample it and then you audit it. But right now, you don't have to sample it. You can actually audit the whole documentation.
Jeff Kavanaugh:
Right.
Mit Majumdar:
Because of AI capability, right? And we have agentified around 12 processes across the whole audit lifecycle. And we have created agents for it, and we have created orchestrative agents on top of it which can manage these agents. So it's a beautiful work that we have done for...
Jeff Kavanaugh:
And if you can get the SEC and AICPA and all these organizations to trust it, that's a big step forward for AI as well.
Mit Majumdar:
Yes, absolutely.
Jeff Kavanaugh:
What's the human role in this? Because there's the... Are you being replaced? Is it who's working for who? What's the role of people going forward?
Mit Majumdar:
Oh, my goodness, it's just going to be a lot. People don't understand it, right? So let me just break it down for you from my point of view, right? One, redesigning of all these things, whether it is if you talk about SDLC and PDLC productivity, right? All this coding will be done by human beings. I know the code writing is by the agents, but designing the processes, the analysis of it, what is the outcome that we are going to get out of this code is all actually by the human beings, first of all. Second, once it is done, who is going to validate that if the work is,
Jeff Kavanaugh:
The quality assurance aspect, yeah.
Mit Majumdar:
Right.
Jeff Kavanaugh:
Judgment, reasoning, critical thinking.
Mit Majumdar:
So there's a need for people who are going to write these whole processes. There's a need for people who are going to apply Claude Code to do this agentification of some softwares. Then there is a need for people to validate the work that got done. This is just in this area. Legacy modernization, if you talk about again, same thing, you know, what code to be modernized. If there's a reverse engineering being done, that is one of the processes in legacy modernization. If reverse engineering is done, who is going to validate the business process that got written, right? You can't ask an agent to just validate it and assume that it is done. So there is... every step of the way, human has to be there in the loop either to initiate it or to validate it. Right? And then you go to the agentification, there is a bigger need for human beings. So the roles may change. You may have less roles for coders, but you may have much more larger number of roles. And don't forget, when you are applying AI, much more amount of code is getting written. Right? Earlier if you could write million lines of code, now you can write hundred million lines of code.
Jeff Kavanaugh:
Yeah.
Mit Majumdar:
Who's going to verify all this hundred million lines of code?
Jeff Kavanaugh:
Yeah.
Mit Majumdar:
So there's a lot of roles that are coming. It's just going to change the roles. So people have to be ready to learn, which I can see people are. I mean, you just look at it in this conference. People are learning.
Jeff Kavanaugh:
Many are. Some will be a challenge, but yeah.
Mit Majumdar:
Well, I mean, you know, isn't that the nature of any...
Jeff Kavanaugh:
I'm not saying it's different than before, but yeah, it's the nature of humans. The thing I keep coming back to, we may, and then developers and others, may only be doing 30% of the things we want to do. So in some respects, we'll finally get to the rest of our to-do list. Yes. That's the way I view this as well.
Mit Majumdar:
Absolutely.
Jeff Kavanaugh:
How many projects have sat because we didn't have a chance to? How many markets do you not serve? How many processes do you not look at? How many analyses do you not run because you never had the time or the money?
Mit Majumdar:
True.
Jeff Kavanaugh:
If now you have the time, it's like Amazon with the long tail. You can finally reach these things.
Mit Majumdar:
See, once you have the tool in your hand, your creativity will just blossom. You'll start thinking new ways to use the same tool. And some of those ways will be AI slop and some of these ways will be amazing work...
Jeff Kavanaugh:
I tell people it's an Iron Man suit for you.
Mit Majumdar:
It is an Iron Man suit.
Jeff Kavanaugh:
You can go faster, think more. If you're not good, it doesn't help. But if you have skills, it just augments and expands them.
Mit Majumdar:
Absolutely.
Great. Well, what is the one thing, given some of the vantage point you have that executives should be thinking about or will be seeing as an opportunity the next nine to twelve months that they haven't seen yet?
Mit Majumdar:
I think more than opportunity, I will tell you something that will help the organizations succeed in this era is the word execution. I don't think strategy is a challenge anymore.
Jeff Kavanaugh:
Yeah.
Mit Majumdar:
It is the ability to execute is the biggest differentiator across every organization. You will see a lot of people talking. Few companies will actually land up executing. Execution needs dealing with people, making people do something that you have already planned for it.
Jeff Kavanaugh:
Well, it's the hard side of change management. It's adoption, it's impact.
Mit Majumdar:
Because I'm noticing that across the board, right?
Jeff Kavanaugh:
Yeah.
Mit Majumdar:
These LLMs organizations, right, they created something which they expected to be used. So they made investment for it to be used by the enterprises. They noticed that it is not getting used fast enough. So their tokens are not getting consumed.
Jeff Kavanaugh:
Okay.
Mit Majumdar:
So they probably expected that all the system integrators can help them get it done. So system integrators like Infosys got into act. We are all working towards it. Their expectation is more faster. You know, usage of tokens. So what they did, they started creating their own companies now. They're creating companies with FDEs. Point is whoever helps the tokens to be consumed faster will get there faster. So execution is the key in all this.
Jeff Kavanaugh:
Great. Mit, I know we have to run. Thank you so much for your time.
Mit Majumdar:
Thank you, Jeff. Thanks a lot. Always appreciate it.
Jeff Kavanaugh:
Always appreciate it. I'm Jeff Kavanaugh. Until next time, keep learning and keep sharing.
