UiPath’s shift from RPA to AI agents — and why India is key to scaling it

In today's time where digital workflows are quickly outmoving traditional automation, UiPath is making a determinative pivot—away from robotic process automation (RPA) and into what it’s summing “agentic” AI.

With more than 10,000 customers worldwide in more than 100 countries, the company is not riding the AI wave but actively shaping it. At the forefront of this evolution is India, not just as a talent nursery but also as a strategic hub of innovation. 

In an exclusive conversation, UiPath’s CTO Raghu Malpani and VP Gautam Goenka detail how AI agents are poised to go beyond rote task execution to understand goals, make decisions, and collaborate seamlessly with humans.

What Comes After RPA?

UiPath has clearly made a shift toward AI agents. What triggered this evolution from traditional RPA? Was it driven by competition, or was there a deeper strategic intent?

Raghu Malpani: Business processes, as you know, can be highly complex. They often span long durations and involve multiple stakeholders. While traditional RPA has been effective in automating structured, repetitive tasks, there are always components of these processes that are dynamic, ambiguous, and judgment-based—areas where RPA, being procedural, falls short.

We’ve always known that these “non-deterministic” elements existed—parts of a workflow that can’t be scripted easily. So when large language models (LLMs) started becoming viable—roughly three years ago—it felt like the right moment. It wasn’t just about industry pressure or a tech trend. It was a clear opportunity to extend automation deeper into areas that were previously untouchable. Our customers were already asking about agents, but for us, it was also a proactive move. There’s significant untapped potential within our existing customer base, and agents allow us to unlock it.

Today, we’re seeing thousands of agents built across hundreds of customers. While many of these implementations are still in the proof-of-concept (POC) stage, the momentum toward scaled deployments is very real.

How long does it typically take to move from POC to production? Are there specific sectors leading the charge?

Raghu Malpani: It really depends on the use case. In some scenarios, the transition from POC to production can be fairly quick. We’ve seen that financial services and healthcare, in particular, are moving faster—these industries often have repeatable, high-value use cases where agents can immediately add value.

However, it’s also true that many customers are still in the discovery phase—evaluating where agents fit best in their operations. It takes time to identify the right processes and build internal confidence. But once a business understands the value and gets buy-in, the implementation itself is not overly complex. That said, it does require specific skill sets and tools—which we, at UiPath, are deeply focused on enabling.

What models power your agents today?

Raghu Malpani: We’re model-agnostic and work with a broad range of LLMs, including OpenAI, Anthropic, Google’s Gemini, Meta’s LLaMA, and Mistral. Larger enterprise customers often have preferences based on their priorities—cost, accuracy, latency, and control.

Some customers bring their own models—say, a fine-tuned OpenAI deployment hosted on their infrastructure—and we support that. Others use the models we offer out-of-the-box. The flexibility is built into our platform: it’s literally a matter of selecting the preferred model from the UI.

Specifically in the Indian context, what kinds of enterprise problems are agents now able to solve—problems that RPA alone couldn’t address earlier?

Gautam Goenka: We’re seeing strong demand in India across sectors—from Global Capability Centres (GCCs) to manufacturing. Some challenges are uniquely local. For instance, many Indian companies still process handwritten invoices. This is a tough problem for traditional RPA, but agents combined with document understanding are proving to be very effective.

Another area is employee onboarding in the BPO sector, which has high attrition rates. Background verification is a huge pain point, and we’re working with customers to automate these workflows using agents.

SAP testing is another big use case here. India has a large footprint of SAP users, and our test automation capabilities, combined with AI agents, are being leveraged to speed up and improve reliability in testing.

How customisable is Agent Builder for industry-specific needs?

Raghu Malpani: We’re seeing a surge in use cases—dozens, if not hundreds—that customers are now exploring. Agent Builder is designed as a general-purpose platform, so customers can build virtually any kind of agent on it. To accelerate adoption, we also offer pre-built solutions or templates—these are industry-specific agents that solve targeted problems and help customers go to market faster. We already have dozens of these in our marketplace, and we’re continuously adding more as we discover new needs.

But beyond the platform itself, what’s becoming increasingly critical is the ability to offer verticalized solutions. That’s where we’re making significant investments—especially in India. Our engineering and product teams here are working closely with customers to build agents tailored not just for India, but also for specific geographies and industries.

Are there India-led innovations that have gained global traction?

Gautam Goenka: Yes, a few come to mind. One key innovation developed entirely in India is around agentic orchestration—specifically how agents collaborate with humans and robots. For example, if an agent is uncertain, how should it escalate to a human for review? That whole escalation and supervision workflow was designed and built by our teams in India.

Another great example is how we’ve used Agent Builder internally. We’ve deployed it to build customer support companions for our own portals. These agents help deflect queries by matching them with existing documents or solutions, which has reduced support ticket volume by 30–40%. So, we’re not just enabling customers—we’re also using our own tools to serve users better.

Is Agent Builder easy for existing UiPath customers to adopt?

Raghu Malpani: Yes, for our existing RPA customers, the transition is quite incremental. We introduce Agent Builder as an extension of what they already use. For instance, we define something called an identified robot, which customers already understand from the RPA context. So conceptually, it's a smooth learning curve.

That said, building agents does require a new set of skills—prompt engineering, for example, becomes very important. There’s also a need to put in guardrails to prevent hallucinations or incorrect actions. So yes, while it’s an incremental transition platform-wise, it does require some upskilling.

Are customers hiring for these new skills, or do they rely on UiPath?

Raghu Malpani: It’s a bit of both. Many customers are reskilling and upskilling internally. At the same time, for more complex, high-ROI use cases, they rely on our professional services team or large system integrators to implement agentic solutions effectively. So we see all models—customers hiring, leveraging partners, and working directly with us.

Gautam Goenka: And let’s not forget—we’re a low-code platform. We’ve made significant efforts to simplify agent development. For example, we’ve introduced features like Autopilot, which helps developers write better prompts, evaluate their performance, and iterate quickly. So even when new skills are required, we’re lowering the barrier as much as possible to support developer productivity.

How do you ensure smooth agent integration across systems like SAP and legacy platforms?

Raghu Malpani: Our strategy is to remain system-agnostic. Unlike some other agent-building platforms that focus solely on a specific vertical or ecosystem—say, only CRM systems—we don’t restrict ourselves. We support a wide range of systems. SAP automation, for example, is one of our core strengths. Salesforce as well. And all the integrations we’ve built for RPA carry forward seamlessly into the world of agents. We see this as a core advantage as we transition from traditional robots to AI-identified robots.

What lessons have you drawn from past automation challenges?

Gautam Goenka: Two key things. 

In traditional automation workflows, some steps are deterministic—perfect for RPA—but many are probabilistic or ambiguous. These parts were either handled manually or automated with limited effectiveness.

One key learning is understanding how agents can step in for those non-deterministic steps. The goal is to make agents either assist humans to boost productivity or take over fully when possible.

Another issue was UI fragility. In the past, automation would break if a UI element changed. Now, with AI, we can recognize these runtime changes—like a new popup or a moved element—and course-correct on the fly. These kinds of improvements, driven by past experiences, are helping us make automation more resilient and intelligent.

Will agents be able to learn post-deployment?

Raghu Malpani: We’re currently working on a feature in early preview that we’re calling “continuous learning.” The idea is to enable agents to learn from real-world inputs and feedback after deployment. We've put significant effort into building highly capable agents at design time, but like any AI system, there will be drift once they're in production.
Thanks to our large and active customer base—hundreds of customers building thousands of agents—we have both the volume of data and the opportunity to build a robust post-deployment learning system.

How do you evaluate agent performance in real time?

Raghu Malpani: It’s not just about success/failure. We use evaluation-driven development—feed known inputs and expected outputs during design. In production, human feedback plays a key role. We’ve added thumbs-up/down interfaces and traceability tools so developers can see an agent’s reasoning, actions, and outcomes.

Does agent success depend more on data quality or workflow complexity?

Raghu Malpani: Both matter. Agents rely on accurate contextual data from external systems. When things go wrong, humans can intervene. We also have live monitoring dashboards with metrics like pass/fail rates, execution time, and data lineage.

How are you ensuring safe and reliable agent operations?

Raghu Malpani: We’re investing in AI Ops—and more specifically, LLM Ops. Monitoring, guardrails, and improvement mechanisms are baked into the platform. For regulated industries, we allow rule-based checkpoints. For example, an agent trying to approve an invoice over ₹10,000 can be blocked until a human approves.

Do you think AI agents will remain assistive or become fully autonomous in the next few years?

Raghu Maplani: It will be a mix, depending on the use case. For example, customer support, case deflection, and supply chain workflows can be almost fully automated, with humans stepping in only for exceptions. The same goes for HR workflows. But these changes will happen gradually. Humans will shift toward more judgment-intensive tasks and use AI agents as powerful assistants.

With growing reliance on AI, what’s your outlook on sustainability?

Raghu Maplani: That’s a very valid concern. AI currently consumes a lot of power, but in just the last year, I’ve seen compute efficiency improve tenfold for the same tasks. Often, things get worse before they get better. The cost of AI computation will continue to drop, and we’ll see more adoption of clean energy sources, including nuclear power. So, while the environmental impact is worrying, I’m a rational optimist. With better technology and energy solutions, we’ll find a sustainable balance.

Final Word
With its flexible platform, deep enterprise integrations, and commitment to no-code innovation, UiPath is not just adapting to the AI era—it’s defining it. As agents become smarter and more autonomous, the company’s vision is clear: to make automation accessible, reliable, and truly intelligent.