In this episode of CGI’s From AI to ROI podcast, host Fred Miskawi, Vice President and Global AI Innovation Expert Services Lead at CGI, is joined by Imran Aziz, Senior Director of Product Management at UiPath, and Doug Vargo, Vice President of Emerging Technologies at CGI. Together, they explore the shift from “bots to brains”—that is, from traditional robotic process automation (RPA) to agentic artificial intelligence (AI)—and the strategic value this evolution brings to organizations.
The discussion covers what constitutes agentic AI and provides real-world business applications across sectors. It also offers insights on navigating the “bots to brains” shift through a strong foundation in governance, data readiness and workforce transformation.
Key takeaways from the episode:
1. Agentic AI is redefining intelligent automation beyond RPA
The transition from rule-based automation to agentic AI represents more than just a technical advancement; it marks a shift in how organizations approach problem-solving and efficiency. Traditional RPA systems followed hard-coded, deterministic instructions, whereas agentic AI introduces semi-autonomous, adaptive capabilities that can make decisions and dynamically generate outcomes based on general prompts.
Doug Vargo notes that agentic AI represents "intelligent digital workers capable of managing independent business processes," though he emphasizes they are still semi-autonomous with humans remaining "in the loop or on the loop" to guide and validate decisions. Imran Aziz further explains that, unlike earlier systems where paths through a process were predefined, modern agents are built on large language models (LLMs) that can “layer on more semantic capabilities” and reason their way through problems dynamically, opening the door to previously unattainable enterprise use cases.
2. Governance and data readiness are critical to scaling AI solutions
While the promise of agentic AI is substantial, it can be realized only if organizations build on a strong foundation of governance and data integrity. Without proper data preparation, role clarity and compliance measures, even the most sophisticated agents may fall short of enterprise requirements.
Imran Aziz underscores that governance goes beyond basic access control. He describes a real-world scenario in the commercial loan industry where it’s vital to "make sure the data is isolated so that roles can view only relevant information" and stresses the need for auditability, compliance and full activity logs for both agents and human actors. Doug Vargo adds that “use case absolutely matters” and outlines key considerations such as data quality, integration complexity and organizational readiness. Without robust attention to data privacy and regulation, he cautions, “we're going to have lawyers in the loop, too.”
3. Industries like life sciences and manufacturing are achieving real-world value
Agentic AI isn’t theoretical; it’s already delivering measurable impact across various sectors by reducing error rates, accelerating outcomes and scaling previously manual workflows. Organizations that have already embraced RPA are finding that agentic AI builds naturally upon their existing capabilities.
In the life sciences sector, Imran Aziz shares a case where a large company used agents—both human-assisted and fully autonomous—to improve plasma donor onboarding. Given the high regulatory standards, this initiative not only reduced costly errors but also improved process throughput. Similarly, in manufacturing, he highlights a shift from multi-day quoting processes to near-instant turnaround, thanks to agent-powered due diligence tools.
Doug Vargo elaborates on CGI's own success with software development life cycle (SDLC) acceleration. Using agentic AI, clients can automate requirements gathering, generate wireframes, streamline code creation and synthesize test data. He notes that clients are seeing productivity gains of 20–30% in development and testing, marking a significant transformation in how enterprise software is built and maintained.
4. Preparing for an agentic future requires a people-first approach
As digital workers become embedded within business operations, human workers must be equipped with the skills and mindsets to collaborate with these new tools. Upskilling and organizational change management (OCM) are essential to enabling a smooth and inclusive transition.
Doug Vargo emphasizes the importance of persona-based enablement, saying that “everybody from a developer, to a tester, to your sales team or even your shop floor worker” will interact with AI tools differently, and training must reflect these nuances. He stresses the need for OCM strategies that are tailored specifically to AI adoption. Imran Aziz adds that identifying areas of low employee engagement—like administrative work in healthcare—is a great place to begin deploying agents, as this approach increases both employee satisfaction and operational efficiency.
By aligning technical innovation with employee empowerment, organizations can move confidently into a future where digital and human intelligence work hand in hand.
From bots to brains: Navigating the RPA and agentic AI convergence
- Introductions
-
Fred Miskawi (00:00)
Welcome, everyone, to From Bots to Brains, navigating the RPA and agentic AI convergence. In this rapidly evolving world of automation, we're witnessing a pivotal shift—the move from task-based robotic process automation to more intelligent decision-making capabilities that you find in agentic AI.
It's not just an incremental change. It's a significant leap forward, promising to redefine how businesses operate and innovate. I'm your host, Fred Miskawi, VP and Global AI Innovation Expert Services Lead at CGI.
Today, we're privileged to have two experts on this topic. First, we have Imran Aziz, Senior Director of Product Management at UiPath, a recognized leader in automation.
And joining him is Doug Vargo, VP of Emerging Technologies at CGI in the US.
And I'm going to start with maybe a little bit of deeper introduction to our guests. So, I'm going to hand over to you, Imran.
Imran Aziz (00:59)
Thanks for having me, Fred. I lead AI products at UiPath, building products such as Autopilot and also vertical agentic solutions. Prior to UiPath, I've also worked at Apple and Microsoft, building more traditional software products and leveraging AI.
Fred Miskawi (01:16)
Thank you, Amran. And Doug, what about you?
Douglas Vargo (01:18)
First, thanks for having me, Fred. It's a great pleasure to be with you and Imran. I lead the emerging technologies practice here at CGI. And we focus a lot on modernized platforms and what we're going to talk about today doesn't happen if we haven't modernized our overall ecosystems by that journey to the cloud. And then, there's the data management and everything we talk about, will be rooted in traditional data management and data concepts and build upon that with AI which I'm really excited about for the future. I'm excited about where we are today with generative AI.
And I can't wait to see what's next with agentic AI as organizations start to roll out these transformational technologies.
Fred Miskawi (01:59)
Thank you, Doug. A full stack of maturity, but let's go back to the basics. We talk a lot about agents. We keep hearing about agents everywhere at this point. We've been talking about it for a little over two years and the definition has a little bit evolved over time. So, I think it's important to kind of cover the basics. Doug, from your perspective, what is an AI agent?
- Definitions of AI agents and vertical agents
-
Douglas Vargo (02:20)
That is a great place to start this conversation. And you talk about two years and I reflect back on, I think it was just a few months ago, maybe the beginning or the end of last year where Sam Altman came out and said, 2025 is going to be the year of the agent. And that got a lot of buzz across the IT communities. And I've had the pleasure to talk with many clients and where they're going with their current AI journeys. And they all want to talk about agentic.
And when they start to talk about agentic, what we find out very quickly is that there is a lot of different definitions around agentic AI. And it is a living definition and commonly misunderstood. I'll start off with the simple definition.
First of all, they are intelligent digital workers. They're capable of managing independent business processes. But what's important is they're semi-autonomous. Many definitions out there, you'll hear autonomous agents. I don't believe we're there just yet. We are always seeing a human in the loop or a human on the loop as part of the agentic definition.
But what's important, in addition to their digital workers, is that these agents are capable agents that can learn. They can adapt. They produce dynamic outputs.
The other important aspect is that they don't require explicit inputs. It's not like with traditional RPA and intelligent automation. They really, instead, receive a general prompt. These agents can then create a plan and then complete tasks leveraging AI tools on the backend.
Fred Miskawi (03:59)
Thank you, Doug. So, let's take that definition. Let's put it in the context of the enterprise. Let's add a little bit of context. And now, we're going into a different name for agents. We often refer to them as vertical agents. Imran, can you give us a definition of what a vertical agent is?
Imran Aziz (04:16)
Yeah, I really like the definition of an agent. To carry that forward, vertical agents to me are ones that can perform specialized business functions. And you can sort of imagine, traditionally, you've had to buy vertical SaaS software, train employees to perform those functions using both productivity and these vertical software.
And now, with the agentic capabilities, you can perform these functions either by assisting humans to do it or in an automated fashion. So, vertical agents perform specialized functions versus an agent, like Doug was defining, as a broader horizontal use case.
Fred Miskawi (04:56)
Yeah, thank you, Imran. And I suspect we're going to see new specializations or role specializations connected to this concept of vertical agents and to the orchestration and ecosystems that we're deploying in the enterprise.
Doug, so you were mentioning RPA, and you've been working quite a bit in this interaction between RPA and agentic AI. How has the automation landscape evolved over time?
- The evolution from RPA to agentic AI
-
Douglas Vargo (05:20)
Yeah, significantly Fred. It's significantly evolved from the traditional RPA technologies. We went into intelligent automation capabilities now to generative prompt-based solutions that we're seeing. And as we look back and we look at RPA, those were rule-based. They were hard coded instructions of telling an automation what to do.
And we saw a lot of success with RPA, but then we got ended up graduating to that intelligent automation and technologies, like document understanding where we're able to intelligently extract information, for example, out of content and then take action on those content, leveraging those inputs and outputs that we already desired.
And now, we're at the generative AI or that prompt-based automation where we can look at better patterns and the models can ‘think’ for themselves to be able to derive the outcomes based off of the input that are driving forward. And we saw the emergence of some of this with those low code configuration platforms and different interfaces that were out there to get those creative outputs that many organizations now rely on from an overall transformation perspective.
And as we start to get into agentic, we will get to the more semi-autonomous. How do we take independent automations that we depended on, and string them together? That way, from an end-to-end solution, we are achieving an overall outcome.
And that is the promise of digital transformation. It is giving those compound transformations to truly evolve your overall workflow and autonomously do that. I don't think we are there from an end-to-end autonomous or fully autonomous. However, from a semi-autonomous or human in a loop, we're already seeing organizations stacking agents one after the other to achieve tremendous outcomes.
Fred Miskawi (07:20)
The kind of funny thing for me is we've used the term intelligent automation for many years. And I think maybe that was a little future forward looking. Now this is truly intelligent automation, but we can't use the term anymore because we've used it all these years.
But Imran, so you're smack in the middle of this transition that we're seeing evolve in front of our eyes within the enterprise.
Imran Aziz (07:34)
Mm-hmm.
Fred Miskawi (07:43)
How are you undergoing your own transformation from RPA to an agentic future?
Imran Aziz (07:48)
Yeah, I was going to say, RPA, in a way, paved the path towards agents, both the horizontal agents that Doug mentioned and the vertical agents. And the reason is because, RPA, through automation, we've innovated on a basic ability of how you take what humans do and both automate some of the UI APIs that connect different heterogeneous systems and applications to perform tasks, leverage screen reading, document understanding technologies, to be able to get the data flow structured so that robots and now, agents, can act on that data and make sense of it.
Some of these RPA capabilities have had to be cross- platform so that they can work across whether you're using a Unix system, Mac, Windows. You have different databases deployed. And then governance, like I think you mentioned as well, RPA systems because they work with enterprise data, medical data, manufacturing data. They have to be compliant with what the enterprise needs are and highly secure and privacy compliant as well.
So now as we transition to agentic, so in a way, it was easy because RPA had led the charge on a lot of these capabilities. But now, what agents allow you to do is they give you the flexibility.
With classic computer science, ML technology and RPA, we leverage more rule-based deterministic technology where a developer had to code in some building blocks and then stitch them together into an end-to-end flow. With agents and reasoning that these LLMs can do, we can layer on more semantic capabilities. So, you don't have to pre-define all the paths through the system. The agent can reason and figure those things out dynamically.
And what that does is, I think it opens a broad set of capabilities and use cases and business functionality that wasn't possible before. And we'll talk a little bit more in detail about that. But the way we've sort of layered on top of RPA is the capabilities of document extraction, API integrations, human in the loop is still key. It was key in RPA and is going to be especially key in agentic world because enterprises need that confidence that there is a person who's a domain expert in charge of these workflows.
And then, last but not least, is the orchestration of these processes together.
So, I think just to recap, I would say agents existing side by side with robots and humans who are domain experts who can make decisions is the framework of the future.
- Real world examples of transformational impact and governance
-
Fred Miskawi (10:19)
Thank you, Imran. And that brings us to our second segment, real world examples of transformational impact. We talk about governance and we talk about kind of the best practices and what we need to do as we move forward and deploy this technology. But from your perspective, Imran, what are some of the forgotten things that are not necessarily obvious, that we don't necessarily address right away, but they end up being a critical part of the success and the scaling of these solutions.
Imran Aziz (10:50)
Yeah, governance is super critical. Let me ground it in an example. So, for example, when you have commercial loans being processed, there are underwriters and loan officers who are processing them. And then, there are people who might do post-compliance checks, which is sort of spot checking how those loans are processed.
So, governance is not only just being aware of roles and responsibilities of who's carrying out the business function but making sure the data is isolated so that roles can only look and view that information. And then, the workflows that they're responsible for executing or the agents also have sandbox information that they can process.
But at the same time, for audit and compliance, you might want to just make sure there's a full log of what the agent or human is doing and then be able to sort of spot check at a later point.
So, I think governance is a fairly complicated topic. But for the enterprise roles, responsibilities, auditability and compliance are some of the big ones, along with making sure PII is treated extremely carefully.
Douglas Vargo (11:55)
Yeah. And just to add to that, when I think about governance, and I really started to think about use case, and use case absolutely matters when it comes to agentic AI to get that return on investment that every organization is looking for.
And then, when I talked about a lot of those critical characteristics that we need to assess as part of every business use case. And for us, it really starts with the readiness of the organization. If you start to deploy these agentic AI solutions, or even more traditional AI solutions, is your organization ready to adopt that change? And then, you start to think about other aspects around data and information readiness.
And so, many times we start having conversations with our clients around agentic AI or AI. And it always falls back to, is your data ready? And that confidence, or that trust stack that organizations are trying to build solutions on top of that. And then, the complex integrations and the technical assessments and the readiness of your overall modernization.
And then, I have to talk about regulations and data privacy, because you're talking about human in the loop. Well, if we don't get regulations and data privacy right, we're going to have lawyer in the loop too. And it's important that we start to make sure that we evaluate all of these different use cases across these critical components to truly understand where can we get transformational value for the organization and get that ROI that everybody at the executive board is looking for when they approve these AI initiatives.
Fred Miskawi (13:36)
I love lawyer in the loop, very appropriate in the legal industry. So, as we're going into this next segment, we've talked about examples and I'll open this to both of you.
When we talk about RPA and this transition from pure RPA into this hybrid environment and eventually into a pure agentic ecosystem, there's definitely one core example of value that we're seeing in the industry today and that's related to software acceleration, software development.
Imran, what have you seen on your side as it relates to that topic?
Imran Aziz (14:11)
So, I think software development and coding agents are definitely super important and accelerate time to market for a lot of solutions.
We've been focusing more on industries such as manufacturing, retail, banking, and also healthcare. And in these industries, there's a lot of repetitive processes which RPA already automated, but there was a point where it couldn't go further.
And now with the agentic platforms, we can advance those use cases. I'll give you a couple examples. We're working with a large life science company where they onboard patients to do plasma donations, which end up delivering life-saving medications at the end of the process.
It's a highly compliant industry where there's a lot of rules and regulations on who you can onboard and double-check. Often, when you have thousands of people working on this from a onboarding perspective, there are mistakes that can be made. And those mistakes can be super costly, including being able to take medications off the shelf if there's an error.
So, what we've done now working with this company is enable them to use agents, both human-assisted agents and also unattended agents to be able to process that information, reduce the error rate and increase the throughput at which you can onboard blood donors.
There's another example in the manufacturing industry where we're working with customers to reduce the time it takes for someone to ask for a quote on a complex order and be able to, in the traditional days, it would take human salespeople and other internal folks to be able to come back in days with the quote, for, let's say a multimillion dollar order. And now with agents, that time is reduced significantly because a lot of the due diligence can be done by agents and a human just needs to approve it or be able to modify that with some autopilot or copilot kind of capabilities.
Fred Miskawi (16:00)
Imran, I think what you're saying is this is not just hype. We're actually seeing this value in the enterprise today and it's continuing to accelerate.
Doug, let's go over to you.
- Accelerating go-to-market by transforming software development life cycle (SDLC)
-
Douglas Vargo (16:12)
Yeah. I see it, not only in the work that CGI is doing internally, but I see it with our customers and where we get the most excitement is around that SDLC acceleration. When I think back at intelligent automation, it's something we've been doing for years with UiPath, where test automation and leveraging test suite to be able to automatically create the test script scenarios, manage those test scripts, and then autonomously execute those in a UAT fashion.
And now with agentic AI, we are able to apply agents across the full end-to-end life cycle. And it starts very early at the planning phase and leveraging agents around automated requirements gathering and documentation. That is a painful process in many ways. And now, we're able to leverage traditional business process automation technologies, in addition with agents, to fully understand the ins and outs—the data flow, the swivel chair activities that are happening to be able to gather those requirements and document and then, draft those requirements specifically to how an organization likes to see that in a very accurate and consistent manner.
And then, once we have those requirements, in design and conception stage, leverage AI to mock up wireframes and to take what we hear from the business, and develop real mockups of a UI/UX design screen, and very quickly turn those around based off of brainstorming sessions.
And then, once we get that design, we get into the development. And we're seeing a lot of focus with code generation—being able to automatically generate the code, orchestrate the code, manage the code in a very highly accurate way, and facilitating a lot of transformation for the traditional developers and being able to do that in a much more efficient way. And then, sometimes, we're seeing 20 to 30% productivity gains from our developers in their overall coding.
We talked about testing and testing is another area that yes, we had success in the past. We are using intelligent automation, but now we're getting AI-driven test case generation, being able to create synthetic data, which is so important. It's so hard to find scenario-based or contextual-based data that we're basing our solutions on. Now with these technologies, we can leverage that within our end-to-end orchestration flow, leveraging agentic AI.
And at the end, maintain these systems with seamless integration of AI testing and our overall CI/CD pipelines. So, I just walked through the whole end-to-end and how these agents are able to truly transform the SDLC acceleration. And as I mentioned earlier, many times, those are still in, I'll say, in silos of between each of those phases. But now, we're starting to orchestrate those together with human in the loop, to get those big outcomes that we're all looking for.
Fred Miskawi (19:07)
Thank you, Doug. And a big part near and dear to my heart as well. It's going to be the topic of many podcasts in the future.
- Preparing for the agentic future
-
So, let's close this podcast talking about a one final question to both of you. And I'll go over to you, Imran. What should organizations prepare for this agentic future?
Imran Aziz (19:23)
That's a great question. When I talk to customers across manufacturing, healthcare, and life sciences, one thing that's clear is that, I think Doug alluded to this, as well, is that you need to make sure that if there's fragmented data, it needs to be normalized. The cleaner that aspect is for input, the better for the AI systems to process.
The second thing would be security and compliance. Just be prepared to know how the AI system should process the data or the compliance checks and the rules and responsibilities for human in the loop that are important.
And then, the third is, really, just know the use cases that are the key pain points for the organization. That can be the first leading charge for AI. And if you've already got an RPA implementation, I think that's easier because then, you've already identified things that robots can do and then, agents can take to the next level.
Fred Miskawi (20:23)
Thank you for those words of wisdom, Imran. Doug, let's go over to you.
Douglas Vargo (20:26)
I heard a quote this morning, which kind of took me back, but it's so true. We are the last generation to have a fully human workforce. And if you let that sink in, for me, it's inspiring. I'm excited to see where we're going in the future, but we're the last generation to have a fully human workforce. And we are going to be assisted by digital workers or agent workers as we move forward. So, we must get prepared.
The biggest part is getting your talent, your employees, upskilled on these technologies. And I'm responsible for a lot of the AI enablement inside of CGI. What I've learned is, make sure that as you start to roll out technologies, you make them persona-based.
Everybody, from a developer to a tester to your sales team or even your shop floor worker, they will have a different way of interacting with these tools. And you need to make sure as you're training and rolling out these tools, you understand the nuances of persona-based training and make sure that you have a solid organizational change management approach that is adapted for AI. And it is a very specific skill set. So, it's just important that we focus on upskilling our employees and get them AI ready for that future AI workforce.
Imran Aziz (21:49)
One thing, if you don't mind me adding, I think if there's a department or a survey in the company where people are not enjoying what they're doing. So as an example, doctors don't necessarily enjoy filling out paperwork or notes with patients and getting that into the system. They'd rather do more work with patients. Then, those areas are perfect for agents and AI to solve.
Because then, you increase the satisfaction of employees. You supercharge their work and you're taking care of the monotonous tasks that people don't want to do.
Fred Miskawi (22:22)
And this is a topic that we've been addressing, right? For the past couple of decades, with RPA looking for those repetitive processes that are not necessarily pleasurable to the individuals that were progressing them or operating them and bringing a little bit of relief from that perspective.
That brings us to the end of a really insightful discussion on the convergence of RPA and agentic AI. If there are a few core messages to our listeners that they should take away from today's conversations, it might be those.
Number one, the strategic convergence of RPA and agentic AI is no longer a future concept. We are going through this transition today.
Next, the real world adoption of these integrated solutions underscore significant ROI. This isn't just about efficiency gains. It's about fundamentally reshaping enterprise operations and bolstering strategic competitive positioning in the market.
And finally, strategic partners and their insights truly matter for successful adoption. The path to intelligent automation is dynamic and as we've heard, full of opportunity.
So, thanks again to both of you, Imran and to you, Doug. And to our listeners, thank you for tuning in.
Imran Aziz
Senior Director, Product Management | UiPath
Imran leads AI products product management at UiPath, including the management of Autopilot and vertical agentic solutions. He has over more than 15 years of experience in leadership positions at Microsoft, Apple and Meta. Prior to UiPath, he worked on a wide range of products such as iOS Notes, iBooks, Bing, Meta’s Ads platform, and Windows Workflow. He is a co-author of over more than twelve U.S. patents in the domain of modern applications, internet search, e-commerce and workflow systems.
