Scaling Enterprise AI with Agents: Insights from AWS

Chad Watt: Welcome to the Infosys Knowledge Institute podcast, where business leaders share what they've learned on their technology journey. I’m Chad Watt, Infosys Knowledge Institute researcher and writer.

Today, I’m speaking with Dominic Murphy, North American Generative AI Solution Architect Leader at AWS, and Guruprasad N.V., Associate Vice President at Infosys and Head of the Infosys Topaz Center of Excellence. Both have deep experience in making artificial intelligence useful for enterprise organizations.

We’ll be talking about AI agents—what they are, how they're being deployed, and how enterprise leaders can build successful AI strategies. Welcome, gentlemen.

Dominic Murphy: Thanks for having me, Chad.

Guruprasad N.V.: Thank you, Chad. Great to be here.

Chad Watt: Let’s start with a bit of history. Between 2000 and 2020, how were you both putting AI to work?

Dominic Murphy: It was a progressive journey from predictive analytics to machine learning. We worked on recommender systems across industries—retail, media, travel, and more—focusing on better product recommendations and customer experiences.

We also built risk and fraud models for financial services, capital projects, and logistics. I recall working on fraud detection where a shipper claimed to send feathers but actually shipped car batteries—impacting weight and fuel costs.

In healthcare, we helped predict hospital readmission rates and risks of post-discharge mortality. It was all about using machine learning to serve customers, patients, and employees better.

Guruprasad N.V.: At Infosys, we’ve been on an AI-first journey since 2013–14, starting with internal transformation—HR, finance, onboarding, IT support, and employee experience.

We deployed a chatbot through our enterprise app, InfyMe, enabling natural language interactions. That had a big impact on employee experience.

We later applied AI to our learning platform and enterprise search. As we matured, we embedded these AI solutions into our offerings, which we now deliver to clients.

Chad Watt: Dominic, when did you become aware of generative AI’s broader potential?

Dominic Murphy: In 2018, one of my architects introduced me to the "Attention Is All You Need" paper about transformer models. Later, another architect built a simple mobile app called Bot or Not, using GPT-3. Employees would guess whether a response came from a bot or a human.

Seeing non-technical users engaging with it showed me the broad consumer potential of generative AI. That was a turning point.

Chad Watt: What is agentic AI?

Dominic Murphy: Agentic AI refers to generative AI models combined with tools and function calling to perform tasks on behalf of humans. Agents can operate 24/7, scale infinitely, and augment human work—improving customer experience.

For example, I once called a hotel concierge and resolved a Wi-Fi issue in seconds. Only later did I realize I was talking to an AI agent—it was seamless.

Chad Watt: What types of AI are you deploying in production now? Guru, let’s start with you.

Guruprasad N.V.: We’ve completed over 400 generative AI projects in the past 12–18 months. Key use cases include enterprise search, document summarization, retrieval-augmented generation (RAG), contact center transformation, and developer productivity with tools like Amazon Q.

In finance operations, we deployed agentic AI in our invoice-to-cash process—generating invoices, dispatching, and following up autonomously. It’s significantly improved cash flow and reduced manual work.

Dominic Murphy: Developer productivity is a major focus—writing, debugging, and migrating code faster, even across languages.

Document summarization and analysis is another high-impact area. For example, one energy client processes documents from 1,200 suppliers. Generative AI reduced decision-making time and freed employees for more strategic work.

We also see AI being used in personalization, synthetic data generation, content creation, and design assistants. And, of course, agents that handle workflows, research, and communication.

Chad Watt: When leaders approach you to start using AI, how do you guide them?

Dominic Murphy: I begin by understanding their desired business outcomes—customer experience, operational efficiency, innovation, etc. Then, we work backward to prioritize workloads that support those goals.

We assess AI maturity, readiness, data quality, and skills. We also build evaluation frameworks to measure impact—KPIs like cost, speed, and output quality.

Generative AI demands change management. It affects how people work, so adoption and training are critical. We also address responsible AI—bias, toxicity, compliance—and continuously measure and refine performance.

Chad Watt: Infosys recently studied ROI from AI and found results vary widely by industry and use case. Dominic, how can companies keep business value in focus?

Dominic Murphy: Start with a thesis: What value should this POC or pilot deliver? Then define clear KPIs upfront.

For example, if an analyst spends eight hours summarizing one document at $800 in cost, a generative AI system should improve speed, output volume, and cost-efficiency.

Track technical success and business impact. Then weigh those against development and operational costs to decide whether to scale the solution.

Chad Watt: Guru, what are some special considerations when deploying AI agents?

Guruprasad N.V.: Flexibility is key—AI tech evolves quickly. Observability is critical; we must track what agents are doing and how they’re performing using telemetry.

Responsible AI is a must—ensure compliance with local laws, mitigate bias, and protect privacy. Set up monitoring agents, dashboards, and safeguards to manage these systems at scale.

Chad Watt: Dominic, what are the biggest challenges with responsible AI?

Dominic Murphy: Operationalizing ethical AI at scale. Many firms have high-level principles, but lack concrete processes.

Technical teams often lack clear guidelines. Governance structures might not be designed for AI-specific risks. Collaboration across legal, technical, and business teams is essential—but alignment can be tough.

Laws are changing rapidly, so keeping up is part of the challenge.

Chad Watt: Guru, how should companies move from experimenting with AI to scaling it responsibly?

Guruprasad N.V.: Start with experimentation, but centralize when you scale. At Infosys, we use an AI Adoption Playbook—eight stages to scale AI responsibly and efficiently.

Use a platform-based architecture instead of siloed tools. This makes it easier to manage cost, performance, compliance, and adoption across business units.

Chad Watt: Dominic, where do you see the greatest potential for AI agents?

Dominic Murphy: It’s not about replacing humans, but enhancing our capabilities.

Developer productivity agents are already driving innovation velocity. Customer service agents reduce wait times and improve satisfaction.

Workflow automation across backend systems—CRMs, ERPs—will be huge. And data-driven insights from large document sets can reveal connections a human might miss.

We did one project at Amazon analyzing thousands of internal documents—an AI agent found patterns we never would have seen manually. That’s the future.

Chad Watt: Dominic, Guru—thank you both for a powerful conversation.

Dominic Murphy: Thank you so much. A pleasure.

Guruprasad N.V.: Thank you, Chad.

Chad Watt: This podcast is presented by Infosys in partnership with MIT Technology Review. Visit our Enterprise AI Hub at technologyreview.com to learn more.

You can find more details, show notes, and transcripts at infosys.com/iki in the podcast section.

Thanks to our producers, Christine Calhoun and Yulia De Bari. Audio by Dode Bigley.

I’m Chad Watt with the Infosys Knowledge Institute. Until next time—keep learning, and keep sharing.