Episode Transcript
Dave Ferrucci: The reality is—and I’ll say something maybe somewhat controversial—the most intelligent conversations I can have today are with AI.
Sastry Durvasula: There’s no question that the level of cognitive tasks and cognitive workload this technology is doing is absolutely unprecedented. I just don’t want us to also overhype it to a point where it never happened before.
Gary Hoberman: Welcome to Architecting the AI Enterprise. I’m Gary Hoberman, founder and CEO of Unqork, joined as always by my co-host, Dave Ferrucci, our Chief Technology and Chief AI Officer. In this series, we bring top leaders together to explore what it really means to put AI to work at enterprise scale: the architecture, the decisions, and the lessons you won’t find in a press release. If you’re leading technology strategy in the enterprise, this is the show built for you. Let’s get into it.
Hi, welcome to another episode of Architecting the AI Enterprise. I am Gary Hoberman, CEO and founder of Unqork.
Dave Ferrucci: Hi, Dave Ferrucci, Chief AI Officer and CTO at Unqork.
Gary Hoberman: On this podcast, we bring the most amazing leaders to you to hear their stories and their journeys. There are very few I’m prouder to bring on than Sastry, whom I’ve seen climb the corporate ladder since I first met him as a digital officer at Marsh. We pitched him. I’ve gotten to watch Sastry move from there to McKinsey to TIAA. At TIAA, we’ve spoken together on panels even recently, where you moved up to be the Global CIO. That was just a few weeks ago when we planned for this podcast, and all of a sudden, in just a few weeks, you’re now the Chief Operating Officer at TIAA. For those in the technology world, COO is the ultimate next step from CIO—though CEO comes next, and if there’s anyone that could do it, I know you can.
This is going to be interesting because we want you to tell your story outside this room. We were just having a conversation about C, VB, and MSDN back in the ’90s.
Sastry Durvasula: I know, man.
Gary Hoberman: Give us the story of how you got here.
Sastry Durvasula: Yeah. You know, I started as a software engineer and architect, so I did a lot of work pre-internet or during the early stages of the web. As we were running campaigns like “Drive to Web,” you would have a website built, launch a big marketing campaign to bring people to the website—obviously pre-Y2K.
Gary Hoberman: And—
Sastry Durvasula: Then the Y2K thing happened, and that started a different set of discussion threads. But I did a lot of coding back in the day, and I still like to code—though now you have agents running for you.
I grew up the corporate ladder leading technology delivery, large-scale architecture initiatives, large-scale transformation initiatives, and then ultimately moved into various CIO and CTO roles. Then this thing called digital transformation happened, where every company realized that they needed to go genuinely digital. That’s when I started getting into digital transformation roles at American Express to reach our customers where they were. This was the advent of social media and SoLoMo—social, local, mobile—if you remember that.
Gary Hoberman: I do remember that. I actually have a vivid memory of writing our first digital strategy for the company, which was crazy back then.
Sastry Durvasula: Those were my days at Amex. We built a range of capabilities like Amex Offers in the app, and digital Pay with Points when you go to Amazon or use your Membership Rewards points in NYC taxis and the Uber app.
Gary Hoberman: Those were fast transactions.
Sastry Durvasula: Think about massive mainframe systems running Membership Rewards in the back end, combined with traditional payment networks. You’re basically making an instantaneous decision on which rails to use, making it seamless, and getting back to the customer that, yes, you just bought this on Amazon or completed a New York City taxi ride.
Those were my digital days, alongside the advent of big data when Hadoop was the big thing. If you dig deep into the internet, you’ll find an article from the Strata conference at the Javits Center quoting me saying, “Hadoop is the new backbone of American Express.”
Then you saw me at Marsh. I was hired to build the first digital strategy and digital organization, covering payments and financial services for the largest P&C insurance broker on the globe. I had an opportunity to work with insurers and major global clients. From there, I went to McKinsey, where I ran technology and digital internally while serving as a senior advisor to technology, financial services, insurance, private equity, and fintech clients.
Then I joined TIAA to lead the global technology and global client services organization. Since then, as you said, Gary, I stepped into the Chief Operating Officer role.
Gary Hoberman: Not many people know this, but TIAA manages trillions in assets.
Sastry Durvasula: Yeah, $1.5 trillion right now. If all goes well, we will be close to $2.5 trillion in AUM following our acquisition of Schroders, a major global asset manager, joining forces.
The story of TIAA is amazing. Andrew Carnegie realized that teachers and healthcare workers—primarily teachers back in Carnegie’s day—dedicated their lives to educating others yet lacked a pension plan. He set aside a million dollars to catalyze something for this community. One hundred and eight years later, we are a global Fortune 100 firm helping millions of teachers, healthcare workers, and administrators, while investing that capital through our asset management arm. It’s a truly mission-driven company.
Gary Hoberman: In your role as COO, what are your core responsibilities?
Sastry Durvasula: Largely enterprise operations for the entire company across our three primary businesses: our core retirement business, asset management, and wealth management. That covers B2B, B2B2C, and B2C.
I run enterprise operations and business transformation across all three. We have large-scale modernization efforts underway, whether across front, middle, and back-office operations in asset management, a major record-keeping transformation in retirement, or ongoing modernizations in wealth. Then, of course, there’s the AI transformation.
Gary Hoberman: That’s the exciting part.
Sastry Durvasula: From an organizational remit perspective, I lead global technology, global digital and client experience, client operations, fraud operations, shared services, and our global capability centers. It represents roughly 50% of the firm’s total workforce.
Gary Hoberman: As an Amex customer and with my mom having taught in New York City public schools until retiring at 81 with TIAA, it’s amazing to see this from all sides. A truly mission-driven organization. I didn’t realize the Carnegie connection.
Sastry Durvasula: When you visit Charlotte, our office is on Andrew Carnegie Boulevard.
Gary Hoberman: Amazing. Getting back to AI—Dave, I’ll let you ask the first question. Let’s drive the conversation.
Dave Ferrucci: Sastry, in your position, how do you see AI affecting the workforce as this plays out, especially in our industry? There are so many perspectives with tremendous variation, and that shift won’t happen overnight. How do you see that evolution?
Gary Hoberman: Where is it helping? Where is it not? Where do you think it’s going?
Sastry Durvasula: It’s a layered question because you have the workforce, the workflows they use, and the end clients you serve.
I developed an 80/20 hypothesis that I often get quoted on: 80% of jobs will change by at least 20%, and 20% of jobs will change by as much as 80%. As a former consultant, I have to come up with frameworks like this! But if you look back at historical transformations—electricity, the internet, mobile—you could apply similar logic.
There are roles that will follow the path of lamplighters, steam engine operators, or typists—specialized roles that will be impacted by 80%. When looking at the workforce, roles like strategy consultants or human resources officers that require extensive interpersonal interaction and customer service will see different types of shifts.
The impact also depends on enterprise readiness. If you are a new, cloud-native startup, it’s easier to disrupt at speed and scale. But for a 100-year-old enterprise with significant regulatory and architectural complexity, that workforce transformation takes longer and looks very different.
Dave Ferrucci: Many historical examples apply, but one could argue that AI is fundamentally different. It isn’t just targeting perfunctory tasks that previously required manual human labor; it is competing directly with core human cognition. I don’t mean to use adversarial phrasing—that touches on broader debates around existential risk—but looking purely at task automation and business impact, AI is unique in its ability to compete with us on a cognitive level. How do you think about that distinction, especially with the rise of agentic AI? There may not be a clean historical parallel.
Sastry Durvasula: I agree there isn’t a direct parallel. Electricity and the internet didn’t take on cognitive tasks.
You could ask: is an ATM a smart machine? We used to stand in line for a human teller who counted out your money. Now an ATM handles that, trust is established across banking networks, and we’ve added biometrics, facial scanning, and automated touchpoints.
Gary Hoberman: Like airport security setups.
Sastry Durvasula: Exactly. That used to be a person manually reviewing documents; now automated systems verify your identity instantly. In medicine, machines can now read scans that once required a radiologist’s manual review.
Dave Ferrucci: Just to press on this: pattern recognition has continually advanced, and we’ve accepted machines doing that. What remains unique is that human language and human cognition are so deeply interwoven. Language defines how we evolve cognitively. The reality today—and I’ll say something potentially controversial—is that the most intelligent conversations I can have right now are with AI.
Sastry Durvasula: The cognitive workload this technology handles is absolutely unprecedented. My only caution is not to overhype it to the point of claiming nothing like it has ever occurred.
Before electricity, a master baker practiced an artisan craft. When electrical appliances arrived, half their manual tasks were automated, but the artisan element remained.
To your point, Dave: this shift is hitting hard. I use AI every day as a thought partner. There are internal thoughts and draft ideas I wouldn’t run by either of you that I run through Claude or ChatGPT first.
Gary Hoberman: I do the exact same thing, and Dave does as well. Our daily usage often centers on validation, rewrites, edits, or analyzing complex correlations. It excels at synthesis.
Sastry, I have to ask about another massive topic: the SaaS apocalypse. We see headlines suggesting Salesforce is being disintermediated by models like Claude, while frontier labs are debating whether capabilities are plateauing or need to slow down.
Sastry Durvasula: It’s a fascinating time. Every week brings a new headline.
Gary Hoberman: In your view: is SaaS dead? What is happening to SaaS?
Sastry Durvasula: That’s the right question to ask someone named Sastry—especially since I used to be a technical advisor for the analytics firm SAS!
This debate will be as fascinating as the ongoing “when will the mainframe die” discussion. The mainframe is still running. It runs in my shop, and it runs across most of the 100-year-old institutions we’ve worked with.
With SaaS, the transition will be dramatic. I don’t take an extreme pro- or anti-SaaS stance because some providers will thrive and others will die. If you look at the architectural evolution—from mainframes where UI, business logic, and databases lived on one system, to client-server, distributed computing, the web, and cloud—we went from monolithic to ubiquitous computing.
Now look at SaaS: will the human-seat-driven model survive? I don’t think so.
Gary Hoberman: The interface itself?
Sastry Durvasula: The system of action. You have systems of record, systems of intelligence, and systems of action.
The system of action is where traditional SaaS vendors sell per-seat enterprise licenses across CRM, ERP, and ITSM. That layer will be disintermediated because autonomous agents—not humans—will be querying and updating those systems.
This connects to the idea of Salesforce being consumed through Claude. Claude becomes the front end. You tell the agent, “Prepare me for this client meeting, pull recent team activity, and summarize open issues.” The agent queries Salesforce in the background.
The system of intelligence holds business logic, governance, compliance rules, and regulatory controls. That layer retains value in the near term, though AI will steadily streamline it too.
Then you have the system of record. Do we still need ERP, CRM, HR, and incident management databases? Absolutely. But in what form? SaaS companies will deploy AI agents directly on their chassis, enabling agent-to-agent consumption.
Today’s front end is tomorrow’s back end. If you’re a marketer, your primary interface becomes an intelligent dashboard, with legacy SaaS systems running behind it.
Gary Hoberman: Many legacy SaaS companies haven’t executed the one necessary move: a foundational reset. Their core architectures haven’t meaningfully changed in 25 years.
Sastry Durvasula: They moved to the cloud and evolved the tech stacks, but—
Gary Hoberman: The data structures and schemas remain the same. If they stepped back, realized their true IP is those underlying data structures, and redesigned them for an agentic paradigm, they would have to risk existing subscription revenues to bring customers forward. Otherwise, these structures remain inefficient for autonomous agents to interact with.
Sastry Durvasula: Some incumbent firms will evolve, though they will make mistakes along the way. I remember when legacy database companies insisted customers buy proprietary on-premise hardware appliances because the cloud couldn’t match their security or performance. Today, nobody talks about database appliances.
Tech providers must be willing to disintermediate and cannibalize their own revenue streams. Microsoft and Oracle have shown an ability to reinvent their models across tech cycles.
Dave Ferrucci: Moving past SaaS to how LLMs are consumed: there is a noticeable tension around enterprise commitments. On one hand, companies commit massive spend to Anthropic or OpenAI, instructing teams to burn down those tokens. On the other hand, enterprises find model selection and cost optimization overwhelming, preferring vendors to abstract away the AI complexity and simply deliver an SLA-backed business outcome.
Gary Hoberman: When we built Unqork, we committed to being cloud-agnostic. The same logic applies to models: should enterprises remain model-agnostic and switch dynamically, or should leadership standardize on specific models?
Sastry Durvasula: You have to remain model-agnostic. Having an intelligent, dynamic orchestration layer that abstracts models away and optimizes tokenomics is critical.
Dave Ferrucci: “Model-agnostic” sounds appealing, but it implies the underlying LLMs are interchangeable commodities.
Gary Hoberman: In many respects, they are becoming commodities.
Dave Ferrucci: Perhaps, but whether AI becomes contextually embedded in vertical solutions or remains an undifferentiated commodity is an open question.
Sastry Durvasula: It ultimately depends on workloads and job families. Even within a model-agnostic architecture, what a cancer researcher needs versus a software developer or marketer differs dramatically.
Whether models are proprietary, open-weight, or open-source, the real questions center on token consumption, pricing, orchestration governance, and avoiding unnecessary layers of vendor disintermediation.
Look at the developer’s role: a mainframe engineer faces a very different reality than a mobile or front-end developer, and some traditional coding roles will cease to exist.
Gary Hoberman: It’s refreshing to hear a business leader who understands architecture answer that question with technical precision.
Sastry Durvasula: You wouldn’t burn expensive tokens on a high-parameter frontier model designed for biotech discovery just to write boilerplate CRUD code. Token governance, cost-to-performance matching, and security remain critical considerations.
Gary Hoberman: Let’s turn to security.
Sastry Durvasula: It brings us back to the early days of enterprise Linux. Dave, you remember this from IBM: how do you verify that open-source components won’t introduce critical vulnerabilities into your core systems?
Dave Ferrucci: With agentic AI, that’s the multi-billion-dollar question. When approaching a requirement, should you prompt an autonomous agent or generate deterministic code? My concern is that the rapid democratization of AI means people are building production systems using agentic approaches without understanding the difference between probabilistic token prediction and deterministic software engineering.
Sastry Durvasula: That divide is critical. Think about the evolution of document creation: I wrote my master’s thesis in LaTeX, compiling source code with manual formatting tags into PostScript files. Today, an LLM drafts long-form content from a single prompt.
Over the past thirty years, humans have steadily outsourced mission-critical decisions to machines once trust was established. In digital payments, moving funds instantly via Zelle requires complete confidence that the transaction executes compliantly without compromising underlying accounts.
We will overestimate what AI changes in the next two years, but we will radically underestimate the shift over the next ten. Decades ago, nobody envisioned getting into a stranger’s car via Uber or allowing delivery workers into their homes. Societal boundaries around trust evolve.
Dave Ferrucci: We must also evaluate automated systems against baseline human error rates. In autonomous driving, an incident involving a self-driving vehicle makes international headlines, yet human error causes the overwhelming majority of road fatalities every day.
Sastry Durvasula: Exactly. The headline is always the autonomous vehicle, not the human baseline.
Gary Hoberman: I recently took my daughter for a test drive in a new Tesla and told it to find a parking spot at Costco on a weekend—it handled it flawlessly. We’re keeping humans in the loop for now, but that baseline is moving fast.
Sastry, with technology reporting up to you as COO, how does the technology function look today versus five years from now? What is your expectation for the modern CIO?
Sastry Durvasula: The CIO role has expanded into “CIO-plus.” It is one of the most transformative positions in the modern enterprise.
Gary Hoberman: Must they understand the business?
Sastry Durvasula: One hundred percent—if not one thousand percent.
If you are a CIO, CTO, or Head of Digital and you understand how to rewire the business, rethink workflows, and guide your C-suite peers, you have an opportunity to drive monumental enterprise impact. If you remain confined to traditional technical solutioning and infrastructure management, the role becomes limited.
We need to consider AI for Operations—modernizing infrastructure and IT operations—as well as Operations for AI. The same applies across AI for cyber and cyber for AI, AI for finance and finance for AI.
For pure software developers, the craft has fundamentally changed. Developers used to spend nine months writing, unit testing, and integrating code. Today, a prompt generates working code in thirty seconds.
Whenever I mentor developers entering the workforce, I ask: “When was the last time you read your company’s investor relations deck or 10-K filing?” Technologists must understand corporate strategy, unit economics, and business context to stay indispensable.
Gary Hoberman: Technologists must learn the business, the industry, and the market.
Sastry Durvasula: And business leaders must learn the technology. AI is directly reshaping product design, distribution channels, operational servicing, and corporate strategy. That cross-functional fluency is mandatory on both sides.
Gary Hoberman: Turning to security, board oversight, and regulatory compliance: how do you manage cybersecurity risks and ethical testing in an AI-driven enterprise?
Sastry Durvasula: Cybersecurity is consistently the most critical topic on our agenda. Our Chief Information Security Officer, Upendra Mardikar, reports directly to me.
We recently expanded his remit to include enterprise fraud operations. In a major financial services institution, fraud prevention and cyber defense are converging rapidly under AI, making that alignment essential. Our retired participants are often targeted by sophisticated phishing campaigns, deepfakes, and social engineering.
Looking at cybersecurity and AI more broadly, several interconnected forces demand attention:
- Physical Infrastructure and Sustainability: The rapid expansion of data centers raises significant economic and environmental questions regarding energy consumption and local water resources. I recently hosted Erin Brockovich on our internal podcast to address community and environmental impacts surrounding large-scale data infrastructure.
- The LLM Capability Race: Managing the pace of model innovation while ensuring enterprise-grade guardrails and compliance.
- Geopolitical Competition: Maintaining technological leadership while navigating foreign AI advancements and differing regulatory standards.
- Resilience and Third-Party Risk: Managing systemic risk across core banking, healthcare, and critical infrastructure. When major third-party IT systems fail, airlines are grounded and operations halt. In an agentic AI ecosystem, an uncontained vulnerability could compromise core operational rails.
Gary Hoberman: Do these topics dominate board conversations?
Sastry Durvasula: Absolutely. It is a central focus for our board of directors, risk and compliance committees, and audit committees. We are fortunate to have board members who bring deep expertise across academia, business leadership, and enterprise technology.
Gary Hoberman: Let’s jump into our lightning round. Dave, you have the first question.
Dave Ferrucci: Sastry, on a personal note: my family is of Italian descent, and my father listened to traditional Neapolitan music. One of the modern singers who carried that tradition forward is Lina Sastri. Your name reminded me of her. What is your favorite genre of music?
Sastry Durvasula: You’ll appreciate this, Dave: my mother tongue is Telugu, a South Indian language historically referred to by linguists as the “Italian of the East” due to its melodic, vowel-ending phonetics.
I enjoy both Western and Eastern classical music, but I’m also a big fan of country music. When I first moved to the US, I listened to artists like Shania Twain, and having lived in Texas, Arizona, and now Charlotte, country music stuck with me.
Gary Hoberman: Dave, what music do you listen to on that audiophile sound system of yours?
Dave Ferrucci: It spans genres, but recently I’ve been revisiting Led Zeppelin with a deeper appreciation for their musicianship and songwriting.
Gary Hoberman: Sastry, what show did you last binge-watch?
Sastry Durvasula: I love crime thrillers, but my most recent binge was Young Sheldon. As a prequel to The Big Bang Theory, it’s brilliantly executed—especially the scene where a young Sheldon sketches out reusable rocket boosters, only for Elon Musk to find the notebook.
Gary Hoberman: Have you seen The Beast in Me?
Dave Ferrucci: The Beast in Me—great series.
Gary Hoberman: What is your favorite book?
Sastry Durvasula: Electrifying America by David Nye. It examines the social and economic transformation of the United States as electricity was adopted between the late 19th and early 20th centuries.
When electricity was first installed in the White House during Benjamin Harrison’s presidency, staff were so terrified of electric shocks that they employed dedicated switch operators just to turn on the lights. It offers a great historical lens for understanding the anxieties and structural shifts surrounding AI today.
Dave Ferrucci: That historical parallel is profound. It reminds us that technology’s value often comes down to handling the necessary decisions we simply overlook.
Sastry Durvasula: Just like automated bill pay—we delegate recurring operational tasks so we can focus our cognitive energy where it matters most.
Gary Hoberman: Sastry, thank you for joining us. This was a masterclass in enterprise technology, architecture, and operational leadership.
Sastry Durvasula: Thank you both. Great discussion.
Dave Ferrucci: Thank you, Sastry.
