Architecting the AI Enterprise

Coding agents aren’t the end of software engineering. They’re the next layer of abstraction. In Part 1 of our conversation on Architecting the AI Enterprise, we sit down with Grady Booch, Co-Creator of the Unified Modeling Language (UML) and former Chief Scientist for Software Engineering at IBM. We examine why today’s agentic harnesses are rebuilding the Watson pipeline, why natural language can’t replace formal systems with grounded semantics, and why vibe coding has no place in high-consequence architectures.

We revisit Watson’s generative, evidence-scoring pipeline, which Grady documented in UML after the Jeopardy! win, and compare its source-aware provenance with the LLM, RAG, and evaluation stacks enterprises are assembling today. Grady maps three golden ages of software engineering. We debate whether formal specification survives once classic programming fades, compare P(doom) estimates around the table, and land on where the real AI risk sits.

What You’ll Learn:

• How Watson’s generative, evidence-scoring pipeline anticipated today’s LLM, RAG, and evaluation stacks
• Why many agentic harnesses are reinventing pipeline architectures and established design patterns
• The three golden ages of software engineering, and why coding agents mark the start of the third
• Why the history of software engineering is a history of rising levels of abstraction
• Why formal systems with grounded, deterministic semantics won’t be displaced by natural language
• Where vibe coding fits in low-risk, disposable systems, and why it doesn’t belong in high-consequence ones
• Why the real AI risk lies in human misuse and concentrated power, not autonomous superintelligence
  
  

Grady Booch is a Scientist, Storyteller and Philosopher, Co-Creator of the Unified Modeling Language (UML), former Chief Scientist for Software Engineering at IBM, and one of the most influential figures in software architecture. He was a pioneer of object-oriented design and part of the design patterns movement, and he spent decades studying how complex, software-intensive systems are built and what makes them last. At Rational Software, Grady’s work shaped tools and methods used by a generation of engineers. Grady is also a historian of computing and is developing a documentary on computing in the spirit of Carl Sagan’s Cosmos.

Highlights and YouTube Chapters:

• 00:00 Welcome and Meet Grady Booch
• 01:35 Watson as a Generative, Evidence-Scoring Pipeline
• 04:43 Scientist, Storyteller, or Philosopher?
• 05:53 AI as a Philosophical Pursuit
• 07:47 Computing Is Younger Than Its Pioneers
• 08:36 What’s Your P(doom)?
• 11:00 Code Generation, Insecure Runtimes, and Unconstrained Agents
• 11:53 Enjoying Software While Deconstructing Its Architecture
• 14:44 Vibe Coding for an Audience of One
• 15:56 Are Coding Agents Replacing Software Engineers?
• 18:28 Formal Systems vs. Natural Language
• 20:37 Grace Hopper, Compilers, and High-Consequence Systems
  
  

Architecting the AI Enterprise is handcrafted by our friends over at: fame.so

Seat-based SaaS was built for humans working in a system of action. Agents don’t need one. In our latest episode of Architecting the AI Enterprise, we sit down with Sastry Durvasula, Chief Operating Officer at TIAA, to break down why per-seat licensing can’t survive agentic orchestration. We also cover how model-agnostic orchestration and tokenomics should shape LLM strategy, and how his 80/20 hypothesis reframes workforce and workflow transformation for incumbents running on decades of legacy architecture.

We dig into Sastry’s model of the enterprise stack as systems of record, intelligence, and action, and why the action layer is the first to be disintermediated as agent-to-agent consumption scales. We debate whether model agnosticism means LLMs become a commodity, and when to burn down enterprise commitments versus pushing model selection to vendors.

What You’ll Learn:

• How the 80/20 hypothesis applies to workforce and workflow transformation in a 100-year-old enterprise
• How the system of record, system of intelligence, and system of action model explain the future of SaaS
• Why software vendors will have to disintermediate and cannibalize themselves to survive
• How to think about model-agnostic orchestration, tokenomics, and matching models to workloads
• Why technologists need to learn the business, and business leaders need to learn the tech

Sastry Durvasula is the Chief Operating Officer at TIAA, the Fortune 100 retirement, asset management, and wealth management firm founded by Andrew Carnegie more than a century ago to serve teachers. His experience spans enterprise operations, business transformation, global technology, digital and client experience, client and fraud operations, and global capability centers. A software engineer and architect by training, Sastry led digital transformation at American Express, built the first digital strategy and organization at Marsh, and ran technology and digital at McKinsey while advising clients across financial services, insurance, private equity, and fintech.

Highlights and YouTube Chapters:

• 00:00 Welcome and Meet Sastry Durvasula
• 01:18 From Software Engineer to Digital Transformation Leader
• 05:56 Inside the COO Role
• 07:30 How AI Will Reshape the Workforce
• 08:12 The 80/20 Hypothesis
• 10:03 Is AI Different Because It Competes with Human Cognition?
• 12:06 Why AI Is Unprecedented, but Shouldn’t Be Overhyped
• 14:31 Is SaaS Dead?
• 19:29 Why Software Companies Must Cannibalize Themselves
• 21:30 Burn Down Your LLM Commit or Let the Vendor Decide?
• 22:52 Model-Agnostic Orchestration and Tokenomics
• 24:01 Matching Models to Workloads and Job Families
• 27:00 Probabilistic vs. Deterministic: Who Do You Trust?
• 29:50 Remember the Human Baseline
• 31:09 The Future of the CIO and the Technology Function
• 35:03 Why Business Leaders Need to Learn Tech
• 36:19 Security, Fraud, and Protecting Retirees
• 38:41 The Interconnected Forces Behind AI Risk
• 43:11 Lightning Round
• 49:21 Let AI Make the Decisions You Forget to Make

Architecting the AI Enterprise is handcrafted by our friends over at: fame.so

What does it actually take to put AI to work at enterprise scale? Not in a demo, but in the systems that run your business?

On Architecting the AI Enterprise, hosts Gary Hoberman and Dave Ferrucci sit down with the technologists, executives, and builders shaping how large organizations design, deploy, and govern AI.

Each episode cuts through the hype to deliver fresh thinking, hard-won lessons, and actionable strategies for leaders who need results. Whether you’re a CTO rethinking your architecture, a product leader evaluating AI-native platforms, or an executive charting your company’s AI roadmap, this is the conversation that meets you where the real decisions get made.

Architecting the AI Enterprise is handcrafted by our friends over at: fame.so

In this episode of Architecting AI Enterprise, Unqork CEO Gary Hoberman and Chief AI Officer Dave Ferrucci sit down with Robert Pick, CIO at Tokio Marine North America, to discuss the realities of scaling AI in highly regulated industries. Together, they explore the operational limits of "vibe coding," the evolving role of the CIO as an enterprise orchestrator, and why human context remains indispensable in an AI-driven world.

Architecting the AI Enterprise is handcrafted by our friends over at: fame.so

In this episode of Architecting the AI Enterprise, Gary and Dave sit down with Axis Capital Global CIO Stephen Lord to explore the wide-ranging possibilities of artificial intelligence within the complex specialty insurance market. They discuss Lord’s leadership in the shift from basic automation to strategic employee augmentation and why high-risk enterprise decisions still demand robust software architecture and human oversight.They also discuss how the global IT function is changing and compare it with previous technology paradigm shifts.

Architecting the AI Enterprise is handcrafted by our friends over at: fame.so

In this episode of Architecting AI Enterprise, Gary Hoberman and Dave Ferrucci sit down with Kfir Godrich, who formerly served as Chief Innovation Officer and Global Head of Data and AI CIO at BlackRock. They examine how AI agents and generative AI are disrupting traditional SaaS and what that means for highly regulated industries like financial services and healthcare. The conversation explores the limits of AI autonomy, the importance of governance and explainability, and why the future of enterprise software will likely be a hybrid of deterministic systems and AI-driven workflows.

Architecting the AI Enterprise is handcrafted by our friends over at: fame.so

Explore how judgment, accountability, and ethics are reshaping AI in enterprise settings. Join CEO Gary Hoberman, CTO Dave Ferrucci, and former global insurance CIO Jenny Larsson as they discuss the integration of AI with human judgment, ethical design frameworks, and the future of AI-driven organizations. This episode offers insights into balancing rapid AI adoption with trust and accountability, making it essential listening for leaders and innovators eager to harness AI responsibly.

Architecting the AI Enterprise is handcrafted by our friends over at: fame.so

In this episode, Jeff McMillan, former Head of Firmwide AI at Morgan Stanley, shares his journey from data quality engineer to leading AI at Morgan Stanley, where he led the rollout of secure generative AI platforms to 80,000 employees. Together with hosts Gary Hoberman and Dave Ferrucci, Jeff explores how enterprises can realistically deploy AI at scale—and why success depends less on models and more on architecture, governance, and organizational clarity. 

Architecting the AI Enterprise is handcrafted by our friends over at: fame.so

Alla Whitston (CIO of MUFG Americas) shares her journey from technologist to business-focused technology leader, emphasizing that true impact comes from deeply understanding business problems. Then, she, Gary and Dave explore how AI is reshaping enterprise architecture and shifting the role of CIOs toward governance of both human and artificial intelligence. Finally, they discuss the risk of uncontrolled AI-generated code, stressing the need for guardrails, accountability, and strategic alignment across business and technology.

Architecting the AI Enterprise is handcrafted by our friends over at: fame.so

The future of AI architecture is here—and it’s more transformative than you think.

Imagine AI tools that don’t just generate code but govern entire enterprise operations seamlessly. Sound impossible? Experts Gary Hoberman and Dave Ferrucci unravel how AI is rewriting the rules of speed, security, and governance in today’s complex business landscapes.

Architecting the AI Enterprise is handcrafted by our friends over at: fame.so