In their inaugural episode, hosts Gary Hoberman and Dave Ferrucci introduce themselves, their mission, and why architecture is so critical in the AI era
AI is everywhere right now, from headlines to boardroom conversations and product roadmaps. The promises are massive: transforming people, code, processes, applications, and entire business models. But behind the hype lies a more practical question: What’s actually working inside real enterprises, and what isn’t?
That’s the mission behind “Architecting the AI Enterprise,” a new series hosted by Unqork’s Gary Hoberman, CEO and founder, and Dave Ferrucci, CTO, Chief AI Officer, and the mind behind IBM Watson.
Gary and Dave have already lined up some amazing guests, including more than 10 CIOs and CTOs from Global Fortune 1000 companies. Together, they won’t just explore AI’s potential. They’ll dissect the reality of leveraging AI to build, deploy, and manage enterprise-grade applications at scale.
From Research Labs to Wall Street: Two Paths, One Convergence
In Episode 1, Gary and Dave begin by telling their own origin stories. They come from opposite ends of the technology spectrum.
- Gary climbed the enterprise ladder, leading massive engineering teams in financial services, managing billions in tech spend, and grappling with the messy reality of production systems.
- Dave built his career in AI research, spending decades at IBM’s Thomas J. Watson Research Center, culminating in Watson—the system that famously defeated human champions on Jeopardy!.
Their journeys eventually converged: both moved from corporate giants to startups, and now they have joined forces to rethink enterprise architecture in the age of AI.
The Watson Moment: When AI Won Jeopardy!
When IBM’s Watson beat top players on the television show Jeopardy!, Dave and his team didn’t just achieve a technical milestone. They helped transform how enterprises think about AI.
At the time, answering even simple questions with accuracy was a challenge, Dave explains. In answering Jeopardy! questions, systems hovered around 30–35% accuracy. Watson pushed that boundary to ~75% on complex, nuanced Jeopardy!-style questions—while also introducing something even more important: Watson started estimating the confidence that it was right.
Predicting accuracy was critical, so that Watson knew if it should buzz in to answer. Doing so with a wrong answer would hurt its chance of winning.
That idea—AI as a probabilistic system—was a critical step forward.
Deterministic vs. Probabilistic: The Core Tension
However, a probabilistic approach has also clashed with enterprise expectations for computing systems. Businesses are used to deterministic systems. If you input X, you get Y. Every time. And that is how many (but not all) business decisions have to be made.
So, AI introduced uncertainty, and with it, discomfort.
Deterministic Systems
- Banking transactions
- Accounting systems
- Core enterprise workflows
- Expectation: 100% accuracy, every time
Probabilistic Systems
- Market predictions
- Healthcare outcomes
- Natural language understanding
- Expectation: high degree of confidence, not certainty
What is the mistake many companies make today, according to Gary and Dave? Applying probabilistic AI to problems that should be deterministic.
As Dave puts it, it’s like “brushing your teeth through your ear.” It might be a solution, but it’s inefficient, risky, and unnecessary.
The Rise of AI Agents and the Risk That Comes With Them
Today’s enterprise AI solutions are largely driven by agents that can generate code extremely quickly using probabilistic large language models (LLMs). However, they carry real risks.
One of the biggest misconceptions in AI today: If code is cheap to create, software becomes cheap. In fact, the risk is, AI could actually make software both riskier and more expensive.
Drawing from his experience managing massive enterprise systems, Gary highlights the real issue:
- Every line of code = long-term liability
- Maintenance, security, compliance = ongoing cost
- Most enterprise budgets = 80% “keep the lights on”
From Golf Carts to Formula One
So what does this approach to AI mean for the enterprise? The bottleneck shifts from coding to architecture. Dave describes it this way:
- Before: Not enough developers
- Now: Too much code, too fast
- Next problem: Who ensures it all works together?
This is where the architect becomes the most critical role in the enterprise. To explain why, Gary and Dave use a powerful metaphor:
- Traditional development = golf carts (slow, controlled)
- AI-powered development = Formula One cars (fast, chaotic)
Unqork’s original vision—constraining development through structured components—was built for control. Now, in the AI era, that original vision can play a powerful role in harnessing AI’s speed, while retaining control and ensuring security and governance.
The Most Important Role in the AI Enterprise
When asked what role matters most going forward, Dave doesn’t hesitate: The architect.
Why? Because architecture is about:
- Making trade-offs
- Managing dependencies
- Anticipating change
- Balancing speed with control
And in an AI-driven world, those decisions become exponentially more complex, and evermore critical.
What’s Next for “Architecting the AI Enterprise”?
This episode sets the tone for what’s ahead. Stay tuned for future conversations will bring in top technology leaders to discuss their career journeys and answer:
- How should enterprises architect for the future?
- What AI is actually working in production?
- What’s failing—and why?