The Future ofAI - Insight From Industry Leaders

The Future of AI: Insights from Industry Leaders on Competitive Advantage, Architecture, and the Future of Work

A conversation with Andrew Ng, Swami Sivasubramanian, and Sridhar Ramaswamy on the transformative moment in artificial intelligence

---

We're living through a hinge point in history. Artificial intelligence has evolved from decades of research in machine learning and deep learning into the most transformative technology of our generation. In a recent fireside chat, three technology leaders shared their perspectives on where AI is heading and what it means for businesses and developers.

 The Shifting Nature of Competitive Moats in AI

One of the most pressing questions facing businesses today is: where does competitive advantage actually come from in the age of AI? The answer, it turns out, is more nuanced than many assume.

**Andrew Ng**, the pioneering AI educator and founder of DeepLearning.AI, offered a surprising perspective: "I find that developers switch tools all the time, so the moats for developer tools may not be that weak because if there's a better tool, we switch API calls on a dime."

The key insight? Competitive advantage depends more on the nature of your industry than the AI technology itself. Consumer brand recognition remains powerful—ChatGPT's billion-plus users create a formidable moat. Two-sided marketplaces powered by AI can be highly defensible. But in the developer tools space, the traditional software moat is weakening dramatically.

"Previously you had a huge team to build a piece of software over many years that was really hard to replicate," Ng explained. "With AI-assisted coding, that moat is weaker."

**Swami Sivasubramanian** from AWS added another dimension: "The best competitive moat if you're in a business is really your business model. What is it that you're really going to drive value for your customers and for your business?"

The message for builders is clear: focus less on the technology itself and more on the unique value you deliver to customers.

 From Subscription to Consumption: The Business Model Shift

The conversation revealed a fascinating shift in how AI products are priced. Traditional SaaS relied on predictable subscription models, but AI solutions are increasingly moving toward consumption-based pricing.

Why? The productivity gains are so significant that developers are willing to pay far more than traditional subscription prices. Ng shared a personal anecdote: "Many of my developers spend hundreds of dollars a month, sometimes more than a thousand dollars a month, and we love it because they want to use even more of it."

But there's a catch. Unlike traditional software with near-zero marginal costs, AI inference has real costs that scale with usage. This creates a unique dynamic where open-source innovation is driving down the marginal cost of accessing models, while cutting-edge frontier models continue to command premium pricing because they can do substantially more.

Sivasubramanian noted that the industry is evolving toward a hybrid model: "There are always going to be certain tools that are subscription-oriented, especially user-driven ones. But as we scale to virtual autonomous agents, you're going to see a lot more elastic, cloud-like pricing models."

 Building Winning AI Architectures: Optionality is King

For product leaders and developers, the question of whether to use public APIs or invest in specialized models is critical. The consensus from both leaders? Start with whatever works to achieve product-market fit, but architect for choice from the beginning.

"For a lot of new innovative projects, the biggest challenge is identifying and building a product that customers love, not cost," Ng emphasized. "Just do whatever works to build that good product. Then when costs start climbing, we often find engineering methods to bend the cost curve back down."

The key is building in model optionality from the start. Small architectural choices early on—like owning your data layer rather than storing it in siloed SaaS products—can provide enormous flexibility later.

Sivasubramanian added practical wisdom: "When people test their products in beta or in the lab, they think certain workload patterns will come in, but then it turns out you're consuming a lot more input tokens or output tokens, and your cost model goes out the window."

His advice? Build in model choice, optimize prompt caching, and learn modern optimization techniques like fine-tuning and distillation. "Just like we used to teach students about compilers and performance optimization, now we need to teach them how to do distillation and fine-tuning."

 The PDF Revolution and Unstructured Data

One unexpected insight came from Ng's focus on a seemingly mundane data format: PDF files.

"The single most valuable form of unstructured data sitting in all of our businesses is PDF files," he declared. "We have so many of them—massive financial tables, healthcare forms—tons of value locked inside."

New agentic document extraction techniques can now pull complex tables from SEC filings, merge cells, and preserve formatting in ways that were impossible before. This unlocks decades of dark data sitting in enterprise systems.

As Ng joked, "Do you know what my most powerful PDF search engine was for much of my life? Command F."

Those days are ending.

What This Means for Developers and Students

Perhaps the most important part of the conversation addressed the anxiety many young developers feel about their future relevance.

Sivasubramanian offered reassurance grounded in history: "Somewhere along the way, we as an industry confused programming and computer science. Core fundamentals are still very much in demand—how to build compilers, database query planners, and understand system architecture."

He drew parallels to previous technological shifts: "When C and C++ came along, the number of developers increased. When Java eliminated memory management concerns, the number of developers increased again. We're at a similar cusp now where everyone is going to be a builder."

His advice for students? "Focus on core CS fundamentals—all the math, physics, and statistics that somehow get overlooked. These fundamentals really help you understand the new era."

Ng was even more emphatic: "This is a wonderful time to build something you're passionate about. The set of things that are now possible to build with less time and lower cost is much greater than ever before."

But he also noted a skills gap: "I can't hire enough people that really know AI. Our universities haven't adapted curricula fast enough."

His message to everyone, not just computer science majors: "Learn to code. Don't code by hand the old way—get AI to help you. That will make people in all job functions much more productive and have more fun."

 The Bottom Line

Three key takeaways emerged from this conversation:

1. **Competitive advantage in AI comes from business fundamentals**, not just technology. Focus on creating customer value, owning your data layer, and building products people love.

2. **Architecture for optionality**. The AI landscape is evolving rapidly. Design systems that let you swap models, optimize costs, and adapt to new capabilities as they emerge.

3. **Everyone should learn to code**—with AI assistance. The bar to building has never been lower, and the opportunity to create has never been greater.

As Ramaswamy concluded: "The future is not something we are waiting for. It is something we are all now more than ever equipped to create."

The question isn't whether AI will transform your industry—it's whether you'll be among those shaping that transformation.