Beyond the Hype: 5 Takeaways on the Future of Building with AI

The energy at AI Dev 25 x NYC was electric. Organized by DeepLearning.AI, the company founded by AI pioneer Andrew Ng, the conference halls in NYC were packed with over 1,200 developers buzzing with ideas.

But beyond the excitement, the event revealed some surprising truths. Talks from industry leaders like Anthropic, Vercel, Databricks, and CrewAI cut through the noise to show us what’s really happening. They painted a new picture of what it means to build software, today.

Andrew Ng about to begin his Opening Keynote session

Here are the top five takeaways that are changing how we build, test, and deploy software.

1. We’ve gone from “coding bottleneck” to “ideas bottleneck”

Andrew Ng shared a pretty disruptive idea that might reframe how teams build software, fundamentally. At the heart of that idea was a new truth; that the main limit on innovation is no longer speed to code.

AI tools have sped up coding so much that the bottleneck has moved to Product Managers (PMs), defining the what and why informed through speedy, frequent intake of user feedback. 

The result? We are seeing team structures change from one PM for every seven engineers to nearly a 1:1 ratio. To put it plainly, that’s mind-blowing.

As Andrew put it, “Having good ideas to build is now the hard part.”.

And he shared a personal story to prove it. Recently, Andrew built a game for his kids in just 10 minutes, using AI, but it took him an hour just to track down his daughter’s friends to play it and give feedback. The “product work” took six times longer than the coding!

This is great news for engineers. By learning product skills, you can become a “team of one,” building and validating ideas faster than ever.

2. AI makes coding more essential, not obsolete

At AI Dev 2025 x NYC, we also tackled what is, arguably, concern #1 in the AI era: Is AI going to take (programming) jobs? Andrew Ng gave us his answer – a powerful, straight-forward “No.” 

AI isn’t killing coding; it is making it more accessible.

Think of this like way back when we moved from punch cards to C++. Or when Javascript simplified interactive web development. Every time coding got easier, more people joined the field, not fewer.

Andrew Ng joined by Malte Ubl (CTO, Vercel), Fabian Hedin (CTO, Lovable) and Laurence Moroney (Dir of AI, Arm) to discuss how is it like to develop software in the age of AI

While some call this “vibe coding,” Andrew prefers “AI Coding” because it respects the complexity involved. The truth is, knowing the “language of software” lets you instruct AI with precision. Just like an art historian can describe a painting better than a novice, a developer knows exactly what to ask the AI to build. Andrew was clear:

Andrew was clear: “Telling people they no longer need to learn to code is the worst advice you can give people.”

Now is the best time in history to learn to code.

3. The real speed boost is in prototyping

AI makes code generation faster, but the real magic happens in the early stages.

For building production-ready software, AI gives us a solid 50% speed boost. That’s huge. But for building prototypes, the boost might be 10x or more.

It’s an important distinction that gives nuance to the generic “how much productivity can we gain from AI in coding?” style questions.

It also changes the risk game entirely. Instead of betting everything on one big idea, teams can now build 20 different prototypes cheaply to find the two that truly work. As Vercel CTO Malte Ubl said, “Iteration velocity solves all problems.”

As we put it here at Wizeline, “cheaper failure / relentless innovation.”

5. “Agents” are the new software

The future is Agentic. Speakers from Anthropic and CrewAI described Agents as the next big shift in computing.

But not every problem requires an agent to solve it. Both Anthropic and CrewAI drew a line between AI-powered workflows, and true agents. So what’s the difference? A workflow follows a set path, while an Agent can direct its own process to reach the defined goal.

Tanveer Mittal and Utkarsh Lamba from Anthropic presenting the latest and greatest from Claude Agent SDK

Just like Marc Andreessen famously declared over a decade ago that software was eating the world, CrewAI’s CEO João Moura declared that agents will eat traditional software. We are moving from static, pre-defined instructions to dynamic, goal-oriented systems. 

As this new frontier develops, an “Agent Operations Platform (AOP)” stack is already beginning to form.

Conclusion: Welcome to the age of the AI builder (that’s you)

The takeaway from NYC is clear: the technology is moving fast, but the biggest change is in how we build, who can build, and what we choose to build. 

Building has never been easier. 

We have the tools to prototype at 10x speed, the agents to handle more and more complex and long tasks, and the opportunity to focus on great ideas rather than just syntax.

But this isn’t just about productivity and speed. It’s a complete inversion of the traditional risk model, making experimentation cheaper and ambitious bets more affordable. 

Let’s move fast and build things.