Learn to Code vs Build to Learn: Why the Old EdTech Model is Dead

The old promise of coding bootcamps and tutorials was straightforward: learn to code, get a job. But on the latest episode of the Levels Podcast, Punit Jajodia explained why that model is collapsing—and what Programiz is doing about it.

The shift isn't just about AI writing code. It's about a fundamental rethinking of what coding education needs to accomplish in 2025 and beyond. For edtech founders and anyone building learning products, Punit's evolution from "learn to code" to "build to learn" offers a glimpse into the future of skill development.

The Old Model: Linear Learning

For decades, coding education followed a predictable path. You started with "Hello World." You learned variables, then loops, then functions, then data structures. You built up knowledge systematically, piece by piece, until you felt ready to build something real.

Programiz operated this way for years. Their tutorials took learners through fundamentals methodically. Master C, then C++, then Python. Understand the syntax before attempting projects. Learn the rules before breaking them.

This approach made sense in a world where writing code was the bottleneck. If you couldn't write a for-loop, you couldn't build anything. Education was about removing that constraint.

But something changed.

The New Reality: AI as Coding Partner

With the arrival of ChatGPT, Cursor, and tools like Lovable and Bolt, the bottleneck shifted. Now anyone can describe what they want to build and get working code in seconds. The constraint isn't writing code anymore—it's knowing what to build, how to architect it, and how to fix it when it breaks.

This created what Punit calls the existential crisis in coding education.

"Learning to code is no longer enough evidence to get a job or an internship because you're really competing with AI now."

If AI can write code faster and more accurately than a junior developer, why spend months learning syntax? This question was killing motivation across Programiz's entire audience. The old value proposition—learn these skills to become employable—no longer held.

The First Pivot: Learn to Build

Programiz's initial response was to shift from "learn to code" to "learn to build."

"So we kind of pivoted from learn to code to learn to build, right?"

The distinction matters. "Learn to code" focuses on the craft—the syntax, the patterns, the conventions. "Learn to build" focuses on the outcome—creating functional software that solves problems.

This reframing acknowledged that knowing Python isn't the goal. Building things with Python is the goal. And if AI can help you build faster, that's not a threat—it's a tool.

But even this wasn't quite right.

The Second Pivot: Build to Learn

The real insight came when Programiz realized they were still preserving the old sequence: learn first, build later. That's backwards in an AI-assisted world.

"And now we are converting to build to learn, which means you can't wait to learn. You have to build right from the beginning in the process of learning."

Think about what this means in practice. Instead of spending weeks on Python basics before attempting a project, you start building immediately. You have an idea for an app? Start building it today. Don't know how to implement authentication? Figure it out when you need it. Stuck on database design? Learn it in context.

AI makes this possible. You can generate working code even when you don't fully understand it, then learn by modifying, breaking, and fixing it. The building becomes the education, not the reward at the end of education.

Why This Actually Works Better

There's educational theory backing this up. Constructionist learning—learning by making things—has been proven effective for decades. But it was hard to implement in coding because beginners got stuck on syntax errors and couldn't progress.

AI removes that friction. You can focus on the concepts—what do I want this to do?—while AI handles the implementation details. Then you learn by examining the code, tweaking it, and understanding why it works.

Punit uses a perfect analogy to explain why this matters:

"We had a coffee maker in the office and one of the springs broke, right? And we had to buy a new one. It was just a spring. But I did not know how to fix that. There's a hole in the coffee machine."

The coffee maker worked for 10 years. But when it broke, it became useless because nobody understood how it worked internally.

"That's what AI generated code is going to be like. It will work... But the day it broke, everything broke. So do you want your code to break at a point where you are incapable of fixing it? Maybe not."

If you only know how to prompt AI but don't understand the code it generates, you're helpless when something breaks. But if you learn by building—examining AI's output, modifying it, understanding its structure—you develop genuine competence.

The Product Implications

This philosophy shift drove concrete product changes at Programiz. They introduced two types of projects:

Guided projects, where the goal is defined and learners build toward a specific outcome with support along the way.

Unguided or "freestyle" projects, where learners pursue their own ideas and get help when stuck.

Both types emphasize building from day one, with learning happening in context rather than as a prerequisite.

"The idea is to do projects, do your own projects and submit them to the wall of inspiration and while you're building the project, if you ever feel lonely or if you want help, you go to the Discord community."

They also built a "wall of inspiration" where learners showcase what they've created, and a Discord community for real-time help. The focus shifted from consuming tutorials to creating alongside others.

Monthly Challenges: Streaks for Builders

To operationalize "build to learn," Programiz introduced monthly challenges—specific projects to complete within 30 days. This creates what Punit calls a "streak" of building.

"We moved to monthly challenges, which was like, so we then pivoted from... this idea of streaks, this idea of leaderboard can be, they're not just a product strategy. They are also content strategy."

The genius here is that building becomes both the learning mechanism and the engagement hook. Each month presents a new challenge, a new reason to show up, a new opportunity to level up skills by shipping something real.

The Broader Lesson for EdTech

Programiz's evolution reveals a pattern that extends beyond coding. In any field where AI can automate core skills, education needs to flip from "learn then do" to "do then learn."

Want to learn graphic design? Don't spend months studying color theory—start designing with AI tools and learn principles as you go.

Want to learn marketing? Don't read textbooks about customer psychology—launch a campaign and figure out what works.

Want to learn writing? Don't study grammar rules—start writing with AI assistance and develop your voice through iteration.

The old model assumed knowledge had to precede application. The new model assumes application is the fastest path to knowledge, especially when AI can scaffold the process.

The Skills That Still Matter

Does this mean fundamentals don't matter? Punit doesn't think so. In fact, he argues the opposite:

"So coding is not just about writing Python. It's a way of thinking. It's a way of problem solving. So problem solving will always remain."

The difference is when and how you learn fundamentals. Instead of mastering them abstractly before building, you learn them contextually while building. The problem-solving mindset, the architectural thinking, the debugging skills—these emerge through practice, not lecture.

"And if I was a 20 something going through this question of should I learn to code? The reframing I would do for them is do you want to bet your career on something every person in the world can do? Or do you want to learn a skill that survive for ages?"

AI democratizes code generation. But it doesn't democratize taste, judgment, or deep understanding. Those still require time and practice. Building is how you develop them.

What This Means for Learning Products

If you're building any kind of educational product, Programiz's journey suggests some questions worth asking:

Are you still operating on the "learn then do" model when "do then learn" might work better? What would it look like to flip the sequence?

Can you reduce time-to-first-creation? How quickly can someone make something real, even if they don't understand it fully?

Are you treating AI as a threat or as scaffolding? Can AI help beginners jump straight into building while learning along the way?

Does your product enable learning in context, or force abstract learning first? Can people get help exactly when stuck, not beforehand?

How do you showcase creation as the goal rather than knowledge as the goal? Is building visible and celebrated in your product?

The broader principle is about meeting learners where they are—which is increasingly "I want to build something" rather than "I want to understand everything first."

Key Points

• The old model—learn systematically, then build—is collapsing because AI can now write code faster than juniors can learn
• Programiz pivoted from "learn to code" to "learn to build" to "build to learn"
• Build to learn means starting projects immediately and learning in context, using AI as scaffolding
• The coffee maker analogy: AI code works until it breaks, then you need understanding to fix it
• Programiz now focuses on guided and freestyle projects, with Discord support and a wall of inspiration
• Monthly challenges create building streaks that double as learning mechanisms
• Problem-solving and architectural thinking still matter—they just develop through building, not beforehand
• The pattern extends beyond coding to any field where AI can automate core skills

Listen to the full conversation to hear Punit's complete vision for the personal software era and why he believes coding skills will outlast the AI hype cycle.