From Application Developer to Application Orchestrator

A few months ago, I decided to build an Apple Vision Pro application as a side project.

Two years ago, that idea probably would have remained on my someday list.

Not because I lacked the technical ability. I've spent most of my career building and leading web technology teams. But VisionOS was a new platform. Learning the framework, finding tutorials, reading documentation, searching Stack Overflow, and stitching together sample code would have required a significant time investment.

Like many professionals, I balance side projects with a full-time job and other responsibilities. Realistically, that first prototype might have taken me months to complete.

Instead, I opened Claude Code.

About 30 minutes later, I had a working prototype.

That experience fundamentally changed how I think about software development.

The First Wave of AI Coding

When AI coding first appeared, the workflow was fairly straightforward.

You asked ChatGPT to generate some code.

It generated a response.

You copied and pasted that code into your IDE.

Then you built, tested, debugged, and refined everything yourself.

The AI was helpful, but it was still functioning primarily as a code generator.

The developer remained responsible for almost every step of the software development lifecycle.

Today's AI coding agents feel fundamentally different.

Instead of generating isolated snippets, they can participate throughout the development process. They can help plan, implement, test, refactor, debug, and iterate.

The conversation is no longer about writing code.

It's about building software.

My Role Changed More Than I Expected

When I look at how I spend my time developing with AI today, only a small percentage is spent writing code myself.

Most of my effort goes into:

• Deciding what to build

• Defining requirements

• Providing context

• Reviewing plans

• Evaluating generated solutions

• Debugging edge cases

• Refining the user experience

The bottleneck is no longer typing.

The bottleneck is thinking.

That's when I realized something important:

The role isn't disappearing.

It's evolving.

The Rise of the Application Orchestrator

For years, the industry has referred to us as application developers.

But increasingly, I think a more accurate description is application orchestrator.

An orchestrator doesn't manually perform every task.

They coordinate people, systems, processes, and outcomes.

That's increasingly what AI-assisted development feels like.

I provide the vision.

The AI helps execute.

I review the result.

We iterate together.

The quality of the outcome depends less on how quickly I can write code and more on how effectively I can communicate intent.

In many ways, working with AI agents feels remarkably similar to working with human development teams.

Clear requirements produce better outcomes.

Context matters.

Architecture matters.

Review matters.

Communication matters.

AI Doesn't Eliminate Judgment

One misconception I see is the assumption that if AI can generate code, developers become unnecessary.

My experience has been the opposite.

The more capable the AI becomes, the more valuable judgment becomes.

In my role at NYU, AI doesn't understand the organizational context, stakeholder expectations, institutional constraints, or business priorities that influence every technical decision.

In my Vision Pro projects, AI can generate implementation options, but it doesn't possess taste.

It doesn't know why lyrics should appear beside album artwork.

It doesn't understand why an interaction feels intuitive or awkward.

It doesn't understand emotional resonance.

Those decisions still belong to humans.

As AI becomes better at implementation, human value shifts toward context, judgment, architecture, and user experience.

The Biggest Mistake People Make

I've noticed that most failures with AI coding agents aren't actually coding failures.

They're communication failures.

When I provide vague instructions, I get vague results.

When I fail to explain the technology stack, requirements, or constraints, the agent makes assumptions that may not align with what I need.

One of the most effective techniques I've adopted is using planning mode before implementation.

Instead of immediately asking an agent to build something, I first ask it to present a plan.

I review the approach.

I modify it.

I provide additional context.

Only then do I move into implementation.

This process consistently produces better outcomes.

Not because the AI became smarter.

Because the collaboration became better.

What Developers Should Focus On

I don't believe coding skills are becoming obsolete.

Far from it.

The ability to understand code, evaluate quality, identify problems, and recognize good architecture remains essential.

You cannot effectively review what you do not understand.

But I do think another skill is becoming equally important:

The ability to collaborate with AI agents.

The developers who thrive over the next decade won't simply be the best coders.

They'll be the people who can:

• Clearly define problems

• Communicate requirements

• Provide context

• Evaluate solutions

• Make architectural decisions

• Guide AI systems toward better outcomes

In short, they'll become exceptional orchestrators.

We Got the Story Wrong

For years, the conversation around AI and software development has been dominated by fear.

Will AI replace developers?

Will software engineering disappear?

Will coding become irrelevant?

I think those were the wrong questions.

The biggest thing we got wrong about AI and software development was assuming it would threaten us.

In my experience, it gave us superpowers.

Projects that once sat on a backlog for months can now be explored in an evening.

Ideas that previously required learning an entirely new platform can now be prototyped in minutes.

The opportunity isn't to compete with AI.

It's to learn how to collaborate with it.

The future belongs to developers who can effectively direct intelligent systems and combine them with uniquely human strengths: judgment, context, creativity, and taste.

The future isn't developer versus AI.

It's developer and AI.

And together, they can build things neither could create alone.