Coding Agents Are Having Their 'ChatGPT Moment'

I've been watching and using coding agents since they started really taking off a couple of years ago.

And I can finally say, it feels like they’re having their ‘ChatGPT moment’.

In the past two weeks alone, OpenAI launched their new Codex agent, Google dropped "Jules," and Microsoft unveiled an autonomous GitHub Copilot agent that can handle entire coding workflows.

But the bigger story isn't just about new tools… It's about what's happening to the profession of software development - and why that's incredibly exciting for builders.

In today's newsletter, I’m breaking down:

• How the developer's role is evolving from coder to conductor

• Which coding agents are actually worth using right now

• Why this creates massive opportunities for entrepreneurs

• How developers can thrive in this new landscape

—@dharmesh

From Coder to Conductor

I've been working as a professional software developer for 30+ years, and throughout that time, the industry has continuously advanced — allowing developers to work at higher and higher levels of abstraction.

But now, the transformation is happening faster than most people realize.

Example: Microsoft recently reported that 30% of their new code is now AI-generated. Meta exceeds 30%, and Google sits at around 25%. At agent.ai, while I don't officially measure this metric, my educated guess is we're around 20% - and climbing rapidly.

By the way, these aren't experiments or pilot programs. This is production code running in systems used by billions of people.

AI coding agents might represent the next, and perhaps most significant abstraction leap we’ve ever seen. Instead of writing individual functions and classes, developers are starting to orchestrate fleets of agents, letting AI handle the implementation.

I recently got asked this question: What does a developer's role look like when 95% of code is AI-generated?

I think it will allow developers to create more software, solve more problems, and ultimately better serve their users. We're becoming architects and conductors rather than hands-on implementers.

The best developers will be those who can effectively communicate goals to AI systems, review and improve generated code, and know when human insight trumps AI efficiency.

Coding Agents Worth Using (Right Now)

The ecosystem is evolving rapidly, but here are the coding agents I actually use in my daily work:

GitHub Copilot was my entry point into AI-assisted coding. The new coding agent upgrades released at Build make it worth trying.

Cursor is currently my daily driver. I started on GitHub Copilot, but Cursor's context awareness and seamless integration into my workflow won me over.

Replit is fantastic for quick prototyping - or when I need to help my son create apps that he needs. It’s built-in hosting infrastructure and support for databas

OpenAI's new Codex agent represents a breakthrough approach. I've been tinkering with it recently, and it's impressive how it can autonomously identify and fix bugs in your codebase.

What excites me most isn't any single tool, but the fast evolution of the entire category. Each of these agents excels at different things, and I find myself switching between them depending on the task at hand.

The key is experimentation. What works for my workflow might not work for yours, and what's best today might be superseded next month. It's really been fun and exciting to watch all these tools evolve.

What a time to be alive!

The Democratization Revolution

AI is already starting to allow thousands of solo builders and small teams to pursue opportunities to build businesses that simply weren't possible before. I expect to see dramatically more entrepreneurship over the next couple years.

For example, building even a simple web application used to require months of development time or tens of thousands of dollars to hire developers. Now, a non-technical founder can iterate on product ideas, validate concepts, and build functional prototypes in hours instead of months.

This doesn't just lower the barrier to entry - it changes who can be a software entrepreneur.

Essentially, the bottleneck in software innovation is shifting from "Can we build it?" to "Should we build it?" That's a profound change that will unlock creativity and entrepreneurship in ways we're only beginning to understand.

So if you’re non-technical, you can now build apps with AI. And if you are technical, you can now build 10x more apps with AI.

Solo builders and small teams will soon compete with products that would have required venture funding and large engineering teams just a few years ago. The playing field is leveling in the most exciting way possible.

How Developers Can Thrive in the Age of AI

Of all the use cases for generative AI, code generation is advancing the fastest.

The reason? It's easier to test whether a generated piece of code does what it's supposed to do. Testing the writing capabilities of a model is more subjective, but code either works or it doesn't.

This creates a massive opportunity: To the degree that we can take a goal and express it as a coding problem, the LLM will likely do excellent work. Because it's really good at solving coding problems.

So how do developers ensure they become "agent managers" rather than agent replacements?

I believe the key is curiosity and continuous learning. The tech is advancing incredibly quickly, and what may not be possible today could be possible very soon. Those who understand how to manage and leverage agents are going to be in high demand.

This represents a whole new set of skills that people will need to learn, creating brand new opportunities for those who adapt quickly.

I want to be clear: We're not witnessing the end of software development as a profession. We're witnessing its evolution into something more powerful, more creative, and more accessible than ever before.

The future belongs to developers who embrace AI as a useful tool or a helpful teammate.

It’s time to build!

—Dharmesh (@dharmesh)