What Should Developers Do in the Age of AI Writing Code?

·Dante Chun

If there's one topic that's been dominating developer communities lately, it's AI. Two years since ChatGPT launched, tools like GitHub Copilot, Cursor, and Claude Code have now deeply embedded themselves in actual development workflows. Running an outsourcing development company myself, I feel this change firsthand.

Honestly, I was anxious at first. "Will developers become unnecessary?" I wondered. But after applying AI tools to real work for over a year, my perspective has changed significantly. Today, I want to share what I've learned through that process.

What AI Does Well, and What It Doesn't

Let's face reality. AI definitely excels in certain areas.

Repetitive CRUD code, boilerplate generation, regex writing, error message interpretation—the productivity gains from AI in these tasks are remarkable. An API endpoint that used to take 30 minutes can now be knocked out in 5.

But there are clear limitations.

Something happened last month. While designing a database schema for a new project, I asked AI for help. The result looked plausible, but it completely missed the business's unique context. Integration with the client's legacy systems, future expansion plans, the team's tech stack—contexts that don't appear in the code.

Ultimately, AI is great at telling you "how" to implement something, but "what" to build and "why"—that's still human territory.

From Coder to Orchestrator

If developers used to be 'instrument players,' we're now closer to 'orchestra conductors.' With AI as a powerful player joining the team, what we need is the ability to lead that player well.

  • Problem Definition: When a client says "build me a bulletin board," understanding the real needs hidden behind that request
  • Architecture Design: AI builds parts well, but humans must draw the big picture
  • Communication: Software is ultimately for people

Lessons from the Field

Running Dante Company, I've come to realize something clearly. What determines a project's success or failure isn't code quality—it's trust and communication.

There was a time when we delivered technically near-perfect results, yet the client was unsatisfied. They said there wasn't enough mid-project sharing. Thanks to AI reducing coding time, I can now spend more time communicating with clients and deeply understanding their requirements.

An Era of New Possibilities

Thanks to AI tools, things that were impossible before are now possible. I can handle fairly large projects alone or with a small team. Perceived productivity has at least doubled.

Adaptation Over Anxiety

Change has already begun and cannot be reversed. The key is accepting AI not as a competitor, but as a collaborator.

It's time to expand our identity as developers from "people who write code" to "people who solve problems." Instead of being anxious, let's adapt. That's the biggest lesson I've learned over the past year.