From Copilot to Pilot

I want to take a moment to reflect on where we are, and where we’re headed, in the world of software engineering as AI continues to accelerate. In particular, I’ve been thinking about how it’s already reshaping the developer experience today, and what that might look like 10 years from now.

We’ve all seen how fast this field is moving. Not long ago, people were saying large language models (LLMs) had plateaued, that most of the gains were in prompt engineering or minor improvements in inference. Some even thought we’d trained them on everything they could be trained on. But here we are, not even a year later, and these models are still evolving rapidly, becoming far more capable, and more “agentic”, able to take multi-step actions, reason over complex codebases, and act more like autonomous collaborators than autocomplete tools.

As a self-taught software developer, someone who grew up tinkering with Angelfire and GeoCities, reverse-engineering web pages by inspecting the source code, I’ve always loved how accessible and creative programming can be. And for a long time, the pace of change was manageable. You learned a stack, picked a framework (React, Angular, Ember, remember when those were new?), and committed to a language or two. The tech evolved, sure, but it never felt like it was racing ahead of you.

That’s changed.

How AI Is Already Shaping the Workflow

Over the last year or two, I’ve found LLMs like ChatGPT, Claude, and Gemini increasingly valuable, not just as novelty tools, but as real extensions of my workflow.

One of the first places I saw serious value was in writing SQL queries. Companies like Uber have demonstrated that AI-driven tools can generate high-quality, context-aware SQL queries from natural language, significantly speeding up development workflows.

Same goes for pull request descriptions. GitHub has highlighted how Copilot can automatically draft PR descriptions by analyzing code changes, saving developers considerable time.

Copilot in VS Code has been a game-changer too. It doesn’t just autocomplete, it completes like me. It sees how I write tests, how I name functions, how I structure my code. I can ask it to stub out tests or help with small UI changes, and it fits right in with my coding voice. That level of personalization wasn’t possible before.

The Death of Stack Overflow, and the Rise of Bespoke Code

Another thing that’s fading fast: copy-paste coding from Stack Overflow. Reports have shown that Stack Overflow’s traffic has declined notably as developers increasingly turn to AI-driven coding assistants instead.

Now? The answers I get from AI are tailored to my codebase. They’re bespoke. They’re contextual. It’s like pair programming with someone who’s read every line of my repo and every related Stack Overflow thread, and distilled it down to what I actually need.

That’s a major shift.

What’s Coming Next: From Copilot to Pilot

This brings me to my prediction: over the next 10 years, developers won’t just be copilots using AI as an assistant. We’ll be pilots, guiding the AI agents, curating and shaping their work, ensuring quality and direction.

Think about a plane. Autopilot has been around for a long time. It can handle most of the flight. But we still expect a human pilot to be there, to take off, to land, to step in when things go sideways.

The same is going to be true for software engineers. Tools like Cursor and Devin are already starting to fill this role, automating complex coding tasks from debugging to implementation. But there will still be a human developer at the helm, steering the ship, catching edge cases, making judgment calls, and bringing creativity, empathy, and context to the process.

We won’t be replaced. But we will be augmented, and the skill set required to thrive is changing fast.

The Shrinking Gap Between Junior and Senior Devs

One of the more controversial things I believe is that the experience gap is shrinking. A junior dev who’s been using these tools daily for four years might outperform a senior dev who’s been reluctant to embrace them. Not because they’re more seasoned, but because they’ve learned how to pilot the AI more effectively.

Architectural knowledge and system design experience will still matter, but day-to-day implementation? That playing field is leveling out. Fast.

The Future of Hiring (Please, No More Live Coding)

Hiring needs to catch up too. We still see companies asking candidates to do whiteboard algorithms or awkward live coding challenges, often stripped of the tools they’d use every day. That’s like asking a carpenter to build a chair using only a hand saw and whatever wood they can find nearby. It’s not realistic.

In the near future, I think companies will expect candidates to know how to wield these tools. AI won’t be a cheat code, it’ll be the new standard. And devs who learn how to work with it fluently will thrive.

The Only Constant: Change

Ethan Mollick, in his book Co-Intelligence, advises, “Assume this is the worst AI you will ever use.” Today’s AI is the least advanced it’ll ever be — each new iteration will only get smarter. Embracing AI now positions developers for continuous adaptation and success.

The AI isn’t here to take your job. But someone who knows how to use it probably is.