This summer, our team had an intern, and one question that she asked has stuck with me for a while: "Is AI going to replace frontend developers?" I found myself thinking deeply about this question because it came from someone not yet fully immersed in the industry but trying to find their footing in the software development world. This question has been circulating extensively in the past year since AI-assisted coding took off, but hearing it from an industry novice gave it a different weight.
My AI Journey (so far...)
Looking back at my experience with AI tools, I can trace my first significant encounter to the GitHub Copilot Technical Preview back in 2021. I recall seeing it being discussed by tech influencers on Twitter and watching demos on YouTube. I was genuinely excited to try it out and was fortunate enough to get access shortly after its announcement.
Playing around with GitHub Copilot back then feels pretty primitive compared to the AI tools I use today with their agentic workflows and conversational interfaces. Back then, it was straightforward: add a comment to describe a function you wanted to generate, and it would create it for you. My personal lightbulb moment came with test generation. I've never been the biggest fan of writing tests (even today), but writing a comment asking Copilot to generate a test and having it work correctly felt like pure magic.
At that point, I wasn't particularly familiar with LLMs or OpenAI. It's worth remembering that this was July 2021, over a year before ChatGPT entered the mainstream and forever changed the trajectory of AI. I simply accepted it at face value, assuming it had been trained on StackOverflow answers and GitHub repositories.
The Rapid Evolution
Between the initial release and what we see now, there was GitHub Copilot Labs which in my mind introduced the primitive building blocks of what we now know as the agentic editing experience. It was an important step that gave Copilot the ability to reason and transform code versus just providing plain AI autocomplete, which is what I'd classify the original release of Copilot as doing. This was the beginning of AI tools having more agency in the development process.
At the same time the Technical Preview was ending, I joined Microsoft and got access to GitHub Copilot in it's entirety as it was transitioning to the subscription model. The progression from there to GitHub Copilot X which seemed heavily inspired by ChatGPT's overnight success allowed me to converse directly with the AI from within VS Code rather than just generating code from annotated comments.
In the last year, I've seen yet another leap with agentic modes in tools like Cursor, Cline, and GitHub Copilot, alongside the emergence of specialized models such as Anthropic's Claude 3.5 Sonnet and Open AI's GPT-4.1 that are designed to excel at AI-assisted coding tasks.
Looking back over these past four years, it's been quite a journey watching how rapidly things have changed in the industry. I feel fortunate to have witnessed this evolution from early on, and being at Microsoft during this period was advantageous as I had access to a variety of cutting-edge tools and LLM models to experiment with as soon as they became available.
Where AI Shines in Software Development
As a whole, I've found AI tools consistently valuable for mundane but necessary work like generating scripts to automate repetitive processes. Like I know how to read and interpret what a PowerShell or Bash script does, but if you asked me to write one from scratch, I'd struggle. That's where AI becomes an indispensable partner; I explain what I need through natural language, and it gets me most of the way there.
Or with scaffolding things for a new feature. I often get stuck trying to make things perfect from the start, so I end up putting them off and procrastinating. AI tools help overcome that initial hurdle. I use AI to do that initial legwork, which gets the ideas flowing, and then I typically rewrite parts and refine them to match my style.
Furthermore, the tight integration between GitHub Copilot and VS Code has fundamentally changed my workflow, breaking the cycle of constantly jumping to Stack Overflow or Google for syntax questions or compilation errors.
That said, it's far from perfect. I frequently encounter situations where I find myself wondering if it's even worth continuing my conversation with AI, especially after receiving several suggestions that simply don't work. Despite these frustrations, the time saved from not context-switching between coding and searching usually outweighs the occasional unhelpful responses.
This fine balance of utility versus reliability becomes more apparent in complex coding situations, where AI isn't (yet) capable of handling everything in a "one-shot prompt" manner. While I sometimes wonder if I need to fine-tune my prompts (or become a better "prompt engineer," as they say), I've found that breaking down complex problems into smaller, more digestible parts typically yields better results than trying to solve everything at once. As odd as it sounds, AI seems to mirror our human approach to problem-solving in that way; it's most effective when the scope is clear and focused.
Will AI Replace Frontend Development?
Circling back to the intern's question, I did respond that: "Yes, it is possible that AI could replace some frontend development jobs. I'm not discounting that possibility". However, with software development, especially enterprise development, it's not always just about getting things to work. Which is what AI is great for.
A perfect example is with frontend styling changes. When addressing a visual bug, I've noticed that AI tends to pile on CSS to "fix" the issue. Yes, it solves the immediate problem, but when I evaluate the solution, I often realize we could just apply an API property from our component library to fix it more elegantly. These maintainability considerations are crucial in the lifecycle of a software product.
It might be cheaper and faster to have AI build a working model, but supporting and maintaining it long-term is something I believe AI agents aren't equipped to handle yet. They will probably get there eventually with larger context windows, but that's one of the many nuances when discussing whether AI will replace human developers.
How I use AI (beyond coding)
Like a lot of us these days, using AI has become almost non-negotiable, an essential tool that feels like a "lifehack" to accomplish tasks more efficiently. Beyond the obvious uses everyone knows such as summarizing long articles, drafting emails, generating basic code snippets, or creating presentation outlines. I've enjoyed using AI in these days:
Creating Analogies and Translations
AI is really good at coming up with analogies that make complex concepts understandable. Many times, I'm trying to explain technical concepts to non-technical leaders or friends outside the tech sphere, and AI helps simplify these explanations with relatable analogies. It's like having a Reddit ELI5 (explain like i'm 5) thread already created with some great answers.
I've also found it helpful with concepts from self-help books. I'm somewhat of a fan of self-help literature in recent years, and AI has been excellent at giving me examples of how to apply theoretical concepts in practical situations. Recently, I was recommended "Loving What Is" by Byron Katie, and I used AI to help understand and apply her framework to real-life scenarios.
Knowledge Expansion
I'm a naturally curious person, and when I watch documentaries or read about something interesting, I tend to go into this rabbit hole of google searches and Reddit threads. AI simplifies this process by connecting different dots and providing a broader context.
During my recent trip to Osaka and Seoul, I visited Osaka Castle and learned that the person who built it was the same historical figure portrayed in FX's Shōgun series (which is excellent, by the way, 10/10 would recommend). Later, at Gyeongbokgung Palace, the tour guide mentioned how the current palace was mostly rebuilt after the Japanese invasion. After the tour, I went on a questioning spree with ChatGPT to dig deeper and connect these historical events. Turns out these two places are more connected than I initially thought.
Trip Planning
On my recent trip to Japan and Korea, I noticed everyday people using tools like ChatGPT on their phones while navigating public transit, not just once or twice, but consistently throughout my travels. And I completely understand why.
ChatGPT and Gemini were instrumental in my trip planning. Depending on the day and what attractions or restaurants we wanted to visit, ChatGPT or Gemini effectively created and revised itineraries on the fly. They accounted for walking time, opening hours, ticket prices, and even suggested adjacent attractions I might enjoy.
While I could have done this planning myself on a computer before the trip, being able to make last minute changes by querying a chatbot on my phone based on weather conditions or if we wanted a more relaxed day was a complete game-changer. Of course, throughout this process, I did wonder if these platforms might eventually become vehicles for targeted advertising based on our travel preferences and queries.
If you can dream it, you can build it
This new era of AI tools in just about every product really embodies that mindset of "If you can dream it, you can build it". Reflecting on this "slap AI everywhere" phase, I see them not as replacements for human creativity and judgment but as powerful tools to achieve more. They remove friction from the creative process, handle routine tasks that would otherwise drain my energy, and serve as a virtual debater when I'm exploring new ideas.
When it comes to going all in with AI, I'm well aware of the slippery slope we're on. I've personally witnessed colleagues submitting sloppy AI-generated code simply because "it works", and I've read about executives at AI-forward companies confidently declaring that AI will replace developers entirely, with no downsides or hidden costs. These oversimplifications concern me, but that's a topic for another day.
So, the question isn't whether AI will replace us, but how we can best work alongside it to enhance what we do.