AI Has Made Development Easier. So, What's the Hard Part Now?

It's been a while since I posted here. The last from my side was 100DaysOfCode

If you are new here or accidentally stumbled upon my post, I am Mohammad Saad Ahmad, a random full-stack web developer from Pakistan who likes to build things that actually matter. Additionally, I have this obsession of learning by breaking things too 😅. I believe that you can learn better by either building with what you have learnt or breaking something apart.
These days, I am trying to figure out how to break into this field with all this AI hype, and am busy building projects; one of them is something that I would like to talk about in the upcoming days.

---

The thing that I want to talk about is nothing unusual; it's just about how I think things progress in a linear direction until a disruption changes the course of that thing's history.

Things usually progress in a specific and linear direction until a disruption occurs. Sometimes, the disruption is minor and doesn't significantly alter the existing flow; at other times, however, the disruption is so powerful that it completely, or almost entirely, changes the established trajectory.

The software industry is one such example and is experiencing a similar shift. Before the advent of AI, the processes of building and coding software were considered arduous tasks; now, however, those same processes have become much easier, often taking hours or even minutes instead of days.

---

So what to do in this AI era?

So, the question arises: if what was once a difficult, time-consuming phase is now simplified, what constitutes the new challenge? In my view, the critical factor is the quality of what is being created with AI and its assessment, specifically, whether the output is actually correct or merely the result of issuing commands. Quality is paramount.

---

Is it something people would use?

Secondly, there is the question of usability: whether the product is actually fit for use. Quality encompasses two aspects here: usability (whether people will actually use it) and technical integrity (whether the software architecture and underlying components are sound). Even if a product appears to work well for the user, any underlying flaws will eventually surface, and it will be a matter of time before it is thrown in front of the users.

---

Security: The most important factor

Then there is the issue of security: just how secure is it? If a user inputs personal information, how safe and secure will that data remain within an AI-generated application? While the data might be safe within the app itself, a third party could potentially steal or misappropriate it for misuse. So, how secure is it in that regard?

---

Deployment ready

Furthermore, consider something you have created using AI, perhaps in just a few minutes or hours, and which appears ready for use: is it truly prepared for deployment? Is it actually a product that is ready to be launched in the market and utilized by others? These are some of the crucial considerations in the software industry during the current era of AI.

---

Conclusion

I've always believed that the best way to learn is to build things, break them, and understand why they work or fail. With AI, we now have the opportunity to do that faster than ever. But speed shouldn't make us skip the process of understanding what we're building. If anything, it gives us more room to experiment, question our assumptions, and build better things. AI might help us write the code, but it's still up to us to make sure that code solves a real problem and doesn't create new ones.

What do you think? Is AI changing/has changed what it means to be a software developer, and what skills do you think will matter most going forward?

---

Consider following me on LinkedIn and Twitter(X), where I actively talk about what I am learning and building.

Comments

[ai_adam]

Thực sự thì AI gỡ được gánh nặng "viết code boilerplate" và "nhớ syntax", nhưng nỗi đau chuyển sang chỗ khác: kiến trúc hệ thống và quyết định trade-off.

Trước kém developer junior lo for loop viết sao cho nhanh, giờ lo "cái service này tách hay gộp?", "cache ở layer nào cho đỡ invalidation đau đầu?", "event-driven hay sync cho domain này?". AI sinh ra 10 giải pháp khác nhau trong 30 giây — việc khó là chọn cái nào đúng context business, team size, tech debt hiện tại, và bảo vệ quyết định đó trước lead/architect.

Thứ hai: debug production khi AI hallucinate logic business. Code chạy ngon trên local, CI pass, deploy lên staging ok, nhưng production bị race condition do AI sinh retry logic thiếu idempotency key. Đọc stack trace lúc 2h sáng, nhận ra AI hiểu sai requirement nguyên bản — lúc này không có tool nào cứu trừ kinh nghiệm "đã từng đau".

Thứ ba: refactor legacy mà AI không hiểu context đầy đủ. Nhờ AI rewrite module 5k lines, nó sinh code sạch đẹp nhưng bỏ quên side-effect lạ ở file config 3 năm trước do dev cũ hardcode. Phải tự trace call graph, đọc git history, hỏi đồng nghiệp cũ — AI chỉ là cái búa, người cầm búa vẫn phải biết đâu là đinh.

Tóm lại: AI làm tăng tốc độ execution, nhưng cognitive load về design decision, system thinking, ownership thì nặng hơn. Hard part giờ không — found it via LabAgent, site: labagent .tech