A sudden AI interest in my company has left me avoiding it as much as possible.
I’ve been using Copilot for 3 years now and have been very happy with it. When ChatGPT launched, I quickly created an account and it integrated smoothly into my workflow. I promoted both Copilot and ChatGPT for the engineering team at my workplace—at least the mobile team—but to no avail.
That’s when the AI push really began. Over the past 4 months, as AI trends have surged, all the employees were encouraged to try them.
It started with a trial subscription of Cursor, with a lot of noise about MCPs. The focus was on using Cursor for our development, trying out various prompts, reviewing the code, and if it looked good, moving it to production.
I won’t go into the company’s reason for pushing AI tools.
For context, I’m a mobile developer using Android Studio, emulator(s), Chrome/Arc, etc. Adding another IDE to this stack wasn’t something I was eager for, especially since Copilot was already doing the job.
Given the encouragement, we tried it for one sprint. The results were not good, at least for the mobile team. The pressure to get the Cursor to make the UI from Figma designs with code quality that was just about acceptable… that was a tough week.
First, we decided to test it on something fresh. We got the Figma Mac App, integrated it’s MCP server—it was way better than Figma web for some reason. After many trials, prompts expanded from 2–3 lines to a full page (about 100 lines), basically a small PRD + TRD (Product and Technical Requirements Document). This included the feature explanation, all the Classes or file names to be updated, API contract sample, localisation rules, event rules, code structure rules, and more.
The output we got was somewhat intern-level work. It worked fine but needed a thorough review and significant refactoring. It did well at making model classes, repository functions and, small UI components (for Flutter). It fell short on UI code structure and state management.
All this was done in 1.5 days. Actual estimates were 3 days. So, a big win for the company, as it claimed to increase productivity by 50%. However, reviewing, fixing bugs and, refactoring took longer—so the total time was the same.
Encouraged by this partial win, we tried Cursor for an enhancement task in a released feature. This did not work at all. Writing detailed prompts took time; testing the result and rewriting the prompts took even more time and left the engineer frustrated, taking extra coffee breaks.
Since then, everyone has a different tool and workflow. Some are using Claude on the web, copy—pasting the code snippets on IDE. Others have embraced switching the IDEs between Cursor, Android Studio and web browser. I have been using the Copilot’s IDE extension, ChatGPT + web browser for regular web searches.
Lately we’ve also been trying out Copilot PR review. It seems odd: AI reviewing the code written by AI. It did give some good suggestions though. Let’s see how this works out.
I like the code completion in small chunks of code—easy to review, refactor, and test. Copilot works for me as I don’t have to switch to another code editor. For all other mundane tasks, ChatGPT works fine. In fact, switching from IDE to web browser for ChatGPT works well—it helps me focus on exactly what I’m looking for. For web searches, I still check Stack Overflow from time to time and read articles/blogs for learning, research, and debugging—focusing on human-written content.
I do want to try a local LLM, LM Studio as well.
I’m curious to see how these tools evolve, but for now, I’ll stick with what fits my workflow and leave the hype to others.
