Have a look at my article published by MoT, “Lessons in quality engineering from working with Cursor and Windsurf.” The article shares a real-life story of how AI-assisted tools like Cursor and Windsurf help and hinder development and what we can learn from it as Quality engineers.
What You’ll Learn:
Real-life experiences of how AI-assisted development tools help and hinder engineering
What AI-assisted tools are capable of
The impact AI-assisted tools will have on quality.
After reading, share your thoughts:
Are your team members using AI-assisted tools? Or are there still concerns?
What are your experiences with working AI-assisted tools like Copilot, Cursor and Windsurf?
What risks do you see in teams adopting AI-assisted tools?
Once again I am coming across the word “overreliance” just like what Automation is vs. Manual testing
We can’t rely too much on AI agents or chat because we can easily lose understanding of what we’re building.
this! I’ve only ever experienced Copilot with GPT and I must say: just like everything AI is at present-day, ultimately it’s still the user or creator that will finish the job. Some requests are hard to put into words (prompts) and a solution/addition only manifests once it is already out there.
There are lots of opportunities to move faster with AI, but it requires informed and intelligent ways of working with them
Time is indeed still needed to use AI effectively!
Thank you for sharing your story interacting with AI and experiencing the current quirks that sound really familiar to me.
I’m on a similar exploration where I use Junie in JetBrains InteliJ IDE as an assistive AI agent to generate code with tests for me. What I observed was that the AI agent wants tests to pass, not by generating quality tests but by changing the configuration to be less strict. Did you experience this as well?
We only use Copilot and Chatgpt. I feel the chatgpt model in Copilot could be a little more intelligent. Many times I found the standalone LLM to be better at reasoning.
Thanks for your article. I agree with you: We can’t rely completely on AI to do all the stuff.
Weeks ago I tried to give a full try to vibe coding for a scenario I automated. I regretted what was happening at the beginning and decided to do the process the artisanal way, and for some things I used AI.
At the moment the risk in our organisation is that the use of AI is starting up in pockets with varying degrees of enthusiasm. They range from a dev that is vibe coding as much as they can, to others not wanting to use AI in case the deskill themselves.
I did a confluence guidelines document on Gen AI and presented it to my quality engineers to encourage them to use AI and share their experience - logging the prompts that gave them real benefit. At the moment the document only contains my prompts.
We use primarily Co-Pilot, but we are now investigating agents such as augmentcode.
this! I’ve only ever experienced Copilot with GPT and I must say: just like everything AI is at present-day, ultimately it’s still the user or creator that will finish the job. Some requests are hard to put into words (prompts) and a solution/addition only manifests once it is already out there.
