Well firstly, are you comfortable with AI writing your test cases and if so, how would you spend the saved time? What tasks would you do instead?
We are starting to implement an AI first approach to Planning so really intrigued to see what comes from this convo. AI’s driven massive growth in output in efficiency from our Dev team which means we need to do something to keep pace. We’ve only just started using it, but anticipate that we may be able to begin to address gaps in our ACs and hopefully much more exploratory testing.
This is an interesting question, to be fair I’ve not written test cases in decades due to waste and low value related thing so this is not something I’d look to AI for. Similarly with automation its not so much a saving but allows me to add automation I would not have covered before.
With AI bringing others challenges to light though I am recognising that a lot more people than I was aware of have been writing and executing test cases or spending a lot of time on maintenance activities. There are some indicators they will just do more of this with faster code churn coming through but I am interested to see what the views of those in this situation are.
Taking on more QE activities from developers does seem one path that is being flagged at the moment but that is a move away from testing,
Based on my past experience of shifting away from test case approaches to other testing models I’d say easily 5x in value and efficiency so an opportunity there but that could me less headcount along the way unless workflow increases to balance that out.
So much tools to experiment with at the moment so that could be a quick option and their would be knowledge not being gained by not writing them anymore, best though to get views from those going through this now, whilst I have experience in that sort of transition from decades ago I’m speculating quite a bit here.
I think I’d spend a lot of time reviewing the test cases AI wrote. I’d also be spending time prompting AI to produce better test cases. The human in the loop will be important for some time to come.
i’d spend the time testing the assumptions behind the generated cases, not just executing more of them.
if the same acceptance criteria shape both the feature and the ai-written tests, you can get a lot of clean coverage around the wrong behavior.
exploratory work, real user journeys, and reviewing why each case exists feel like the higher-value move.
I’d spend more time validating requirements, not writing more test cases.
AI can generate test cases quickly, but it can’t tell whether the requirements are complete or correct. My Validation-First approach is to validate business rules, assumptions, and edge cases before using AI for test generation.
Better inputs lead to better AI-generated tests. Human judgment is still essential where it matters most.
I’m still one for human in the loop. Hence I think I wouldn’t have too much free time at hand to do anything else. However, given the new stuff with agents and sub agents, I think I’d spend time thinking about improving the quality improvement.
Fun question. A lot of the smart answers here already land on “validate the requirements first” and “test the assumptions, not the cases” – I’m with all of it.
Here’s the flip side I’d spend my time on: hunting what isn’t there.
AI is brilliant at answering the question you asked – and blind to the one you didn’t. It’ll faithfully generate 500 tidy cases for the requirements you have (including the wrong or half-baked ones), and the volume looks like thoroughness. It has no idea what it didn’t cover.
So, the job stops being “write the cases” and becomes “find the gaps,” in both directions:
- Upstream – with the Business Analysts and product before a line’s written, dragging out the edge cases and unstated assumptions nobody put in the user story.
- Downstream – staring at the pile AI produced and asking, “what risk did all 500 of these quietly dance around?”
The test case was never the valuable part. The thinking that knows where this product actually breaks was. AI can have the typing – I’ll keep the judgment.
Curious if anyone’s finding AI good at surfacing edge cases it wasn’t handed, or whether that blind spot is holding like I suspect.
I think the more interesting question is what happens to the tester’s understanding if AI writes 90% of the test cases?
Writing test cases isn’t just documentation, it’s a thinking exercise. It’s where I uncover ambiguous requirements, conflicting business rules, missing edge cases, and questions I didn’t know I had. The biggest risk with AI-generated test cases isn’t that they are poor, it’s that we stop asking the questions we used to ask while designing them.
If AI takes over most of the writing, I would want to spend the saved time on the work AI can’t truly do: challenging its assumptions, validating whether the requirements are actually testable, exploring scenarios the requirements never described, and investigating the production risks and customer workflows behind them.
The final step would be asking whether the generated tests genuinely increase confidence or simply increase coverage. Those aren’t always the same thing.
For me, the goal isn’t for AI to replace the thinking that happens during test design. It’s to automate the repetitive parts while preserving and ideally amplifying, the critical thinking that makes testing valuable.
Talking with colleagues, lunch, drinking coffee, scrolling insragram reels, starting working later and leaving earlier + as alway meaningless meetings/calls, implemeting new “visionary” ideas of bosses utilizing AI
@QA_Tidy_Boy I would say with AI in picture it definitely helps in augmenting with all the stuff we do… but humans still own the final call…This is exactly why skills like skepticism, validation, and critical thinking are becoming more valuable and would definitely spend more time learning that… The testers who thrive over the next decade will be the ones who ask better questions, spot better risks, validate output instead of trusting it, understand the system deeply enough to know when the AI is wrong, find that gaps which AI failed to find and pair their own judgment with the AI’s speed. That combination is hard to beat.
Yes I am comfortable AI writing test cases
- With AI the estimated time should get reduced to for all activities. If it was taking 5 days earlier lets say then now with AI we should not estimate the same
- Time is spent in review as AI creates a lot of unnecessary test cases
- To remove extreme negative cases it has created
- In Iterating through by providing more and better context so it can correct itself
- After the test cases are created i try to create a markmap ( an extension in VS code) which converts markdown file to a beautiful map + you can export it as a html
- So you can share that with Dev or have a review call and go through the html which is easy to review
Use it as Co-pilot and not pilot. If we sit with AI it takes time to refine / review.