What current human favouring skills would you be comfortable handing over to an agent?

I was revisiting this really good thread from last year What key skills regularly help you tackle testing tasks?

I did not want to hijack that with a significant tangent but felt those activities listed there would be ideal for having this one.

Some teams seem to be looking at testing from a humans out the loop angle, some with humans in the loop and my own personal favourite of humans at the helm.

If we consider the skills listed in that thread how do they fare when we consider the Ai angles.

Firstly do those activities lean towards human or machines strengths, there will be a mix and of course even human lean activities can heavily leverage from tools but are there still fairly clear activities that suit humans or machines. Not in this list is say scripted testing, I’d tend to say this does lean towards machine strengths but on this list on the face of it many seem to lean towards skilled human strengths but it is still worth challenging that in my view.

Then if we consider trained agents, can we give it human strengths to a level or is there still very clear activities that we would rule our for agent execution?

My off the cuff view on the four activities listed, I am open to be persuaded on all of these.

Investigating a bug report. I’d go with agents being capable of doing this fairly well.

Collaborating on a new feature - still human leaning in my view.

Exploratory testing - I’m still all in on human with this, though could drift towards human at the helm with lots of agent usage for individual experiments and the human skills on the critical thinking and decisions between each loop. I sort of get the idea that machines cannot explore that some people take but its also good to challenge that thinking.

Prioritizing bugs - agents can likely do a first round of this given appropriate skills but final decision for me is still a collaboration activity, too many tacit knowledge nuances with this one which is usually why its a 3 amigos decision.

In all cases though I would still have skilled humans at the helm but with a consideration of critical distance in place, not in the loop, at the helm.

What about everyone else, anyone all the way in on agents yet?

I’ve seen systems that have little to no coded tests (regression, unit, integration etc)
Testers working on such systems usually don’t find time to write them whilst ensuring coverage too.

This is where AI should help, the tester should explain to it how the system or part of it works once, and then the AI becomes the regression tester. Should something not work, the AI is free to log a bug.

but then, what is or isn’t a bug? maybe that should be a human decision?

Yeah, there is a very recent thread on this kind of decision, it depends on whether you can explain the core value of a product - by giving the agent access to the bug database, it will have to look for past duplicates for example first. All possible, but often inefficient, so just logging is going to be enough I would think.