# Manual testing is dead. Automation is dead. What’s really changing?

> AI is killing mechanical test execution, not testing. Four testing leaders on what dies, what grows, and what testers should do on Monday.

- Series: The Future of Software Testing (No. 02 · Discussion one)
- Published: 2026-10-02
- Reading time: 9 min read
- Canonical URL: https://thefutureofsoftwaretesting.com/articles/manual-testing-automation-dead-whats-changing/

> Automation is dead. Manual testing is dead. Everything in testing is dead, because AI is going to take over.

That was the opening provocation when four leaders - Vu Lam, Mush Honda, Cristiano Caetano, and Alex Martins - sat down for the first discussion in The Future of Software Testing series.

Nobody dismissed the claim. Nobody fully accepted it either. What emerged was a more precise answer - and a more uncomfortable one. So what is actually dying in testing?

## The mechanical layer is disappearing

Look at where AI is advancing fastest in testing, and a pattern starts to emerge. Pre-written, step-by-step checking. Regression suites repeated again and again. Routine test generation. Test maintenance. Result analysis. Much of this work has historically consumed enormous amounts of tester time without always demanding equally high levels of judgment.

> The repeatable layer, the mechanics of it, is what’s going away. The judgment part, however, absolutely continues.
>
> - Mush Honda

That distinction, mechanics versus judgment, became the spine of the whole conversation. As AI absorbs more of the mechanics, the value of the tester moves upward. But this is not the first time technology has changed where testing work happens.

## The bottleneck doesn’t disappear. It moves.

Cristiano Caetano has heard “testing is dead” for almost 30 years, and he’s watched what actually happens each time: the bottleneck relocates. Manual execution became a bottleneck, so teams automated it. Automation then created new bottlenecks: building and maintaining brittle test suites, analyzing large volumes of results, and separating genuine failures from false positives and false negatives. Now AI is beginning to accelerate those tasks too. And the bottleneck moves again.

> Upward and onward to the human - because the human needs to validate whether all those aggregated results make sense: should we release, or should we double down somewhere else?
>
> - Cristiano Caetano

His analogy is the modern car factory. Automation can perform most of the repetitive work, but people remain responsible for the tasks that require finer judgment. Testing is heading the same way. People will still be there, but they will contribute at a different level. And that shift has consequences for the kinds of testing roles most exposed to AI.

## The easy jobs go first

For generations of testers, the first rung of the career ladder has often looked something like this: Here is a set of tests that someone else already wrote. Go run it.

> The easiest testing job in the world is to run the test. Well, guess what? That’s the easiest job for AI to do as well.
>
> - Vu Lam

The instructions already exist: what to do, what data to enter, what outcome to expect. That makes repetitive execution one of the parts of testing most exposed to automation.

Running pre-written tests - and increasingly writing routine ones - is becoming less defensible as a standalone role.

“The pool of purely manual testers will shrink. There’s no denying it, and pretending otherwise doesn’t help anyone,” Vu emphasized. But this is where the [“manual testing is dead” argument](https://katalon.com/resources-center/blog/will-ai-replace-manual-testers) starts to break down. Because not all manual testing is mechanical. And not all valuable testers were hired primarily because of their ability to execute tests.

## Where human expertise still changes the outcome

For all the talk about autonomous testing, Alex Martins sees a different reality across many large enterprises: advanced technology often coexists with substantial manual testing. Some of those testers were not hired primarily for technical depth. They were hired because they understand the domain.

> Some organizations can be incredibly advanced in certain areas and still have a lot of manual testing. In healthcare, I also see manual testers who come from nursing and clinical backgrounds. Their domain knowledge is exactly what makes them valuable.
>
> - Alex Martins

That expertise becomes even more important when AI enters the process. AI can generate plausible-looking tests for requirements it does not truly understand. Without enough domain depth, a tester may struggle to recognize when a confident-looking output is actually wrong.

> If you’re a manual tester and you don’t have domain knowledge, you’re basically going to introduce more problems into the whole life cycle.
>
> - Alex Martins

Giving an inexperienced tester AI does not automatically create an accelerated tester. Without enough context and expertise, AI can simply accelerate the wrong decisions. And that points to a larger shift in what creates value in testing.

## What becomes more valuable

If AI can take on more execution, what becomes more important for the people who remain accountable for quality? Three capabilities stood out in the discussion.

### Judgment

AI can produce more tests, results, and evidence than teams have ever had to process. Someone still has to decide what that evidence means - and whether the product is ready to ship.

### Domain expertise

AI can generate plausible outputs without fully understanding the operational, regulatory, or customer context behind them. Domain experts know when something that looks correct is actually wrong.

### Accountability

AI can contribute to a decision. It cannot own the consequence. When software fails in production, accountability still sits with people. As Mush put it: “You can’t blame the AI.” There is another reason these capabilities matter. AI does not automatically fix a weak testing process. It can scale one.

> AI magnifies the state of your testing process. Good practices get amplified into great results. Bad coverage gets amplified into confident-looking noise. More AI-generated tests do not equal more confidence - sometimes they just mean more volume.
>
> - Mush Honda

The implication is simple: More output is not the same as more confidence. So the goal for testers is not simply to use more AI. It is to learn where and how to trust it.

## What testers should do on Monday

> **Stop evaluating AI like a tool and start onboarding it like a teammate.**

![Fig. 01 - Onboard, supervise, extend trust: context in, confidence earned step by step](https://thefutureofsoftwaretesting.com/_astro/g-teammate.wvBq1ojV.png)

*Fig. 01 - Onboard, supervise, extend trust: context in, confidence earned step by step*

“AI is a member of the team, not a takeover of the team,” Vu shared. A talented teammate still needs context: the domain, the data, the definition of good, examples of strong work, and clear boundaries. Then you supervise the work and gradually extend trust as evidence builds.

> If we hired an amazing tester who didn’t have the domain context, they’d write test cases that are acceptable - but not at the level we expect. AI is no different.
>
> - Mush Honda

> Until I have enough evidence that AI delivers the result I’m expecting, I’ll continue validating.
>
> - Cristiano Caetano

And not every testing problem needs generative AI. Deterministic automation is still cheaper, faster, and more reliable for plenty of scenarios. Generative AI is non-deterministic - powerful where judgment-heavy scale is needed, wrong where repeatability is the point. The goal is not to replace deterministic automation with AI.

## So, what actually died?

Manual testing? Not exactly. Automation? Certainly not. As Vu put it: “The easy part of manual testing is dead. Manual testing in another way is not dead. It will continue to thrive, because there are limitations to what AI can do.”

What is disappearing is mechanical manual execution: pre-scripted, repetitive, low-judgment work that technology can increasingly perform faster and at greater scale. [Manual testing as exploration](https://katalon.com/resources-center/blog/exploratory-testing), investigation, and human judgment is not disappearing. Automation is not disappearing either. In many cases, deterministic automation remains the best solution. What is dying is the assumption that a tester creates value primarily by executing repeatable steps.

As AI takes on more of that execution, the center of gravity moves upward - toward judgment, domain expertise, and accountability. The question is no longer whether AI can do parts of your testing job. It already can. The more important question is whether you are building the capabilities that become more valuable as it does.

> **We would rather be argued with than agreed with. If your team has run this experiment - more rigour upfront, or more validation throughout - tell us what actually happened.**