Skip to content
QAREVOLUTION
Subscribe
SOFTWARE TESTING

Why Now Is the Worst Time to Quit Software Testing

Ron Wilson · QA Revolution
Oct 12, 2026
Comment Share

I broke this down in the video above, with four sheets of paper and a lot of ripping. Below is the written version with every source, the two reasons I tore up on camera, and the one reason I could not bring myself to tear.

Somebody on our team is about to quit software testing at the exact moment it becomes the most valuable job in the building, and it might be you. If you are a tester, an SDET, or a QA lead who has looked at the layoff headlines and the AI demos and quietly started updating a resume for a different career, this article is for you. I wrote down three reasons to stay. Two of them are good reasons, but only one of them is the real reason, and the difference between them matters more than the reasons themselves. By the end you will know which one survives, why the other two fail under pressure, and what we can do about it on Monday morning.

Here is why the timing matters. In the last twelve months the question changed. It used to be “will AI replace testers?” and now it is “developers are shipping AI-written code at a pace no human can read, so who is checking it?” That is a different question with a very different answer. Testers who leave now are leaving in the middle of the single largest increase in untested code our industry has ever produced.

The cost of getting this wrong runs both ways. If we stay for the wrong reason, the next bad quarter or the next automation tool will push us out anyway. If we leave for the wrong reason, we walk away from a job that is about to be redefined by the people who stayed in the room.

I have hired testers for more than twenty years, and I have watched this profession get declared dead by offshoring, by agile, by DevOps, and by the promise that developers would write their own tests. Every single time, the job changed and the people who stayed got paid more to do the new version. What I learned from those cycles is that the reason we stay decides whether we survive the next one. So let us look at the three reasons, in the order I ripped them.

Reason One: There Is More AI-Generated Code Than Ever and Fewer Testers to Check It

More code is shipping faster than ever, almost half of the AI-written portion carries a security flaw, and the humans reviewing it are tired.

Let us start with adoption. The 2025 Stack Overflow Developer Survey found that 84 percent of developers now use AI tools in their workflow, and that is not a vendor slide. It is the largest annual survey of working developers on the planet. Code volume is up, and the time any single human spends reading any single line is down.

Now look at what is inside that code. Veracode published its 2025 GenAI Code Security Report in July after running 80 curated coding tasks across more than 100 large language models. The AI introduced a security vulnerability in 45 percent of cases. Java was the worst performer with a failure rate above 70 percent, cross-site scripting failed 86 percent of the time, and log injection failed 88 percent of the time. These are not exotic bugs. They are the same OWASP Top 10 items we have been testing for since before most of these models existed.

Developers used to tell us it works on my machine, and now they tell us it works in the chat window. It is the same energy with less evidence behind it. The math is simple. We have more code, written faster, by something that is wrong almost half the time, reviewed by a human who has 400 lines in front of them and a standup in ten minutes. That is a testing problem, and testing problems need testers.

So why did I rip this one up? Because it is a volume argument, and volume gets automated. If the only thing keeping us in testing is that there is more to check, somebody will build a bot to check it. It might be a static analysis tool, or it might be a second AI agent reviewing the first one. The pile getting bigger is true, and it is still not enough. We do not stay because the pile grew.

A flood of teal and red code blocks pouring down toward a single small green magnifying glass

Reason Two: AI-Skilled Testers Keep Their Jobs and Get Paid More

The U.S. Bureau of Labor Statistics projects 10 percent growth for software developers, quality assurance analysts, and testers from 2025 to 2035, and workers with AI skills now earn a 56 percent wage premium.

Pull up the BLS Occupational Outlook Handbook and look at the combined category for software developers, quality assurance analysts, and testers. It is projected to grow 10 percent from 2025 to 2035, with about 106,100 openings per year on average, and median pay for software quality assurance analysts and testers was 104,300 dollars in May 2025. That is the government forecast, not a recruiter’s pitch.

Now stack the PwC 2025 Global AI Jobs Barometer on top of it. PwC analyzed close to a billion job ads across six continents and found that workers with AI skills earned a 56 percent wage premium over the same role without those skills. A year earlier that premium was 25 percent, so it more than doubled in twelve months. PwC also found that jobs grew even in the occupations most exposed to automation. The roles did not vanish. They changed shape, and the people with the new skills got paid for the new shape.

What does that mean for us? The testers who learn to test with AI, and to test AI itself, are not getting replaced. They are getting a raise. The testers who refuse are competing with a bot for the boring half of the job, and the bot does not take lunch. We have to pick a lane.

This is also a good reason, and it is also not the reason. It is a market argument, and markets are weather. Every tester who was laid off in 2023 was told the market was great in 2022. A forecast is a forecast, and if a wage premium is the only thing holding us here, the next bad quarter takes us out no matter what the BLS says. That is why I ripped this one up too.

A balance scale tipping toward a tester figure with a glowing green AI skill node, with a forecast line fading into clouds

Reason Three: Somebody Has to Be the Person Who Does Not Believe the AI

The one skill AI cannot generate is the instinct to say “prove it,” and professional skepticism is the entire job description of a software tester.

Go back to that Stack Overflow survey. The same 84 percent of developers use AI tools, but 46 percent of them say they do not trust the accuracy of what it gives them, up from 31 percent the year before. Usage is going up while trust is going down. And what was the number one frustration, cited by 66 percent of developers? AI solutions that are almost right, but not quite.

Almost right. We have lived in almost right for our entire careers, because that is the whole job. Almost right is the gap between the demo and the incident report. It is the login flow that passes every test and breaks for one user in one browser with one accent character in their last name. We find those for a living.

The 2024 DORA report from Google Cloud adds the system-level view. As AI adoption increased, delivery throughput dropped an estimated 1.5 percent and delivery stability dropped an estimated 7.2 percent. Code goes in faster and comes out more broken. That is exactly what we would predict after handing every developer on earth a fast, confident teammate who is wrong 45 percent of the time and never once says it is not sure.

Here is the part that matters. The developer’s job is to ship and the AI’s job is to agree, because it was trained to be helpful rather than suspicious. Nobody in that loop is paid to doubt, except us. That is our entire profession: professional skepticism with a reproduction case attached. Developers invented AI to write more code, and testers are the reason that code gets to stay in production.

This is the sheet I could not rip. Reason one was about the pile and reason two was about the paycheck, but this one is about us. The pile shrinks, the market turns, and the person whose default setting is “I do not believe you yet” still has a job. That does not get automated. That gets promoted. I taped it to the monitor, and the rip that does not happen is the moment in the video I would send to any tester who is wavering.

A row of grey figures bowing toward a glowing AI orb while one green-lit figure stands upright holding a magnifying glass

Why the Volume and Market Arguments Fail and the Skepticism Argument Holds

Volume and market arguments depend on conditions outside our control, while professional skepticism is a skill we own and AI structurally lacks.

Let us be precise about the difference, because it is the difference between a reason that survives a layoff and a reason that does not.

The volume argument says there is more to test, and that is true today. But volume is a condition, not a skill, and conditions change. If AI code quality improves, or if a second layer of AI review catches more of the 45 percent, the pile shrinks and the argument goes with it. A reason that depends on the other side staying bad is a weak reason to build a career on.

The market argument says testers with AI skills earn more, and that is also true today. But a wage premium is a signal of scarcity, and scarcity erodes as more people pick up the skill. The 56 percent premium is an invitation to move now, not a promise that it will still be 56 percent in three years.

The skepticism argument is different in kind. Large language models are optimized to produce the most plausible next token, and plausible is not the same as true. The model has no internal mechanism that flags the gap.

The Stack Overflow data shows that developers feel this every day, the Veracode data shows what it costs, and the DORA data shows what it does to delivery stability at scale. Every one of those problems is a gap between what the system claims and what it actually does, and closing that gap is exactly what we are trained for. It is not a condition of the market. It is a trait of the technology, and it is a trait of the tester.

What to Do Monday Morning If You Are Staying in Software Testing

Stop testing only what the AI wrote and start testing how the team uses it, learn one AI tool well enough to break it, and bring a bug story to your next one-on-one instead of a certification.

There are three things we can do this week, and none of them require a budget.

First, test how the team uses AI, not just the output. Find the pull request that says generated with AI, lightly reviewed. Lightly reviewed is a tester’s trigger phrase. Ask who reviewed it, how long they spent, and whether the tests in the pull request were generated by the same model that wrote the code. A model grading its own homework is a process defect, and process defects are ours to raise.

Second, learn one AI tool well enough to break it. Not well enough to use it, well enough to break it. Hand Claude a failing test and ask it to fix the problem, then check whether it fixed the code or simply changed the assertion so the test passes. That difference is our new job description. When I tried this on my own suites, the pattern showed up more often than I liked: the model will happily make a red test green by making the test worse. I walk through what to look for in the video.

Third, write down the last bug you found that no tool would have caught. It might be the race condition that only appeared under a specific data load, or the business rule that was never written down anywhere. Bring that story to your next one-on-one, because that is our leverage. Not a certification. A story with a reproduction case attached.

If you want the prompts I use to catch almost right code, including the one that asks the model to list what it did not test, they are in the companion resources for this series on QA Revolution.

The honest answer about AI and software testing is not optimistic or pessimistic. It is specific. There is more code than ever, and that alone will not save our jobs. AI-skilled testers keep their jobs and get paid more, and that alone will not save them either. What saves them is the one thing the technology cannot do and the one thing our profession has always done, which is to refuse to believe the system until it proves itself. That is the reason that survives, and it is the reason I could not tear up.

I went deeper on all three, with the sheets and the rips, in the full video. It is worth the eight minutes if you are on the fence.

Here is my question for you: what is the one bug you caught that AI would have shipped? Not one you read about, one from your own team. Drop it in the comments and I will respond to as many as I can this week.

Subscribe for more practical breakdowns on AI and software testing for working testers. No tutorial farms. Real data, real experiments, and honest results.

Sources

Every figure in this article was pulled from the primary source on October 12, 2026.

More from QA Revolution

All videos →