AI in software testing is changing the work on two levels: as a tool for generating test data, analyzing documents and reviewing code, and as a threat to existing testing roles. Deep system knowledge and domain expertise remain indispensable, because AI agents can cover superficial tests but cannot yet reliably spot side effects in grown code bases.
Key Takeaways
- AI-assisted code generation brings short-term productivity gains, but in the medium term it will trigger a wave of major refactorings and software failures, because nobody keeps the big picture of grown systems in view.
- QA lead positions and senior testing roles are currently being cut, while regulated industries have so far been hit much less by these job losses.
- Non-functional requirements such as security, accessibility and performance are standard knowledge in testing teams, but product management and C-level decision-makers are often completely unaware of them.
- Knowledge of system architecture and domain expertise are the skills that make testers hard to replace, because AI can deliver high-level coverage but has no context about your own product domain.
AI Testing Is Changing the Work, Just Not the Way Management Hopes
Artificial intelligence dominates the conversation in the testing community, from conference stages to the local newspaper. For AI in software testing itself, the topics mostly revolve around generative AI: generating test data, creating test cases, summarizing and evaluating documents.
The next step is agents and the Model Context Protocol (MCP), which lets you build agent capabilities on top of an API. Testers should start exploring this now, even if the company strategy isn’t pushing them. The tools are freely available, and you can build and try out your own systems.
AI is ready for production wherever a lot of the work involves text and documents. From his work at Techniker Krankenkasse, a large German health insurer, Daniel Knott cites analyzing, summarizing and evaluating documents as a case where AI already runs in production applications. Many other companies are still finding their way and are building first prototypes into existing products.
Where AI in Software Testing Pays Off: Test Data and New Views of the System
Test data generation is one of the strongest uses of AI in testing, especially when access to your own data sources is secured through private LLMs. It adds concrete value to daily work.
A second effect is a new way of looking at data. If you let Copilot or another LLM access your own repositories, you gain new insights into how the data fits together. That leads to better decisions: do I want to test at this level, or automate completely?
None of this makes the tester’s role as a domain expert disappear. Quite the opposite. Only someone who understands the business context can judge the AI’s results in the first place. AI provides more information about a system, but making sense of it is still human work.
A Wave of Generated Code Is Coming, and It Will Be Expensive
Vibe coding and AI-assisted development drastically lower the barrier to building software. For young founders, this is a good moment: with few people and little technical know-how, a product gets to market faster.
The bill comes later. Over the next few years, expect a wave of major refactorings, failures and software outages. A “fix” is generated in no time, and a code review is quickly handed off to the AI. What is missing is the big picture: whether one changed line in a grown application triggers a side effect somewhere else.
That is exactly where new demand comes from. We need people who can understand, dig into and assess entire system landscapes. Daniel expects this demand to return once companies understand what it means to use AI sensibly, rather than only as a way to cut staff.
“Unfortunately, we first have to get through this valley of layoffs. On LinkedIn, very good people in testing and engineering are looking for new jobs. I believe that demand will come back, because companies first have to understand what it means to use AI sensibly.”
(Daniel Knott)
The Worry About Your Own Job Is Real
Besides AI, what moves the community most is uncertainty about the future. Anyone who checks LinkedIn every day sees people looking for work and posts being shared on their behalf. Right now, AI has a noticeable impact on jobs in testing.
Senior positions are especially scarce. QA lead roles and roles leading toward test management or general management are hard to find at the moment. Those responsibilities are being cut or piled onto someone else.
The pressure isn’t coming from AI alone. The economic situation and old-school management thinking reinforce each other. At the C-level there is often little real insight, and then testing is the first place the budget gets cut. Communicating the value of testing to the business is an old task for the community, but in moments like these it is rarely in any one person’s hands.
Why Regulated Industries Are Less Affected
There are differences, though. In regulated environments, the impact has so far been smaller than in SaaS applications and startups. And in German-speaking countries, the hire-and-fire mentality that is more pronounced in the US is less common. Neither is a guarantee for the future.
Low-Code and No-Code Are Here to Stay, Because Testing Know-How Is Missing
The trend toward low-code and no-code automation tools continues. Companies turn to them because they lack testing expertise. Instead of a dedicated test automation engineer, the tool fills the gap.
At the same time, teams are looking more closely at which levels they actually want to automate. Products have become more complex: lots of back-end code, large interfaces, several front ends, and in some cases desktop and embedded systems on top.
This leads to a rework of automation strategies. UI-driven tests with Cypress or Selenium matter, but they are expensive to run. Mature test suites are being refactored on a larger scale to restructure the strategy along the classic test pyramid and get the value back out of them.
The Test Pyramid and Agile Testing Quadrants Get Teams Talking
In many companies, the people involved simply don’t talk to each other. Developers write their unit tests and small integration tests, and a test department somewhere does UI automation, often for things that are long covered at lower levels.
Two models help close that gap. The test pyramid works well for talking with developers about which levels code is automated at. The agile testing quadrants bring non-functional requirements, acceptance criteria, performance and observability into the picture, and with them a whole-team view of quality.
Daniel uses both models in the software testing training he teaches for a product bootcamp, with participants who have no testing background. The feedback from the companies: the models help mainly because they get the conversation started at all. If that works, a lot has been gained.
Quality Belongs in the Team, Not in a Separate Department
Whoever does the testing should be deeply embedded in the product team and the product lifecycle. That is the only way to build up, over time, the business and domain expertise needed to get to the bottom of a system.
That has consequences for the skills question. AI will be able to cover broad, superficial high-level testing. Agent tools aren’t perfect yet, but they already handle simple high-level topics well. That is not a loss. It takes load off the team.
The tension remains, though: more shared responsibility within the team makes everyone broader, but nobody deep anymore. That is exactly why there is a strong case for deliberately keeping testing expertise in the team, instead of spreading it across everyone until it disappears.
Which Skills Matter Beyond AI
Blanket answers fall short, because everyone starts from a different place. For beginners: build the fundamentals with books, blogs and YouTube, and consider a certification to learn the vocabulary of testing.
Beyond that, these areas are especially worth the effort:
- System architecture and system modeling. Modern applications are no longer just a back end, a front end and an interface. Cloud systems and nested landscapes are part of the picture. Testers who understand them ask sharper questions.
- Industry and business knowledge. This is the basis for judging AI results. AI hallucinates often, a lot and convincingly, sometimes so well that you don’t notice.
- Non-functional requirements. Old news for testers, often not for product teams. Ask when non-functional requirements were last discussed, and you frequently get blank faces.
- Security and accessibility. Even without being an expert, you can read up, experiment and try new things here. The field is endless.
The most important skill is the attitude behind all this. Don’t stay stuck at Foundation Level. Keep developing and keep learning throughout your career, but in moderation. It is easy to overwhelm and stress yourself with the flood of new topics. A steady flow takes you further than panic.
The Cycles Are Getting Faster, and New Territory Rewards Curiosity
The mobile wave around 2010 was considered fast: the iPhone in 2007, the first app stores in 2009, then the big push from around 2010 to 2012. AI swept through the industry even faster. Computing power on mobile and desktop devices keeps speeding the cycles up.
In phases like this, there is no standard yet. AI brings new quality criteria for which there are neither books nor established methods. The only option is to experiment until a standard emerges.
That is the opportunity. If your company gives you time to dig into a new research topic, you have a rare chance, a bit like a science fair for professionals. Daniel traces his own path back to this: he was allowed to share what he learned in the early mobile days, and that led to his book and his YouTube channel. The testing community thrives on this culture of sharing and supporting each other, and that is what moves it forward.
Frequently Asked Questions
Where can AI already be used productively in software testing today?
AI is most advanced in areas that involve extensive work with text and documents: document analysis, summarization, and evaluation are already being used in production applications, such as at Techniker Krankenkasse. Test data generation is also a strong area, especially when access to proprietary data sources is secured via private LLMs. Many companies, however, are still in the prototyping phase.
What are the implications of AI-generated code for the quality of established systems?
We can expect a wave of major refactorings, failures, and software outages. Vibe-coding lowers the barrier to building software; a fix is quickly generated, and a code review is quickly delegated to AI. What’s missing is the big picture: the question of whether a single changed line in an established application triggers a side effect. This is precisely what creates a new need for people who can thoroughly understand system landscapes.
Which testing roles are particularly affected by the current job cuts?
Experienced positions are in the shortest supply. QA lead positions and roles in testing or general management are hard to find; they’re being eliminated or added as an extra burden for someone else. The pressure isn’t coming solely from AI: economic conditions and outdated management thinking are compounding the issue, and C-level executives often lack the insight needed to understand why testing is the first area to face budget cuts.
Are all industries equally affected by AI-driven job cuts in testing?
No. In regulated environments, the impact has so far been less severe than in SaaS applications and startups. There are also regional differences: In German-speaking countries, the “hire-and-fire” mentality (which is more pronounced in the U.S.) is less prevalent. Neither of these factors is a guarantee for the future; they merely shift the timing and severity.
Why are companies turning to low-code and no-code test automation?
Because they lack testing expertise. Instead of filling a dedicated role for a test automation engineer, the tool fills the gap. At the same time, teams are taking a closer look at the levels at which they actually want to automate, since products have become more complex: a lot of backend code, large interfaces, multiple frontends, and in some cases, desktop and embedded systems.
How do you get a conversation about test levels and non-functional requirements going within the team?
Two models can help. The test pyramid is useful for discussing with developers at which levels code is automated. The Agile Testing Quadrants bring non-functional requirements, acceptance criteria, performance, and observability into the picture. Feedback from companies and from training sessions with participants without a testing background shows that the real benefit lies in the fact that communication begins in the first place.
Should testing knowledge be distributed across the entire team, or should a dedicated testing role be maintained?
There are many reasons to deliberately keep testing expertise within the team rather than spreading it across everyone until it disappears. More holistic responsibility broadens everyone’s skill set but doesn’t deepen anyone’s expertise. Those who perform testing should be deeply embedded in the product team and the product lifecycle, because only then can domain expertise grow over time, leading to a deep understanding of the system.
What skills, aside from AI knowledge, make testers valuable in the long term?
System architecture and system modeling are at the top of the list because modern applications consist of cloud systems and nested landscapes, and those who understand them ask more targeted questions. Added to this is industry and business knowledge as a foundation for evaluating AI results, since hallucinations are common and convincing. Non-functional requirements, security, and accessibility are also worthwhile areas to focus on. The key is maintaining the right mindset: don’t stay stuck at the foundation level.
