
30.09.2026
AI accelerates software development. Code can be created faster, prototypes can be developed in shorter time, and new features can be delivered significantly faster. However, this creates a new challenge for software testing.
As development becomes faster, quality assurance must also keep pace with this speed. Classical testing processes often rely on stable requirements, predictable development cycles, and fixed testing phases. AI and agile development change these prerequisites. Requirements can change more quickly, new software increments emerge continuously, and test teams have less and less time to wait for finished software.
Daniel Horn has been working in quality assurance, software testing, and test automation for about 14 years. In this expert interview, he explains why AI is intensifying existing challenges in quality engineering, what the role of early testing and automation is, and how companies can adapt their testing strategy to the new development speed.
In the full expert interview speaks Daniel Horn explains the challenges of quality engineering in the AI era and explains how companies can adapt their testing strategy to the increasing development speed.
[00:06] Software Testing in the AI Age. Why quality assurance needs to be rethought. Warm welcome to today’s accompio focus topic from Dresden. I’m Daniel Horn here. Daniel, nice to see you there. Could you introduce yourself briefly and explain what role you have in the field of Quality Engineering and Software Testing and what challenges you face on a daily basis in your work? Thank you so much for being here. Yes, my name is Daniel Horn, as mentioned, I’ve been working in quality assurance as a consultant for almost 14 years. Mainly testing, manual testing, test automation, test management, everything you could imagine in quality assurance. And over the last few years, my focus has slightly shifted
[00:52] Moving on to: How can quality be better designed in agile transformation? How can DevOps teams be better, that is, everything where you think: high-quality software. How can I achieve this? How must my testing processes be designed for this? Now AI is accelerating this entire development process. What does this new speed mean for the role of Quality Engineering? Yes, in principle, AI will initially only reinforce what we have already seen about agile change. That is, this whole topic, softer requirements, faster iteration cycles, etc., and so on. This has before AI already reinforced the whole of Quality Engineering
[01:39] put to the test. Now with AI, with even faster prototyping by the developers and by the programmers, this shortcoming, which has emerged over the past few years, is naturally becoming even more visible. And yes, the main challenge will be: How do we communicate quality? How can we actually bring quality into the software in the end? Constructively, but also analytically ensuring that what we want is actually delivered. And at what point does the classic development and testing process come under pressure first? Already quite early on. So these classic testing processes assume that,
[02:25] Having stable requirements documents means that you already had to think about what you actually need beforehand. Agility, AI changes the whole thing. We want to become faster, we want to be faster on the market and we are checking whether what we have actually built is already there, and everything you once learned from the textbook assumes that we have stable requirements. We already know what the software should look like in the end. And here we are actually coming under real pressure for the first time. We don't really know that at all. And the second thing is of course that the faster these deliveries come, the more pressure there is on test automation or test execution. We can't afford test pipelines of one to two days anymore.
[03:12] Where are the biggest bottlenecks currently for Quality Engineering teams when preparing, conducting, and evaluating tests? What is actually visible is actually during execution. So as already mentioned, the delivery is getting faster and faster, and somehow this bug wave that is just pushing ahead of the developers with Hey, we have more and more features, we have more and more deliveries, more and more commits. Classic test teams just can’t keep up with that because the testing methods, the testing process, are naturally still designed to deliver on classic systems. So we deliver somehow weekly, we deliver somehow every two weeks; that can be planned.
[03:58] Now that we’re being told that we deliver everything every two hours, we can no longer expect to have two days of testing time at any given time; we have to somehow make it work differently. And what does this change specifically mean in the day-to-day work of a quality engineering team? In a nutshell, it can be expressed in one sentence. So imagine that you’re trying to shoot moving targets in the dark. So, you know, what was valid yesterday is no longer valid today. What worked yesterday doesn’t necessarily have to work today, because of course new software increments have come along. And so on and so forth. And this speed is actually presenting testers with major challenges at the moment.
[04:45] And that often really just means taking a step back and looking at where we can test, where can we improve quality? And over the last few years, unfortunately, it has kind of crept in that functional testing is very, very much end-to-end testing. So we check when the software is finished. We make sure that all the features are there, and we have to go further and further towards what is actually being taught. Many small tests, many early tests, and few tests in the later development stages. And are there any changes that companies absolutely need to make so that the quality can keep up with this enormous development speed? The teams need to work more closely together, so basically everything that makes up the whole agile manifesto, and yes,
[05:33] The agile transformation, DevOps, has been preached for 20 years: Teams need to grow together, teams need to communicate better. As already mentioned, this is all reinforced by AI now, and with this speed it suddenly becomes clear that this communication is essential. A separation between business analysis, developers, programmers, and testers is just not tenable anymore; everyone needs to talk to each other. Everyone should have the same understanding of the software. And from the very beginning, maybe testers should be involved in the requirements analysis as well. What can be tested? Where are the edge cases that an AI or the code ultimately should consider? And the more information I give to an AI, the better the code will be
[06:19] Of course, an AI can only work with what it knows, and often you simply have such intrinsic internal knowledge that is not codified, and we need to learn more to codify that knowledge. We need to learn more, to say from the outset that this is what we want, even if it’s just the first increment at first. But that’s our initial quality barrier that we’re building. And within that barrier, the AI can work best. What mistakes do companies currently make particularly frequently when they use AI in software development without adapting their testing strategy? You’ve already mentioned that they just use AI without adapting their testing strategy. So, in principle, AI is only used for: we want to program, we want to code, we want to develop.
[07:06] The whole system depends on the fact that the entire pipeline or the entire process actually needs to remain in flow more or less. And if I just speed up one place, then suddenly I have a bottleneck behind that accelerated place because I can’t go any further and in front of that place or in that place there’s suddenly such a vacuum because the processes upstream simply aren’t fast enough to deliver new requirements and new inputs. What I’m observing or what we’re observing is actually that companies tend to focus very, very narrowly on individual places and do it. You can do that here, you can do that there, but in the process of doing that you lose sight of the overall process and simply don’t know anymore,
[07:54] What does my development actually need, what does my programming need, and what do these needs then require in terms of requirements that have to be delivered somehow faster? And so on and so forth. So yes, keeping the overall process more or less in view. And that is the biggest mistake in my eyes. Important concepts in software testing have been Shift Left and Shift Right for years. Is that still relevant or does a new understanding of quality assurance need to be developed? I believe the concepts are even more relevant or more current than what they were five or six years ago. Shift Left actually means that we shift quality assurance tasks to earlier stages of the activity or earlier stages of the process. That is, we have topics like constructive quality assurance.
[08:43] We already determine beforehand what we want to test in the first place. During the requirements analysis, we already check that. Those could be edge cases, those could be use cases. Or they could be use cases that we test. And this idea, of starting testing early and starting with quality assurance measures early, that is what Shift Left more or less stands for. And yes, the AI will just get better, the more I give it, the more or better results it will deliver. On the other hand, Shift Right is a concept that says: quality should come from the domain, or rather, the quality requirements should come from the domain, which of course also relates to this whole communication issue. So we need to communicate more to the demanding domain.,
[09:30] The requesting departments should actually say that it is important to us. Designing an entire software system in advance and saying that it must be capable of doing this is not the way the department wanted it. That wasted my time, no matter what happened. Then I burned tokens on the AI. I lost weeks or months in case of doubt, which was not necessary at all. So, all in all, it is important to include all departments in this quality discovery process and requirement process as early as possible, which ultimately improves the system’s quality in a constructive way. The entire analytical part will eventually have to be incorporated into these processes somehow.
[10:16] How is the collaboration between development and quality engineering changing as AI takes on ever larger parts of software development? Communication is actually becoming increasingly important. That means that all the involved parts have to communicate in some way, talk to each other, and exchange information with each other. The tasks of quality engineering are likely also moving away from this whole analytical part towards: How can I enable entire teams and how can I finally get my product team to think about quality; not just “Hey, I delivered a feature, done”; but also to look again if in doubt.,
[11:02] What could happen if the customer or the user enters some meaningless data in some way? Well, there will be much more of this kind of enablement. And what role do test automation and continuous testing play in this? Specifically, in terms of the speed of modern development processes. In the current speed, it is more or less indispensable. So we’ve seen it before. Test automation has sped up the entire process. We were able to shorten our testing cycles more or less, because we implemented test automation. And if we deliver faster now, the more test automation we have to do. So as a small example, for example: I recently built a small software application over the last few weeks
[11:51] And it has some kind of cycles of 2 to 5 hours after I said that. Hey, is that finished or not? And if I hadn’t had a result after five hours, I would have aborted it and started over. And that only worked because I had set a relatively strict quality corset from the start and said: Test everything you write, test everything that’s being done, and at the end I’ll only do a final reduction, a visual inspection: Does it fit? Is that what I actually ultimately asked for? That’s what was actually delivered, what I wanted? If I hadn’t had that, these cycles would probably have been significantly longer. Or in the end, there would actually have been software that didn’t do what it was supposed to do. Nowadays, AI can create test cases, analyze tester results, or prioritize errors.
[12:37] Where do you see the greatest added value currently? You’ve mentioned it already. It’s more or less, AI systems are predictive systems; they’re like, hey, I correlate a word A with a word B and that leads to a result C. Eventually, you’ll learn that correlation doesn’t equal causality. So, that’s what needs to be tested. But these things, they’re not going away. So, what helps a programmer is also going to help with quality engineering testing automation. Code generation. All of that, in a way. We need to do repetitive tasks.
[13:24] We need to regularly look at results, maybe detect patterns. So, especially when it comes to pattern recognition, AI is simply unbeatable because it can simply say: Hey, in my dataset, this error appears 200 times in 300 test cases. Maybe there’s a root cause, a core error that’s ultimately in the code and not just 200 errors. And we don’t need to analyze 200 tests; we need to focus on one error. So, here, AI will help a lot in the analysis and in code generation and all the things that are ultimately repetitive activities.
[14:11] Everything I say is, let's say, artistic or reflective; in that case, one should use AI more as a partner for brainstorming and similar activities and rather check the results again two or three times. That is, there are of course clear limits to the use of AI. Where do you see the human tester, the human tester as indispensable? As a testing body, so with all the hype that AI is currently creating, we should always be aware: AI is only a tool for the time being. So I can of course use tools properly. But I can also use them incorrectly. And it is precisely with these, with this idea, that one should
[14:57] So, I always keep an eye out to see how I can use AI meaningfully. Do I actually use AI meaningfully at this point, or do I use it only as a means to an end? Since I currently have an AI bot or an LLM, I don’t necessarily have to delegate all my activities to my LLM. And then, when I actually have to look again at what tasks are tasks where I’m definitely faster, I might also be more constructive than an AI because an AI often stays in its training data and can only do very, very little out of the box, as I’ve said. Does the human or human aspect still help in this, the human in the loop?
[15:46] Another aspect is of course always a topic of sovereignty. Because now, let’s say something happens or let’s say the case is set, all my systems are based on AI, all my processes are implemented by AI agents. I actually don’t have anyone else who checks somehow what happens if the providers change their price caps? What happens if Anthropic suddenly says no, in the EU we don’t want to do business anymore? And so on and so forth. So this whole dependence, of course, also has to be somehow ensured that you don’t get too dependent on these systems. So I’m of the opinion that everything that an AI does should at least still be able to be done manually by a human in some way. Of course, it takes longer, but the process shouldn’t stagnate as a result.
[16:34] That is another important approach. The dependency you mentioned. Many people involved in this topic are probably not even aware of it. Now we have addressed the human component. Could you perhaps list some very concrete tasks that still need to be taken on by an experienced quality engineer? Basically, assessing how good the software actually is. That is the whole risk management aspect. Testing is primarily just a risk management activity. So, are we going live with how the software currently looks, or not? And testing reduces the risk or provides signals that ultimately allow us to make that risk statement.
[17:21] And this statement, I would always have someone meet with, because that’s a statement; there are liability issues behind it. Those are legal issues behind it, but there are also financial risks behind it. And letting an AI make that statement, yes, deploy it now and see what happens; I see that a bit critically. So I’m more of the kind of person who prefers to handle the critical processes myself than to outsource them. Daniel, in your view, what should a modern testing strategy look like that leverages the benefits of AI but doesn’t compromise on quality and safety?
[18:07] For me, and I have already mentioned this several times, much of this strategy should be about collaboration, communication processes, and the like. So how can we ensure that quality is actually built in automatically by the AI? How can we already ensure in advance that this quality corset or our quality barriers are more or less in place? Perform early testing, that is, really early on, bring the quality in. Plan many unit tests, many integration tests. When I do that, I usually have already caught 80 to 90 % of the critical errors in some way, and they are no longer there. And quality assurance should
[18:53] It doesn't have to be the last link in the chain. So often it's simply seen as: Hey, we develop what we have—our software. That has to be tested. That's more or less the factory mindset. And the conveyor-belt mindset. In the end, it leads to quality decline and the like. We just have to get more into the concepts of having quality built in automatically from the beginning. And we can only achieve that through automation, early automation, parallelization. In case of doubt, parallelization of test systems, etc., and so on. So we just have to move much more in that direction here.,
[19:39] What could we have already technically done in the past that wasn’t necessary at all because the pressure wasn’t there to do it? That’s what we should be focusing on now. So how can we test 200 pull requests simultaneously, all of which have a different version? Of course, that doesn’t work with one or two stable, fixed testing environments. So much more in: How can I automate scaling of all this, in some way or another, into these processes? Automating tests, deploying deployments, automating environments. To finish, I’d like to ask you for a tip for companies that want to integrate AI into their software development today.
[20:24] What would you advise them? I repeat myself: AI should not be a means to an end. So I shouldn’t use AI just because AI is there and because there’s a hype around it. We should always consider which use cases we really want to achieve with it? In case of doubt, are there even automation solutions or processes that are faster and cheaper than an AI? So an AI is not the fastest thing in the world. It just needs to think, it needs to rethink its processes every time. Being critical about which steps I would like to use AI in. Which steps is AI actually necessary? Where do I have disagreements or in case of doubt
[21:13] Unclear requirements or unclear prerequisites? That’s where AI comes in handy. Everything that’s actually a stupid way of doing things, and stupid clicking through, stupid testing of one class. That doesn’t necessarily have to be done by AI. That can also be done through classic automation solutions. Otherwise, we should think about AI as a whole process technology, not in individual processes, but as a whole process technology. That’s where workshops with all the stakeholders are helpful. That really means going back to the drawing board again. What are people’s concerns? So, more or less, we should involve everyone involved in this process. That could be the early adopters, but the critics should also be involved.,
[21:59] in case of doubt, we’ll ask again. Topics like sovereignty: what happens if systems no longer exist? So, all that needs to be considered and not just jump on this hype bandwagon. Daniel, thank you for this tip. Many thanks for the great conversation. I thank you for watching and listening to the talk. We’ll keep in touch, see you next time at accompio Fokusthema. Super.
AI initially reinforces a development that has already begun using agile methods and shorter iteration cycles. Requirements are becoming more flexible, software is developed more quickly, and it is delivered more frequently.
With generative AI, this speed increases significantly further. Developers can create prototypes faster and produce larger amounts of code. For quality engineering, this means that the speed of the current testing processes is increasingly insufficient.
Test pipelines that take one or two days are no longer suited to an environment where new software increments are created in a very short time.
„AI is only a tool for now.“
Daniel Horn, Principal IT Consultant at accompio
A central approach is therefore the so-called Shift LeftQuality assurance is moved as far forward in the development process as possible.
Instead of testing the finished software first, the question should be asked during the requirements analysis: What exactly should be tested? What use cases exist? Which edge cases need to be considered? And what quality requirements are there in the first place?
Daniel Horn also sees communication between the department responsible for development and testing as crucial. Quality requirements should not be created at the end of the development process, but should be defined together as early as possible.
This leads to a fundamental change: Testing is not just used to check finished software, but is an integral part of software development from the very beginning.
If software is developed and delivered more frequently, tests must also be able to be performed more quickly.
Test automation This is not a completely new concept. It has already helped to shorten testing cycles in the past. However, due to the additional speed of AI, it is becoming even more important.
It’s not just about automating individual test cases. It’s also Deployments, test environments, and the parallel execution of tests must be made more automated.
One example illustrates the challenge: When hundreds of pull requests with different software versions are to be tested simultaneously, one or two fixed test environments are no longer sufficient. Test systems must be able to be deployed dynamically and in parallel.
AI offers various possibilities in testing. For example, it can create test cases, analyze test results, prioritize errors, or assist with pattern recognition.
This can be particularly useful for large amounts of test results. Instead of manually evaluating hundreds of results, AI can help identify anomalies and focus on relevant errors.
Nevertheless, human evaluation remains important. AI should be used as a Tools and support to be understood – not as a substitute for the professional understanding of a tester.
For companies, this raises an important question: Where does AI actually bring added value?
Daniel Horn warns against using AI just because it is available or because the current hype suggests so. For some tasks, there are already classic automation solutions that can be faster, cheaper, or more reliable.
Therefore, every process should be considered critically: Where is AI useful? Where is classic automation sufficient? And in which areas is human expertise indispensable?
AI should not be seen solely in terms of individual work steps. It can be more useful to view it in terms of the entire process to analyze and discuss with the participating teams where AI can actually be used to support them.
According to Daniel Horn, a modern testing strategy must primarily accomplish one thing: Integrate quality early and continuously into the development process.
This includes many small tests at different levels – for example unit and integration tests – instead of solely extensive end-to-end tests at the end of the development process.
At the same time, tests, deployments, and test environments must be made more automated and parallelized. Only with comprehensive Test Management Quality engineering teams can keep up with development when potentially hundreds of changes are being made simultaneously.
„We need to move in more of the same direction: What can we actually do technically already, which was previously not at all necessary? We should strengthen this now.“
Daniel Horn, Principal IT Consultant at accompio

accompio helps companies adapt their testing strategy to the increasing pace of development.
AI fundamentally changes the speed of software development. This also increases the pressure on existing testing processes. However, the answer to this is not to hand over all testing tasks to AI. Companies must instead invest in their own testing infrastructure and resources. Further develop the entire testing strategy: Test earlier, automate more, parallelize processes, and consider quality requirements already during development.
AI can help with: creating test cases, analyzing results, or detecting patterns. The decision, where and how AI is used sensibly, However, remaining a central task of the people in quality engineering.

Daniel Horn is a Principal IT Consultant at accompio and helps companies adapt their testing strategy to the increasing development speed.
AI accelerates software development, leading to more software increments and shorter development cycles. Testing must therefore be made faster, more automated, and integrated earlier into the development process.
AI can support testers in tasks such as creating test cases, analyzing test results, and pattern recognition. However, the professional evaluation and the decision of which tests and quality requirements are relevant remain important tasks for humans.
Shifting to the left means moving quality assurance as early as possible into the development process. Tests and quality requirements are taken into account during the requirements analysis and development, rather than only being checked on the finished software at the end.
As software is developed and delivered more frequently, testing must also be conducted more quickly. Automated tests, deployments, and testing environments enable the simultaneous review of larger numbers of changes.
AI can create test cases, generate code for test automation, analyze test results, identify patterns of errors, and prioritize anomalies. The technical review and risk and approval decisions should continue to be handled by humans in quality engineering.
Continuous Testing integrates automated tests into the development process continuously. This makes it possible to identify errors more quickly and enables shorter delivery cycles. In order to achieve this, in addition to the tests, deployments and test environments must also be automated and made available in parallel when necessary.
A modern testing strategy starts early with clear quality requirements, unit and integration tests. The department, development, and quality engineering work closely together; recurring tests, deployments, and environments are automated, and human experts retain the authority to make risk and release decisions.
