UiPath Accelerates AI in Software Development and Testing

Software development is in the midst of a major revolution. AI is changing every stage of the software development lifecycle, driven by the insatiable desire to build new applications faster than ever before, at a higher quality for lower costs. The days of having to choose between fast, good and cheap are disappearing with all three objectives now within reach.

Gerd Weishaar, the General Manager and SVP of Product Management for UiPath's Test Cloud, has been involved in software development for over two decades. He says the pace of change is so quick that any report about what is happening is out of date in just a few weeks.

"Most reports suggest that between 70% and 80% of developers are using AI tools to help them code. But the picture is quite different when it comes to quality assurance with adoption of AI sitting well under 50%. I'm a little bit surprised that quality assurance is not picking up much faster because how would you make sure that this code is safe, that it's working in an enterprise context."

While the numbers are quite stark, adapting to this new generation of software development requires more than just tools and software licenses. Gerd says what's needed is a deep transformation that takes a development team from a legacy process to a future process.

Gerd created UiPath's test offering six and a half years ago. When the company started investing in AI capabilities a little more than two years ago it completely shifted the roadmap.

"It was obvious that Gen AI would have an impact on how software was created and tested. We started out augmenting and supporting every single step that a tester does and provide some Gen AI capabilities with Autopilot for Tester. Phase two gave our customers the ability to build agents themselves."

UiPath recently held a hackathon and had 237 different agent ideas submitted. Gerd says the sheer volume of agents that were submitted was astounding. Among the ideas that came were agents that could detect flaky code. Another looked into GIT repositories for commit nodes to derive changes that may impact requirements or tests. Others analysed execution log files to find poorly designed tests

The next stage of UiPath's Autopilot for Tester will use multiple agents and orchestrate them to fulfill more complex processes. Testing will become more autonomous with a person sitting over the process.

"We often say that in AI we need a human in the loop. In this instance, it will be human on the loop with full autonomous agents doing the work. And the human will, at the end say, that is right, this works and will give the release stamp at the end. That's my expectation," says Gerd.

Where does Australia sit when it comes to the use of AI in software development? Gerd travels extensively and says we are in a good position. His observation is that the United States tends to be a very fast adopter while Europe tends to be much more cautious. Australia, he says, sits in the middle with local developers taking a very pragmatic approach.

"AI might be able to create code fast but that doesn't mean developers don't retain responsibility," says Gerd. "Developers now have these power tools. I think it will bring a specific challenge to developers. Seeing huge or large amounts of code being generated in front of you and being responsible that this code does exactly what it's supposed to do is a big burden. I don't think it's easily doable, to be honest."

This is why testing needs an AI infusion. With code being generated faster than ever, quality assurance processes need to adapt. Tools such as Claude Code can produce code that used to take a person days to create in minutes. Traditional approaches to testing simply can't keep up.

Software testing has not changed or innovated substantially for most of the last three decades in Gerd's view. He observes that the level of automation most organisations have achieved is somewhere around 20% to 30% at best. That means still 70% of the testing is done manually.

One of the reasons for this is that building automation has a cost that is not paid back unless it's used several times. But if a piece of software is altered then the automation needs to be updated. That means the cost of automation could not be recouped.

"What AI does, I think, is a significant change here," explains Gerd. "It lowers the cost of creation, and it lowers the amount of maintenance because we have true self-healing now. You can build automation and easily maintain it."

Over time, Gerd imagines a world where software is created and tested using AI agents. He believes that we will have autonomous exploration. AI will explore an application and, while it's exploring it, create tests on the fly. It will document those tests and tell you if they passed or failed. And once it's finished, it will create a list of test cases that can then run as a regression test.

This will lead to some significant change. Organisations will need to assess which people will be able to adapt to this new world.

"You will find people that have already solid skills in some areas and just need to add or augment a few more skills. And then you will find that there are people that are much slower on the skill range. This is where the tough part comes in for organisations because they will have to decide who are the people that we can transform into the future. I do not believe that everyone is going to make it. It may sound harsh. But it's like with every industrial revolution, I think some people will upskill and will have a better work life. And some will be caught in that transition," says Gerd.

The world of software development and testing is facing a revolution. Automated code creation is driving a move to AI-powered testing and that will lead to a need for new skills to be developed. While humans will remain an important part of the software development and quality assurance process, their role will become supervisory. But with that will come new responsibly to ensure that AI is delivering secure code that meets the organisation's needs.