UXMA @ WeAre­Devel­opers World Congress 2026

Birte: Above all, I wanted to come away with ideas that I could apply directly to our projects. Of course, I knew that AI would be the dominant topic. But I was also hoping to discover case studies and some presentations that didn’t rely on the buzzword “AI.”

Milan: My goal was to understand the overall direction of software development, as well as topics like security, accessibility, new web technologies, and advanced tools that make day-to-day work easier.

Thoren: I arrived without any major expectations and was confident that I’d definitely find something of interest among the offerings. My hopes were to gain interesting technical insights into AI, learn about specific projects and tools, get a glimpse of what the major players are currently doing, and gain practical experience at the architec­tural level.

Milan: My concern was that, with so many options and so little time, I wouldn’t be able to prioritize properly and would end up in the wrong talk.

Thoren: I didn’t really have any concerns before the conference. It wasn’t until I got there that I realized how difficult it is to choose the right sessions at an event of this size.

Birte: Many of the talks actually confirmed my initial fears: A brief introduction was usually followed by a pitch for a new AI tool that would solve all our problems. Still, there were a few presentations with valuable insights on database architecture, performance optimization, and current software trends that have stuck with me.

Milan: I was particularly impressed by the scale of the event. The expo, the international attendees, and the wide range of topics show just how diverse the developer community has become. Good on-site planning is essential to ensure you can attend the talks you want to hear – and not just be presented with “shiny new AI products.”

Thoren: It quickly became clear to me: I have to make a choice. Talks, the Expo, or networking – you can’t do everything at once. Either I go from talk to talk or I deliberately set aside time for the exhibition booths and conversations. Plus, masterclasses and workshops fill up quickly. Many panels and presentations, at a maximum of 30 minutes, were too short to go into depth, and talks about specific problems ultimately ended up showcasing a product that helps solve the problem.

Birte: I specifically looked for ideas that we at UXMA could implement right away—best practices, refactorings, and pragmatic solutions. Something I could apply directly at work on the Monday after the conference. Because after a year in front-end development, my preliminary conclusion is this: AI solves a surprising number of problems quickly and reliably, especially functional bugs. But there’s often a snag with the UI, because AI is blind. It doesn’t see what’s happening on the screen, only what’s written as text in the code. What looks logically structured there might, under certain circumstances, appear cluttered or unintuitive in the UI. To address this issue, I chose a talk that I hoped would bridge that gap.

Thoren: I was interested in AI-powered workflows that I could apply directly to my daily project work, as well as insights into how other companies are aligning their architecture with AI. An interesting start in that regard was the main stage talk “The AI-Native Engineering Org: What’s Real, What’s Hype, What’s Next” by Taroon Mandhana (CTO of AI and Teamwork at Atlassian). They’re building a global AI context layer for the entire company—a central knowledge layer that their AI systems access. At UXMA, we’re naturally focusing on exactly these topics right now—from a shared context for AI tools and general collaboration with AI agents to testing with local LLMs. When one of the major players on stage describes what we’re already doing on a smaller scale, it’s a good sign that we’re on the right track.

Birte: My personal highlight was the talk “Design System for the Machines – How to Make AI Understand Your UI” by Jennifer Wjertzoch, Senior Frontend Engineer at DKB Code Factory. The core idea behind it is that a design system is no longer aimed solely at humans. Machines also “read” our UI – for example, our AI assistant, which understands, expands, and modifies our code. But unlike us humans, machines don’t just read text—they read building blocks and their meaning within the UI. And in doing so, they regularly miss three things:

• Missing Meaning: A < div class="button" > is, to the AI, just a div at first. The fact that it’s supposed to be a button isn’t apparent from the CSS class. The meaning is missing.

• Missing Behavior: What does the element do? Open, toggle, submit? If it’s not specified anywhere, one can only interpret. The function is missing.

• Missing Intent: Why is this here, and what is the user supposed to do with it? The intent is missing.

At this point, I was reminded of the last Accessibility Meetup in Kiel. The gap that AI stumbles over is the same one that a blind or visually impaired person stumbles over – and that’s why screen readers need information. So the solution presented in this talk wasn’t a pitch for a new tool, but rather accessibility through semantic HTML (< button >, < nav >, < main > instead of a “div jungle”) and clean ARIA labels and roles when semantics aren’t sufficient. This makes meaning, behavior, and intent machine-readable – for both AI and people with visual impairments. This is what makes it accessible.

Milan: My highlight was the talk titled “Are Classical Automation Frameworks Dead? How AI Agents Are Transforming QA” by Andrei Nutas, Technical Test Architect at Atos. It addressed several issues we face in our day-to-day work: UI tests that are “flaky” – meaning they sometimes pass and sometimes fail – as well as the “selector maintenance tax,” the constant effort required to maintain tests when selectors in the user interface change. The talk presented three approaches: full agentic testing, a hybrid model, and the “classic” manual setup of tests. I wasn’t surprised that a hybrid testing model was recommended and that the speaker cautioned against a testing system that always runs with AI agents and burns tokens with every test run. In the hybrid testing model, it was demonstrated how classic frameworks like “Playwright” or “Skyvern” can be used to write self-healing tests. If a test fails, developers don’t have to look directly at the logs to find and fix errors that are often trivial but hard to detect. Instead, in this specific case, an AI agent steps in, resolves the issue, adjusts the test accordingly (e.g., via an editable database available in the pipeline), and no longer needs to intervene during the next run. This process is simple enough that it could even run on local models.

Thoren: What really stuck with me was the talk titled “You don’t need to write the code. You need to become a verification architect and prove it’s correct” by Guillaume Moigneu, Field CTO at Upsun. The central thesis: The “10x developer” of the agent era isn’t someone who writes ten times more code, but someone who defines, verifies, and delivers ten times more scope. And verification goes from being a roadblock to a speed multiplier, because the more rigorous the verification mechanisms, the more you can let the agents run unsupervised.

To address this, Moigneu distinguishes between two questions that often get conflated in everyday practice: Verification (“Did we build it correctly?”) – such as tests, types, and linting – is a task for machines and, ideally, fully automated. Validation (“Did we build the right thing?”) involves the machine checking against a documented intent. We are responsible for building the verification pipeline so that, in the end, the human-in-the-loop (HITL) is ideally only responsible for validation. The talk’s guiding principle sums it up: “If it’s not proven, it doesn’t exist.”

What I liked about the talk: It provides a good overview of the state of the art and covers experiences and challenges that I’ve also encountered with AI agents, while explaining a structured, reproducible workflow from commit hooks to critic prompts. Some of these approaches are already part of my daily work, and that’s exactly how I can tell that this talk was more than just a good half-hour presentation.

Birte: In addition to the new AI tools, I gained new insights into my own small, everyday topics. What I’m taking away from my highlight talk is that our Accessibility Pod is the right place to turn exactly that into a shared standard. Because sometimes you can use trends to do something that’s always been the right thing to do and finally give it the priority it deserves.

Milan: My impression is that no one is debating anymore whether AI makes sense for software development or not. Instead, the focus is on how it can efficiently support us as a tool. The extent to which AI will be used is still wide open in most areas, but it represents a huge opportunity. If you have specific interests, it’s worth attending workshops and, when it comes to the talks, not just reading the title and description, but also considering whether the topic truly aligns with your own interests.

Thoren: WeAreDevelopers offers a wide range of options. The conference is particularly worthwhile for students, junior developers, and technical managers, while experienced developers should opt for more specialized formats if they’re looking for in-depth technical knowledge. Above all, I came away with a clearer picture of the evolving role of developers and confirmation that UXMA is on the right track when it comes to AI.

Despite our different areas of focus, we all came to a similar conclusion: the most exciting insights emerged when practical experiences were shared. WeAreDevelopers therefore didn’t so much provide us with new answers as it confirmed that we at UXMA are already working on the right questions and can now develop them further in a targeted manner—just as we can our own skills.