Selenium Conference 2026: AI, BiDi, and the Future of Test Automation

From May 6 to 8, the Selenium and Appium Conference took place in Valencia, bringing together the global automation community for three days of technical discussions and knowledge sharing. Built by the community and for the community, the event welcomed speakers from 23 countries across five continents — with workshops, talks, a panel session, and lightning presentations all pointing toward where browser and mobile automation is heading next.

AI influence

The conference opened with a keynote by Baris Sarialioglu, who presented some striking numbers about AI adoption across automation and testing teams. According to the data, approximately 55% of QA teams already use AI-powered testing tools, around 70% of testers actively use generative AI for test creation and automation tasks, and test pass rates have increased from roughly 55% to around 70%.

Despite this progress, he pointed out that many teams still struggle with managing uncertainty, gaining deeper system understanding, and building confidence when working with AI-assisted processes. This challenge becomes even more important as software evolves from predictable systems into adaptive ecosystems — where AI introduces probabilistic behaviour and non-deterministic outcomes that require a fundamentally different testing mindset.

One particularly memorable part of the keynote was a comparison between concepts from quantum physics and what he called “quantum automation”:

• Superposition: the current state of a system cannot be fully known without measuring it.
• Entanglement: services, AI models, environments, and observability layers form interconnected ecosystems where changes propagate across components.
• Measurement problem: measuring and repetitively testing an AI system may influence its behaviour.
• Zeno effect: excessive observation or intervention may affect system evolution.

He finished with a quote from Professor Brand in Interstellar:

We’re still pioneers. We’ve barely begun. Our greatest accomplishments cannot be behind us, because our destiny lies above us.

He used this to draw a parallel with the early days of automation — even after decades of evolution, the industry is still exploring new techniques, challenges, and opportunities.

The AI discussion continued in other sessions. During “What AI”, Noemi Ferrera focused on sharing practical advice for finding the right AI tools and using them effectively. She encouraged attendees to experiment with multiple options and emphasised that AI should act as a force multiplier rather than a replacement for engineers. She also stressed the importance of defining clear rules and expectations, and reminded the audience to always consider the potential limitations and failures of AI systems while maintaining human oversight.

On the last day, the keynote from Sofia Palamarchuk expanded this discussion by questioning whether AI is truly helping teams understand systems better and make better decisions. Her proposal was particularly interesting: instead of using AI only to generate large amounts of data, teams could use AI agents to identify knowledge gaps, explain features, analyse impacts, summarise evidence, support documentation, and assist testing activities. She also argued that AI adoption should be treated as a managed capability — with structured adoption programs, clear metrics to measure impact on quality and delivery, and a responsible approach that keeps human judgment at the centre.

Selenium BiDi

While AI dominated many conversations, Selenium BiDi quickly emerged as another key theme. It featured in multiple talks, with one dedicated session led by Maksim Sadym, who explored the current state of WebDriver BiDi and how it’s evolving within browser automation.

He started with a brief history of automation technologies, comparing WebDriver Classic, Chrome DevTools Protocol (CDP), and WebDriver BiDi. The comparison showed how WebDriver Classic mainly operates using a synchronous, unidirectional request-response model, while CDP offers deeper browser integration but remains browser-specific. WebDriver BiDi attempts to combine bidirectional communication with cross-browser standardisation.

One practical example showed how teams are already using BiDi capabilities for request interception and response mocking — helping overcome challenges related to unavailable test data, dependencies on external systems, simulating API failures, or validating frontend behaviour under specific responses.

Scalable test architecture

While BiDi explored the future of browser capabilities, Sargis Sargsyan focused his talk on improving the automation architecture itself. He opened with a painfully familiar problem: the classic flaky test that runs perfectly on your local machine but fails completely in CI. To tackle this, he suggested combining the Page Object Pattern with Loadable Components to guarantee that pages and components are fully rendered before any test interactions happen. The Selenium Wiki article is a good starting point, and Sargis also has code examples in his repository.

He also introduced applying Domain-Driven Design (DDD) principles to automation projects, moving away from organising frameworks purely around UI elements. A few practical tips he shared:

• Organise Page Objects by business domain, for example: checkout/, auth/, and similar structures.
• Use business terminology in test code — names like CartItem and OrderSummary make intent clearer than generic names like Component1 or input_1.
• Use Factories and Builders to create readable and scalable test data, for example: UserBuilder.aGuest().withEmail("x").

Beyond DDD, Sargsyan shared general best practices for keeping test suites healthy:

• Drop hardcoded waits: avoid Thread.sleep() and rely on the advanced waiting mechanisms built into recent Selenium versions.
• Avoid “God Page Objects”: don’t let a single page object accumulate too many responsibilities, or it will quickly become a nightmare to maintain.
• Keep assertions out of Page Objects: assertions belong to the test layer — this keeps page objects clean and focused, and even Martin Fowler has written about this.
• Treat automation like production code: building a maintainable framework means applying software design patterns, requiring pull request reviews, and enforcing strict engineering standards.

Final remarks

What stood out most about this conference was the balance between in-depth technical content and open discussions about real engineering challenges. While Selenium and Appium are at the centre of the event, the conversations extended well beyond frameworks and tools — exploring how automation, AI, accessibility, quality engineering, and modern software delivery practices are becoming increasingly interconnected.

The venue itself added to the experience. The conference took place at Veles e Vents in Valencia’s marina, which provided a great setting for both sessions and networking. Even the coffee breaks reflected local culture, with traditional Spanish food such as paella and croquettes adding a welcoming atmosphere.

All sessions were recorded and made available on the Selenium Conference YouTube channel, allowing anyone who couldn’t attend to access the content and giving attendees the chance to revisit presentations afterward.

Overall, it was a very positive experience — an event I would gladly recommend to anyone interested in automation, and one I would be happy to attend again in the future.