Craft Article #2: AI and Software Quality

I wrote this article a year ago in 2025, following ByteDance's announcement of UI-TARS — an AI solution capable of interpreting and interacting with interface behavior — and after the recent announcements around OpenAI Operator and Perplexity Assistant, I realized there was no longer any excuse for not implementing automated UI tests. But today, in 2026, with the proliferation of agentic solutions and the release of agent-browser by Vercel, it was clear the time had come to publish it.

Today, thanks to these AI-powered tools — and all the others released week after week over recent months — the need for validation and acceptance testing is better covered than ever: not only to verify that expected features meet their specifications, but also to evaluate performance and, in part, accessibility.

In this article, I want to share my perspective on the contribution of AI to testing and software quality: how these new solutions are transforming our practices, and how they can help us better address the full spectrum of testing and quality assurance requirements.

The Rise of AI in Test Automation: A New Golden Age for E2E, Usability and Accessibility

At the heart of software development processes, tests guarantee not only the reliability of an application, but also the user experience and accessibility. In recent years, the emergence of increasingly powerful AI models has opened the door to new approaches: generating meaningful specifications through interaction and the systematic addition of acceptance criteria, generating code from specifications, intelligent End-to-End (E2E) test automation, proactive detection of usability (UX) issues, and reinforcement of accessibility best practices.

We will explore how these developments are transforming the way product teams design, execute and maintain their tests. We will also see why AI is on track to become a major lever for delivering more inclusive, friction-free user journeys.

1. AI at the Heart of End-to-End Testing

Automatic Generation of Test Scenarios

AI models are now capable of dissecting an application's architecture (web, mobile, etc.) to automatically generate E2E test scenarios. Instead of spending hours coding and maintaining scripts, teams can:

• Quickly identify critical user journeys: authentication, payment, key forms, etc., through log analysis or tracking tools.
• Detect interface variations: AI spots design or interactive element changes and adjusts tests accordingly.

Test Maintenance and Robustness

One of the most common problems with E2E tests is fragility: the slightest code change can "break" tests. AI solutions continuously learn from the application's behavior and automatically adapt to changes, which:

• Significantly reduces maintenance time
• Improves the reliability of test campaigns, even when the application is evolving rapidly.

2. Usability and UX: Automation "Beyond Functional"

User Behavior Analysis

Tests are no longer limited to "works/doesn't work" validation. Thanks to AI, it is now possible to:

• Analyze navigation to identify friction points or unnecessary steps
• Detect unexpected user journeys that reveal a need for design optimization

These analyses provide a better understanding of the application's ergonomics and make it possible to adjust the user experience accordingly.

Assisted Exploratory Testing

AI can be used to conduct semi-automated exploratory testing — for example, by subjecting the application to varied, often unexpected scenarios — in order to:

• Detect hidden bugs earlier
• Gather UX feedback (navigation issues, blind spots, lack of intuitiveness)

This approach enriches traditional testing while freeing up time for teams to focus on analysis and implementing fixes.

3. Accessibility: AI in the Service of Inclusion

Automatic Detection of Non-Compliance

Accessibility standards (WCAG, RGAA, etc.) define a set of rules to ensure that digital content is usable by everyone, including people with disabilities. AI models can now detect:

• Contrast errors, missing or redundant heading hierarchies
• Incorrect or missing alternative tags (alt) on images
• Structural issues making navigation complex for screen reader users

This rapid diagnosis reduces the risk of missing critical accessibility points.

Generation of Compliant Content

Beyond detection, some AI tools can automatically suggest or generate compliant content. This includes:

• Generating alt tags for images through image recognition
• Automatic captioning for video or audio content

By resolving these issues upstream, teams greatly improve the experience for visually impaired, hearing-impaired, or otherwise differently-abled users.

4. Perspectives and Challenges

Skill Development and Governance

Adopting AI in the testing cycle requires new areas of expertise: data science, MLOps, and ethical auditing. Companies must plan for:

• Targeted training for QA, development and product teams
• Validation processes to avoid algorithmic bias and protect user privacy

Transparency and Ethics

The more central AI becomes to processes, the more essential it is to ensure transparency:

• Respect for personal data: AI often analyzes browsing logs or sensitive usage data
• Fairness and inclusion: ensuring that models do not reproduce discrimination in their detection or suggested fixes

5. Conclusion: Toward a Future Where AI Elevates Software Quality

Recent advances in AI are not simply "boosting" E2E tests — they are redefining how the overall software experience is conceived. By automatically detecting bugs, adjusting UX based on behavioral data, and ensuring accessibility from the design stage, AI contributes to building products that are more reliable, more inclusive, and more enjoyable to use.

For many teams, adopting these tools will represent a genuine qualitative leap — provided that questions of ethics, data protection and skills are properly anticipated. In short, AI is establishing itself as a strategic ally in software quality, from testing to user experience, while laying the foundations for sustainable and responsible innovation.