Gothenburg-based drone software developer Airpelago has redefined its quality assurance workflow by adopting AI-driven agentic testing to transform automated test failures into actionable product telemetry. The Swedish company, which develops autonomous flight and traffic management systems, previously relied on Playwright for end-to-end testing but encountered persistent flakiness. Traditional selector-based scripts failed to interact reliably with the platform’s core interactive mapping interface, rendering automated validation impractical and producing build notifications that offered little diagnostic value. In June 2025, Airpelago integrated QA.tech, a goal-driven testing platform that replaces rigid scripts with adaptive AI agents. Rather than following predefined DOM paths, these agents interpret on-screen elements and execute user journeys based on natural language objectives. This approach successfully automated the testing of complex map-based interfaces that previously resisted standard frameworks. The team defined twenty primary operational workflows and deployed the agents to validate them in a live browser environment. The initiative implements friction as telemetry by distinguishing between infrastructure noise and genuine product friction. Airpelago CTO Tobias Fridén reported that the shift eliminated the hesitation that previously accompanied deployments. The platform runs multiple times daily, integrating with GitHub pull requests via keyword triggers to validate preview environments without blocking routine commits. Failed tests generate narrated reports containing screenshots, network traces, and agent reasoning, effectively automating bug reproduction. Analysis of agent failures categorizes friction into four engineering priorities. Navigation friction indicates broken or obscured user flows requiring architectural adjustments. Labeling friction highlights ambiguous interface elements warranting copy revisions. State friction exposes timing or loading discrepancies that traditional suites typically mask with hardcoded waits. Data friction identifies mismatches between user actions and system responses, a critical issue for applications handling sensitive visual data. Tagging failures by category enables precise routing of defects to appropriate teams. The telemetry model also establishes a direct feedback loop with AI coding assistants. Friction reports function as optimized prompts for tools like Claude Code and Codex. Platform integrations allow development agents to query test results, trigger reruns, and apply fixes directly within the coding environment before commits. This tightens the development cycle by resolving defects while developer context remains active. Airpelago’s deployment demonstrates how adaptive AI testing secures previously unautomatable product surfaces while converting continuous integration failures into improvement signals. As agentic validation systems mature, treating automated testing difficulty as early-warning telemetry is expected to become standard across organizations managing complex visual interfaces.