Call for submissions: Platform Engineering meet-ups
Rootconf
Rootconf

Call for submissions: Platform Engineering meet-ups

Share your journey of building platforms that power engineering teams

All submissions

Rana Roy

Beyond Static Test Scripts: An Agentic Approach to Adaptive yet Reliable Android Testing

Submitted Aug 15, 2026

Session type: Talk (30 mins) Submission made for: Bangalore meet-ups

What if the test suite became the bottleneck to shipping software - not because the application was broken, but because the tests could no longer keep up with it?

Modern Android applications can change significantly from sprint to sprint, while their end-to-end tests often remain tightly coupled to implementation details: selectors, UI hierarchies, navigation paths, timing assumptions, and exact assertions. Every UI evolution can introduce test-code churn, stale automation, and growing test technical debt. In an Agile environment, this creates a fundamental mismatch: application development can move at the speed of continuous delivery, while test automation must repeatedly be rewritten to catch up.

This session presents an intent-driven, agentic approach to mobile test automation that shifts the focus from how the current UI works to what the test is intended to validate. An LLM agent can interpret natural-language test skills, establish application state, interact with evolving UIs, reason about semantic outcomes, and adapt its execution when the application changes. But simply replacing deterministic automation with an LLM introduces a new problem: probabilistic execution is difficult to trust, reproduce, diagnose, and scale.

The framework therefore combines agentic reasoning with deterministic engineering controls. Deterministic setup and teardown enforce preconditions and postconditions; complete execution traces provide observability and evidence; adversarial verification scrutinizes agent decisions; bounded self-healing handles recoverable UI and state drift; and structured failure classification distinguishes product defects from environmental, infrastructure, precondition, and agent failures. Once an agent successfully executes and validates a test, the execution is converted into a deterministic replayable artifact where the workflow admits a stable, reproducible representation. Tests that inherently depend on dynamic reasoning remain agentic, while deterministic portions can be replayed mechanically. This allows subsequent CI/CD runs to avoid unnecessary LLM invocations, while preserving agentic execution where determinism is not appropriate. The agent is brought back into the loop only when deterministic replay detects genuine drift or when the test requires adaptive reasoning.

The result is a complete reason → validate → determinize where possible → replay → heal lifecycle, orchestrated through versioned test skills, Jenkins-based CI/CD, and concurrent MCP-backed Android execution environments. The goal is not to replace deterministic automation with AI, but to use AI where reasoning is required and determinism where repeatability matters.

Key Takeaways

  • Replace implementation-coupled tests with intent-driven automation: Define what needs to be validated and let an agent determine how to interact with the current application.
  • Reduce test technical debt and test-code churn: Decouple stable test intent from continuously evolving UI implementations.
  • Make agentic testing trustworthy: Combine complete observability, execution evidence, adversarial verification, bounded recovery loops, and meaningful failure classification.
  • Separate reasoning from execution: Use LLMs for semantic understanding, exploration, validation, and healing while relying on deterministic automation wherever the workflow can be reliably reproduced.
  • Reason once, replay many times where possible: Convert successfully validated executions into deterministic replayable artifacts when a stable representation can be derived, while retaining agentic execution for inherently dynamic test scenarios.
  • Scale agentic testing through CI/CD: Version test skills, orchestrate execution with Jenkins, and run tests concurrently across multiple Android devices through MCP-backed execution environments.

Who Will Benefit?

This session will be valuable for:

  • Android and mobile engineers dealing with rapidly evolving application UIs and growing test-code churn.
  • QA and SDET teams looking to reduce test authoring, repair, and regression-validation effort.
  • Platform and DevOps engineers building scalable CI/CD infrastructure for mobile testing.
  • Test automation architects exploring LLM-driven, intent-based, self-healing, and deterministic testing architectures.
  • AI and agentic-systems engineers interested in building production-grade harnesses around LLM agents, including bounded execution, state management, observability, verification and recovery loops, and deterministic controls around probabilistic workflows.
  • Engineering leaders evaluating how agentic AI can be introduced into software delivery without sacrificing reliability, reproducibility, observability, or operational control.

Speaker Bio

Rana Roy is an AI Engineer at Sasken Technologies Ltd., focused on using AI and agentic systems to automate engineering workflows. His work involves building AI-driven automation frameworks that enable developer and QA teams to automate complex workflows, reduce manual effort, and improve engineering productivity. He is currently working on applying agentic AI to mobile testing, enabling intelligent test execution, validation, failure diagnosis, and self-healing within CI/CD environments.

All submissions

Comments

{{ gettext('Login to leave a comment') }}

{{ gettext('Post a comment…') }}
{{ gettext('New comment') }}
{{ formTitle }}

{{ errorMsg }}

{{ gettext('No comments posted yet') }}

Register

Hosted by

Rootconf
Rootconf
We care about site reliability, cloud costs, security and data privacy