Intel / article

DragonCrawl: A Generative, Intent-Based Framework for Scalable Mobile End-to-End Testing

Source: arxiv.org
Author: Sowjanya Puligadda, Mengdie Zhang, Ali Zamani, Dhruva Dixith Kurra, Eric Chen, Juan Marcano
Date: Published Aug 3, 2026

Context

DragonCrawl uses AI to check whether mobile-app user flows still work after code changes. The authors' production account describes a system using OpenAI's GPT-4o to reason about intended actions, inspect the screen for the expected final state, and call tools that change backend state. This connects interface interactions with the underlying system during regression tests. Across 1,013 automated tests, the authors report pass rates of 91.6% on iOS and 92.2% on Android. Those figures describe test outcomes in the deployment; they do not measure the proportion of real bugs detected or establish that test failures from automation instability have disappeared.

Key takeaways · AI-distilled

DragonCrawl targets regression, not exploration: it validates specific user flows on every code change and blocks commits that break critical functionality - 91.6% pass on iOS and 92.2% on Android across 1,013 tests in CI.
The economics are the headline: test onboarding drops from 96-120 hours to under 4, with an estimated 27 developer-years saved in maintenance.
Architecture evolved from V1 -similarity matching to V2 generative intent-based reasoning with GPT-4o - vision handles end-state detection while tool calls bridge backend state transitions.
Production lessons include fighting explosion and memory constraints - the operational half most -testing papers skip.

Terms in this piece · Glossary

embedding — A list of numbers representing a piece of text's meaning, so that similar meanings end up numerically close and can be searched.
multimodal — A model that works with more than text — reading images, audio, or video, and sometimes generating them too.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.

Key quotes

“DragonCrawl validates specific user flows on every code change, blocking commits that break critical functionality.”

“The system reduces test onboarding time from 96-120 hours to under 4 hours and has saved an estimated 27 developer years in test maintenance effort.”

Read the source arxiv.org