--- title: Replay: Faster Mobile QA for AI Agents with agent-device description: agent-device v0.20 introduces a replay system that cuts repeated mobile QA flows from minutes to seconds by eliminating model calls for known routes. A... canonical: https://daily.dev/posts/replay-faster-mobile-qa-for-ai-agents-with-agent-device-f1em81tdh twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Replay: Faster Mobile QA for AI Agents with agent-device | daily.dev og:description: agent-device v0.20 introduces a replay system that cuts repeated mobile QA flows from minutes to seconds by eliminating model calls for known routes. A... og:url: https://daily.dev/posts/replay-faster-mobile-qa-for-ai-agents-with-agent-device-f1em81tdh og:image: https://api.daily.dev/og/posts/F1EM81TdH.png og:image:alt: Replay: Faster Mobile QA for AI Agents with agent-device og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Replay: Faster Mobile QA for AI Agents with agent-device **[Callstack Blog](https://daily.dev/sources/callstack-blog)** · 20 min read · 1 upvotes · 0 comments ## Summary agent-device v0.20 introduces a replay system that cuts repeated mobile QA flows from minutes to seconds by eliminating model calls for known routes. A 3-minute 18-second Claude-driven Android task replayed in 8.8 seconds with zero model calls. The post covers experiments with the --settle flag (reducing model turns by 25% and cost variance significantly), replay script mechanics with identity-binding checks, and a 36-trial repair study. Key finding: 22 of 36 repair trials passed but only 10 preserved identity evidence intact. A hybrid approach—replaying the known route then handing off to the agent—reduced median task time by 42% and cost by 46%. The author concludes that re-recording a flow can be cheaper than repairing it, and that repair quality should be measured by evidence preservation, not just pass/fail status. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How much faster is replaying a recorded agent-device flow compared to letting an AI agent redo the whole mobile QA task from scratch? A nine-action Android Settings QA flow that took 3 minutes 18 seconds when driven live by Claude Sonnet via agent-device finished in 8.8 seconds, including app launch, when replayed as a saved script with zero model calls. Across 20 repeated replay runs, the median was 8.0 seconds with a range of 7.2 to 8.5 seconds. _Anyone benchmarking AI agent QA speed can track tools like agent-device on daily.dev._ ### What does the --settle flag do in agent-device and does it actually help? Introduced in agent-device v0.18, --settle makes the CLI wait for the UI to stop changing before returning a snapshot, avoiding half-transitioned screens that trigger extra model observations. In tests on React Navigation's example app, it cut model turns by 25% and tokens by 22% for a Haiku-class model, and reduced cost variance for a Sonnet-class model from a 1.8x spread to a 1.1x spread; it is now recommended by default. _Developers tuning agent-driven UI automation can follow tool updates like this via daily.dev._ ### Can an AI agent reliably self-heal a broken end-to-end test script without breaking test validity? Not reliably: in a 36-trial repair study on agent-device using Haiku and Sonnet models across three UI drift types, 22 of 36 repair attempts produced a passing script, but only 10 preserved the original identity evidence intact, meaning most passing repairs were actually invalid edits that a naive pass/fail check would have missed. Re-recording the flow from scratch was cheaper in the one clean comparison available. _Teams weighing self-healing test claims can follow real repair benchmarks like this on daily.dev._ ## Similar posts on daily.dev - [How We Optimized Agent Device for Mobile App Automation](https://daily.dev/posts/how-we-optimized-agent-device-for-mobile-app-automation-creb13roy) · Callstack Blog · 1 upvotes · 0 comments - [Your Agent Aced the Task. Will It Do It Again?](https://daily.dev/posts/your-agent-aced-the-task-will-it-do-it-again--yetkhtezb) · Hugging Face · 0 upvotes · 0 comments - [Teaching Claude to QA a Mobile App](https://daily.dev/posts/teaching-claude-to-qa-a-mobile-app-43pxxdxv9) · Hacker News · 0 upvotes · 0 comments --- Tags: [#testing](https://daily.dev/tags/testing), [#ai-agents](https://daily.dev/tags/ai-agents) [View this post on daily.dev](https://daily.dev/posts/replay-faster-mobile-qa-for-ai-agents-with-agent-device-f1em81tdh) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Replay: Faster Mobile QA for AI Agents with agent-device","url":"https://daily.dev/posts/replay-faster-mobile-qa-for-ai-agents-with-agent-device-f1em81tdh","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/replay-faster-mobile-qa-for-ai-agents-with-agent-device-f1em81tdh"},"datePublished":"2026-07-31T07:09:42.969Z","dateModified":"2026-09-13T21:20:33.359Z","description":"agent-device v0.20 introduces a replay system that cuts repeated mobile QA flows from minutes to seconds by eliminating model calls for known routes. A...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/0d1ae09b55d5a0a1fda82d11724d8971?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/0d1ae09b55d5a0a1fda82d11724d8971?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Callstack Blog","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Callstack Blog","logo":"https://media.daily.dev/image/upload/s--3qItoroi--/f_auto,q_auto/v1774179048/logos/callstack-blog?_a=BAMAMiWQ0","url":"https://daily.dev/sources/callstack-blog"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/replay-faster-mobile-qa-for-ai-agents-with-agent-device-f1em81tdh","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"testing,ai-agents","timeRequired":"PT20M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Callstack Blog","item":"https://daily.dev/sources/callstack-blog"},{"@type":"ListItem","position":3,"name":"Replay: Faster Mobile QA for AI Agents with agent-device"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/replay-faster-mobile-qa-for-ai-agents-with-agent-device-f1em81tdh#faq","mainEntity":[{"@type":"Question","name":"How much faster is replaying a recorded agent-device flow compared to letting an AI agent redo the whole mobile QA task from scratch?","acceptedAnswer":{"@type":"Answer","text":"A nine-action Android Settings QA flow that took 3 minutes 18 seconds when driven live by Claude Sonnet via agent-device finished in 8.8 seconds, including app launch, when replayed as a saved script with zero model calls. Across 20 repeated replay runs, the median was 8.0 seconds with a range of 7.2 to 8.5 seconds. Anyone benchmarking AI agent QA speed can track tools like agent-device on daily.dev."}},{"@type":"Question","name":"What does the --settle flag do in agent-device and does it actually help?","acceptedAnswer":{"@type":"Answer","text":"Introduced in agent-device v0.18, --settle makes the CLI wait for the UI to stop changing before returning a snapshot, avoiding half-transitioned screens that trigger extra model observations. In tests on React Navigation's example app, it cut model turns by 25% and tokens by 22% for a Haiku-class model, and reduced cost variance for a Sonnet-class model from a 1.8x spread to a 1.1x spread; it is now recommended by default. Developers tuning agent-driven UI automation can follow tool updates like this via daily.dev."}},{"@type":"Question","name":"Can an AI agent reliably self-heal a broken end-to-end test script without breaking test validity?","acceptedAnswer":{"@type":"Answer","text":"Not reliably: in a 36-trial repair study on agent-device using Haiku and Sonnet models across three UI drift types, 22 of 36 repair attempts produced a passing script, but only 10 preserved the original identity evidence intact, meaning most passing repairs were actually invalid edits that a naive pass/fail check would have missed. Re-recording the flow from scratch was cheaper in the one clean comparison available. Teams weighing self-healing test claims can follow real repair benchmarks like this on daily.dev."}}]} ```