MobileBoost

The mobile test automation market has a maintenance problem more than an authoring problem. Writing an Appium test is a day; keeping 400 of them green across weekly iOS releases is a career. MobileBoost attacks both ends but is strongest at the second. The Test Agent consumes a pull request, extracts or generates the relevant cases, runs them in parallel against simulators or real devices, and posts pass/fail plus artifacts and logs to CI status checks. That last step matters: the artifact trail is what makes a failed AI-authored test debuggable rather than an opaque red X. The agent-skill YAML is the piece worth studying if you're building agentic workflows. You hand your coding agent skills plus a pointer to your own test-standards.md, and it produces tests in a declared format (xcuitest in the example) with validation on. That's an unusually honest answer to the "agent writes the feature but not the test" gap — you're not just getting generation, you're getting generation constrained by your conventions. On infrastructure, the claims are specific and checkable rather than broad: offline mode on iOS simulators, App Clips on simulators without physical devices or TestFlight distribution, camera-feed image injection for QR/document/AR flows, audio transcription and injection for voice-powered flows, network capture and mocking to keep payment and failure scenarios off production, accessibility flagging during ordinary runs, and performance regression detection on each PR. Scale reaches 100+ concurrent device setups. Any one of those can be the reason a suite is impossible on another platform. Trade-offs are real. The visual editor and natural-language flow descriptions deliberately trade low-level control for accessibility — teams with intricate state machines or bespoke assertion logic will hit the ceiling. The human-refinement loop the vendor advertises (product managers and QA analysts adjusting generated flows) is a genuine strength for coverage volume and a genuine weakness for anyone who wants the test to be the spec. This is cloud execution only, so if your compliance posture requires running tests inside your own network, look elsewhere. Customer evidence is specifics rather than logos: Lyft comparing seven tools and citing AI-driven UI interaction plus fast response times; Noom evaluating fourteen and citing native iOS and Android support, prompt-based creation inside their codebase, and reliable CI execution; Compass reporting an 88% nightly pass rate after more than doubling from where they started; Duolingo's Brock Janikowski citing a 70% cut in manual regression workflows. Treat vendor-reported numbers as directional, but the named people and stated evaluation counts are more than most competitors publish. Where it fits: mobile teams with CI already wired up who are drowning in regression maintenance, and teams experimenting with coding agents that need a test-authoring counterpart. Where it doesn't: security-constrained