Everyone figured AI would eventually crack mobile testing’s nut—those brittle locators, endless emulator tweaks, the sheer pain of scaling UI tests across iOS and Android. Tools like Appium promised cross-platform bliss, but delivered nightmares of XPath fragility and maintenance hell. Detox? Faster, but still code-heavy. Enter Finalrun. This Show HN from finalrun.app changes the game: spec-driven testing via English YAML and computer vision, all from a dead-simple CLI. No locators. No scripting loc_id BS. Just describe what you want, point at real devices or sims, and let AI models like Gemini or Claude execute.
Boom. One command.
What Finalrun Actually Does (And Why It Matters Now)
Mobile QA budgets ballooned 20% last year—Gartner’s word, not mine—thanks to app complexity exploding with AR, payments, and cross-device madness. Devs waste 30-40% of cycles on flaky tests, per Stack Overflow surveys. Finalrun? Cuts that with YAML specs that read like Jira tickets.
Take their minimal example:
name: login_smoke description: Verify that a user can log in and reach the home screen. steps: - Launch the app. - Enter ${secrets.email} on the login screen. - Enter ${secrets.password} on the password screen. - Verify the home screen is visible.
That’s it. No ‘tap element with id=foo’. AI vision spots the screens, fills forms, verifies outcomes. Install via curl -fsSL https://raw.githubusercontent.com/final-run/finalrun-agent/main/scripts/install.sh | bash—grabs Node, CLI, even Android/iOS SDKs if missing. Run finalrun test smoke.yaml –platform android –env dev. Artifacts land locally for inspection. Slack community? Waitlist for cloud devices. It’s CLI-first, repo-local, secrets-safe via .env (gitignore’d, obvs).
But here’s my edge: this echoes Cypress’s 2017 web testing upset. Back then, Selenium ruled with verbose WebDriver code; Cypress auto-detected waits, locators. Finalrun does that for mobile—but with vision AI, not heuristics. Bold call: if they nail flakiness under 5%, it’ll capture 15% of the $5B mobile testing market by 2026. Appium’s share? Shrinking already.
Why Finalrun Crushes Appium and Friends
Appium’s locators flake on UI tweaks—5% regression per release, industry avg. Espresso? Android-only. XCUITest? iOS silo. Finalrun’s English + vision sidesteps it all. Describe ‘enter email on login screen’; Gemini-1.5-flash (their rec) sees pixels, acts.
Setup’s a dream. finalrun doctor checks hosts. finalrun check –env dev validates YAML bindings. Suites bundle tests; finalrun suite smoke.yaml scales it. Workspaces in .finalrun/ keep it git-friendly—specs, env stubs (no real secrets), config.yaml for model/env prefs.
Skeptical? Me too, at first. AI vision’s hit 90%+ accuracy in benchmarks (see Google’s Mobile VLM evals), but real apps with dynamic UIs? Early days. Their YouTube demo (linked) nails a login flow flawlessly—app launch to home in 30s. No mocks. Real device.
And the market timing—perfect. With Apple’s Vision Pro pushing multimodal AI, mobile devs crave this. Finalrun’s not hype; it’s pragmatic. CLI over SaaS lowers barriers (no vendor lock). But watch: if cloud waitlist drags, momentum stalls.
Short para: Pricing? OSS agent, cloud TBD.
Is Finalrun Production-Ready for Your Mobile Team?
Not yet—beta vibes strong. Docs scream ‘workspace-root .env for secrets, never commit’. Load order: .env. overrides .env, then process.env wins. Smart. Supports Claude, Gemini, OpenAI—pick your model poison.
Run artifacts? ~/.finalrun/workspaces//artifacts. Serve ‘em: finalrun start-server –workspace /path/to/app. Inspect from terminal or browser. finalrun runs lists history.
Critique time—their PR spins ‘AI-driven CLI’, but it’s really YAML interpreter + LLM agent + native drivers. Solid, but don’t sleep on edge cases: custom fonts, dark mode flips, A/B tests. Vision LLMs hallucinate 10-15% on occlusions (arXiv papers confirm). They’ll iterate.
Teams at scale? Start smoke.yaml suites. Bind ${secrets.email} via .finalrun/env/dev.yaml (placeholders only). It’s dev-friendly—run from subdirs, auto-finds workspace.
Historical parallel: Remember Selenium Grid’s cloud pivot? Finalrun’s waitlist hints same path. Prediction: They’ll hit $10M ARR by partnering Headspin/ BrowserStack for fleets.
Look, if you’re scripting login tests in Kotlin/Swift today—switch. This halves toil.
Why Does This Matter for Mobile Devs Right Now?
Flutter/React Native shops: rejoice. Cross-platform hell solved sans code. Enterprise? Compliance via local runs, auditable artifacts.
Market dynamic: AI test agents (Replicate, Testim) nibble web; mobile’s underserved. Finalrun leads—first with English specs + vision CLI.
Imperfection: YAML’s verbose for complex flows. But beats 200-line Appium suites.
DevOps angle—pipe to CI: finalrun test –platform ios | jq for JSON reports (future?). It’s raw now, but extensible.
Wrapping the unpack: Finalrun isn’t vaporware. Curl it today, test tomorrow. Skeptics, join Slack—feedback shapes it.
**
🧬 Related Insights
- Read more: AMD’s Lemonade: Zesty Local LLMs or Just NPU Bait?
- Read more: Ditching 1C and SAP for Odoo 19: The Hidden Traps
Frequently Asked Questions**
What is Finalrun and how do I install it?
Finalrun’s an AI CLI for mobile testing: YAML specs in English, vision-powered execution on Android/iOS. Install with one curl command—it sets Node, CLI, SDKs.
Does Finalrun work with real devices or just simulators?
Both—local emulators/sims via doctor check; cloud devices on waitlist. Runs artifacts stored for inspection.
Will Finalrun replace Appium for mobile UI testing?
Potentially, yes—ditches locators for AI vision, slashing maintenance. But monitor flakiness; it’s early but promising.
