■ HIGH RISK ■ Technology
In large part, yes for feature construction. The moat is thinner than mobile developers claim but thicker than web, because platform churn, device reality and store gatekeeping keep injecting problems no training set has seen yet.
“AI generates apps from screenshots now. But shipping to the App Store still requires a human's patience.”
Our AI replacement risk score — how we score jobs
Mobile development is UI work with hardware attached. A typical sprint mixes screen building in SwiftUI or Compose or React Native, networking with offline caching, state restoration, push notifications, permissions flows, background tasks, deep links, in-app purchases, analytics, and the perennial fight with battery and memory constraints. Then comes release: signing, provisioning, staged rollout, App Store and Play review, crash triage across a long tail of devices and OS versions, and support for users still on a phone from five years ago.
Generation is strong across the construction phase. Models produce screens from screenshots or designs, write networking layers, translate between platforms, generate tests, and increasingly handle framework migrations, which matters when Apple and Google rewrite their UI paradigms every few years. Cross-platform tooling plus generation means a competent two-platform app needs far fewer engineer-hours than it did, and much of the boilerplate around permissions, storage and navigation is effectively free now.
The friction that remains is unusually physical and unusually bureaucratic. Real devices misbehave in ways emulators do not: thermal throttling, flaky radios, aggressive OS process killing, camera and sensor quirks per manufacturer, and accessibility behaviour that differs by platform. Store review is an adversarial human process with shifting policy on privacy labels, tracking consent, subscriptions and content, and a rejection can be a business emergency. Performance and battery tuning demand profiling on hardware. Platform migrations arrive annually and arrive first as documentation, not as training data, which briefly hands humans an advantage every WWDC. Our score of 66 sits just under frontend for exactly these reasons.
Automatability: our editorial assessment of current and near-term AI capability
Compression is underway and accelerates to 2030. App development budgets already assume far fewer hours per feature, and cross-platform plus generation has cut the cost of a second platform sharply. Expect mobile teams to consolidate, with specialists retained for performance, native platform depth and release management. Annual OS releases keep supplying novel work that models handle poorly at first, which slows but does not reverse the trend.
It can build and ship a simple one, and it can produce most of the screens for a complex one. The failure points come later: crashes only reproducible on particular hardware, battery drain from a background task, an App Store rejection over privacy or subscription policy, and accessibility behaviour that differs subtly between platforms. Those still need a person.
Native depth is the more defensible position, especially for performance-critical apps, on-device machine learning, and anything touching camera, sensors or system integration. Cross-platform work is closer to web development in how easily it generates. That said, cross-platform plus generation is what many companies are choosing precisely because it needs fewer engineers.
It provides friction rather than protection. Store policy shifts frequently, rejections require negotiation and rework, and getting an app approved with subscriptions, tracking consent and privacy disclosures correct is genuine expertise. It is a meaningful part of why mobile automates slightly slower than web, but nobody should build a career solely on knowing how to answer a review team.
Get on the new platform APIs as soon as they are announced, because that knowledge is scarce before it reaches training data. Build profiling and crash-analysis skills on real devices. Take ownership of releases end to end. And move toward deciding what the app should do, since screen construction is the part being absorbed fastest.