Autonomous Coding
Ship an iOS app
without buying a single Mac.
Building an iOS app has always required a physical Mac, because Apple's own tooling only runs on macOS — and that constraint applied to AI coding agents just as much as to people. Cognition's autonomous agent Devin is now taking on that constraint itself, by carrying a full virtual macOS environment into the cloud.
The Constraint
The unwritten rule:
no Mac, no app
Xcode, the iOS Simulator, and app signing only run on macOS. Even a human developer working from a cloud IDE or a Linux CI box eventually has to route through somebody's Mac. That constraint applied to AI coding agents too — leaving a strange division of labor where an agent could write all the code, but the actual iOS build still needed a human's Mac.
According to Cognition's official post, "Bringing macOS to Devin," Devin's cloud agents now carry a real macOS environment of their own — Xcode, iOS Simulator, and signing setup included. The headline of the announcement is that Devin can now go from writing code to building, testing, and shipping a TestFlight beta for an iOS or macOS app end to end, without a single physical Mac in the loop.
Who's Behind This
A company whose valuation
tripled in under a year
Devin's maker, Cognition, is one of the fastest-rising valuations in the agentic coding race.
Cognition was founded in November 2023 by three International Olympiad in Informatics gold medalists — Scott Wu, Steven Hao, and Walden Yan. It launched Devin in March 2024 as an "autonomous AI software engineer" and quickly drew attention. After acquiring rival coding tool Windsurf in July 2025, its valuation climbed to $10.2 billion by that September, then to $26 billion in May 2026, and is now reported at roughly $47 billion.
Per the official blog, this macOS support is built on top of accessibility-cli, an open-source tool originally released by the Dioxus project. That points to something closer to an agent actually seeing and operating a real screen, not just being handed a bare virtual machine. For a fast-scaling company, it also reads as a deliberate expansion from web and server-side code into iOS and macOS app development.
Before / After
What changes when the
Mac dependency goes away
| Before | Devin with macOS support |
|---|---|
| Build environment required a physical Mac | Runs entirely in a cloud virtual macOS environment |
| Simulator testing was done by hand | The agent operates the simulator and self-verifies |
| Progress shared via manual screenshots | Screen recordings sent to Slack automatically |
| TestFlight registration and delivery by hand | The agent issues a TestFlight link automatically |
In Practice
Who it helps, and how
Mobile engineers
iOS work no longer stalls waiting on Mac hardware procurement. That said, agent-driven signing and distribution flows likely still deserve a human check before anything ships.
Startups & PMs
Windows- and Linux-centric teams can build a mobile version in-house without investing in Mac hardware up front. Auto-issued TestFlight links also shorten the review cycle.
Solo builders
Someone without a Mac can now turn an app idea into a real iOS build. Anyone who already owns Mac hardware sees almost no benefit here.
A hardware constraint is quietly losing its meaning, one virtual environment at a time.
What's Next
What to try next, and what to watch
Start small: an internal iOS tool or a test app, not a flagship release. Next, put a mandatory human review step on any signing certificates and provisioning profiles the agent generates, before scaling further. And if you automate all the way to TestFlight delivery, write down explicitly who owns the final approval click — that rule should exist before the first real release, not after.
There are open risks, too. Apple's signing and distribution policies shift over time, and it's not yet clear how well an automated flow keeps pace with future policy changes. A "see and operate the screen" style of automation may also be more error-prone for apps whose UI changes frequently. The pace of this feature rollout fits a company whose valuation is climbing fast — but how reliable it is in production is still something to judge from a track record, not from a launch post.