Sixt's custom Claude agent migrates 780 iOS test files (~7,000 tests, 108,914 lines) from XCTest to Swift Testing for about $106
Sixt's iOS app had more than 19,000 unit tests, all on legacy XCTest. Rather than migrate by hand or use a generic agent, a Sixt staff engineer built a small TypeScript/Bun migration agent: a cheap text-to-text pass through the Anthropic SDK with a 1,199-line rules prompt, structural regex checks and `swiftc -parse` validation with retries, a read-only research agent for harder compile errors, a learned-pattern memory file, and a fixer agent that runs the real test suite via xcodebuild. Across 50 modules it processed 780 files and changed 108,914 lines (~7,000 tests) with a 97.3% success rate in about 3.8 hours of compute for about $106. The full migration is not yet merged, to avoid opening dozens of PRs at once.
The numbers (as reported)
| Files processed | 780 across 50 modules |
|---|---|
| Unit tests migrated | ~7,000 |
| Lines changed | 108,914 |
| Success rate | 97.3% (759 of 780; 21 failed) |
| Total compute | ~3.8 hours over 61 runs |
| Total AI cost | ~$106 (~$0.14 per file) |
Facts
- Company / subject
- Sixt
- Tools
- Claude Agent SDK
- Models
- Claude Opus 4.5
- In-house tools
- Swift Testing migration agent (Sixt)
- Codebase
- brownfield
- Work type
- Test migration
- Stack
- Swift, XCTest, Swift Testing, UIKit, Combine, TypeScript, Bun
- Published
- 2026-06-02
- Last checked
- 2026-11-11
What the source claims Practitioner (own blog/paper)
Claimed by the builder, writing about their own work. Not checked independently.
- 780 test files across 50 modules migrated with a 97.3% success rate 97.3% of 780 files
- 108,914 lines changed for about $106 in AI cost ~$106
“A single engineer day costs more than the entire migration did.”
- About 3.8 hours of agent compute versus an estimated months of manual work ~3.8 hours compute
Caveats
Written by the engineer who built the agent, on a personal Medium blog, with recruiting-style framing. Figures come from the agent's own cumulative stats. The author says the 21 failures were nearly all network errors to the LLM gateway, not migration mistakes. The manual-effort comparison ("months", 5–10 files a day per senior engineer) is the author's estimate. The migration is not fully completed or merged yet. Sixt used its internal LLM gateway; the post names the Anthropic SDK for bulk transforms and the Claude Agent SDK for the research and fixer agents.
More stories by tool: Claude Agent SDK
Related topics: Test migrationsBrownfield codebases