1,370 engineers, a 10,000-line Scala-to-Java migration, four days. What it took, what it didn't.
Stripe
10,000 LOC in 4 days
Stripe rolled out Claude Code to 1,370 engineers and used it for a 10,000-line Scala-to-Java migration. The work that previously would have eaten weeks of senior-engineer time landed in four days, with reviews handled by the same engineers who would have written it. The unlock wasn't typing speed — it was that the agent kept perfect track of cross-file consistency, which is exactly what humans get wrong on migrations.
The Stripe case study is the one most often cited because the numbers are big, but the structure of the work is what's worth studying. Migrations are pure cross-file consistency problems. Every change in one file has to land in fifty others, with type signatures aligned, error handling aligned, and tests still green. Humans bleed precision on the 47th file. Agents don't.
What you can copy from Stripe: scope your migration in passes, not all at once. First pass: rename types only. Second pass: replace one helper. Third pass: update callers. Each pass is a clean diff. The agent excels at this because each pass is a single intent applied across many files.
Check your understanding
Q1.What kind of work is most natural for an agent like Claude Code?