Bloomberg's Automated Uplift ran LLM-assisted migrations across thousands of repositories: 2,333 PRs and 729 successful migrations
A paper from Bloomberg engineers (with UCL's Federica Sarro) reports on Automated Uplift, a multi-month initiative applying LLM-assisted migration pipelines across thousands of Bloomberg repositories. It covers two production campaigns: a mixed Python packaging migration (about 1,900 candidate repositories) and a uniform C++ build system migration (about 6,700 candidate repositories). The pipeline maximises deterministic processing, confines the LLM to bounded file transformations inside pre-computed context, and feeds in production telemetry such as execution logs and usage traces. The authors found that generating migrations was the easy part; the bottleneck was owning teams' cost of reviewing, testing and deploying them.
The numbers (as reported)
| Pull requests | 2,333 generated across two campaigns |
|---|---|
| Successful migrations | 729 |
| Python campaign | About 1,900 candidate repositories |
| C++ campaign | About 6,700 candidate repositories |
Facts
- Company / subject
- Bloomberg
- Tools
- Unstated
- Codebase
- brownfield
- Work type
- Migration
- Stack
- Python, C++
- Published
- 2026-11-03
- Last checked
- 2026-11-09
What the source claims Practitioner (own blog/paper)
Claimed by the builder, writing about their own work. Not checked independently.
- Two campaigns produced 2,333 pull requests and 729 successful migrations 2,333 pull requests, 729 successful migrations
“Together, these campaigns generated 2,333 pull requests, resulting in 729 successful migrations.”
Caveats
Written mostly by Bloomberg engineers who ran the work, with an academic co-author. Drafted from the abstract only; the abstract names no model or coding tool and gives no time or cost figures. 729 successes out of 2,333 PRs means many PRs did not result in a migration.
More stories by tool: Unstated
Related topics: MigrationsBrownfield codebases