Google Code Migration
Combines internal code search, model generation, build, and testing to assist large-scale migrations. Task: Suggested implementation: Use internal code search to identify migration targets, generate diffs with a fine-tuned model, and validate via automated builds and tests. Human engineers review batches of pull requests and manually fix the remaining edge cases. The workflow relies on a state machine where tests, not the agent, determine completion.; tools: Tool described in the source. Human role: Not stated in the source; limitations: Source reliability is secondary, derived from public compilations that may contain vendor promotion. Prerequisites include access to internal code search, fine-tuned models, and robust CI/CD pipelines. Integration uncertainty exists regarding the specific model versions and infrastructure costs. Human approval is mandatory for merging changes..
Tools
Tool described in the source
Published on Agent Work Atlas:
Overview
Combines internal code search, model generation, build, and testing to assist large-scale migrations.
Workflow
Suggested implementation: Use internal code search to identify migration targets, generate diffs with a fine-tuned model, and validate via automated builds and tests. Human engineers review batches of pull requests and manually fix the remaining edge cases. The workflow relies on a state machine where tests, not the agent, determine completion.
Reported result
Reported outcomes include 80% of code changes being fully AI-written in one ID migration and 5,359 files migrated from JUnit 3 to 4 in three months. These metrics are attributed to the source author and have not been independently verified. Actual success rates and time savings may vary based on codebase complexity and model performance.
Human role
Not stated in the source
Limitations
Source reliability is secondary, derived from public compilations that may contain vendor promotion. Prerequisites include access to internal code search, fine-tuned models, and robust CI/CD pipelines. Integration uncertainty exists regarding the specific model versions and infrastructure costs. Human approval is mandatory for merging changes.
Try it
Act as a migration engineer. I need to migrate [replaceable: specific library or API] across [replaceable: number] files in [replaceable: repository name]. First, use [replaceable: code search tool] to list all files requiring changes. For each file, generate a diff using [replaceable: LLM model] based on [replaceable: migration rules]. Run [replaceable: build/test command] to validate. If tests fail, retry with error feedback. Stop after processing 5 files for a dry run. Present the diffs and test results to me for manual review before proceeding. Do not merge any changes without my explicit approval. Measure success by the percentage of files passing tests automatically.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.