---
title: 'Google: LLM-assisted 32-bit to 64-bit identifier migration'
slug: google-llm-code-migration
stable_id: 4a77ab1ce686bbd8
company: Google
function_code: software_engineering
pattern_codes:
  - creator_to_judge
evidence_strength: verified
publication_tier: showcase
freshness: current
reviewed_at: '2026-08-23'
updated_at: '2026-08-22'
source_quality_summary: 1 peer reviewed; publication outcomes are verified.
caveat_summary: >-
  Experience report without a randomized comparator. Time reduction is the developers'
  estimate, not instrumented time measurement.
collections:
  - human-still-decides
bundle_version: 1.0.0
bundle_fingerprint: sha256:c23c6cc2b88153d008ea8fda928f632ce0011fc2d4c5036672a16e5d895bab93
canonical_url: https://brianletort.ai/transformations/google-llm-code-migration
---

# Google: LLM-assisted 32-bit to 64-bit identifier migration

Large migrations can combine machine-found locations, LLM edits, automated checks, and engineer review.

Function: Software engineering. Patterns: Creator to judge. Evidence: verified.

Freshness: current. Reviewed: 2026-08-23. Updated: 2026-08-22.


Source quality: 1 peer reviewed; publication outcomes are verified.

## Before

1. **Software engineer** — Finds migration sites and writes context-specific edits across the monorepo. (control: Manual code review and testing.)
2. **Software engineer** — Repairs failures and submits each change list. (control: Reviewer approval.)

## After

1. **Migration tooling** — Discovers likely change locations and prompts a monorepo-trained LLM to generate edits. (control: Location discovery, compilation, tests, and static checks.)
2. **Software engineer** — Reviews, repairs where needed, and submits validated change lists. (control: Human review and standard submission controls.)

## Decision rights

Automated checks may reject edits; engineers decide whether to repair and submit; reviewers retain merge approval.

## Exception path

Failed or semantically difficult edits return to engineers for manual correction or conventional migration tooling.

## Outcomes

- **LLM contribution to accepted migration work** (verified): Traditionally manual, context-dependent migration edits. → 74.45% of 595 submitted code changes and 69.46% of 93,574 edits were LLM-generated.. Experience report without a randomized comparator.
- **Estimated migration effort** (verified): Earlier manual migrations. → Developers estimated a 50% reduction in total time.. Time reduction is the developers' estimate, not instrumented time measurement.

## Executive lesson

The reusable pattern is not autonomous coding; it is AI generation inside a deterministic discovery, validation, and review envelope.

## Anti-pattern

Allowing model-generated repository changes to bypass compilation, tests, or human review.

## Questions for leaders

- Which migrations are variable enough to benefit from LLMs?
- What evidence must pass before a change reaches review?
- How will estimated savings be instrumented?

## Sources

- [Migrating Code At Scale With LLMs At Google](https://doi.org/10.1145/3696630.3728542) — ACM Proceedings on Software Engineering