Port the Copilot runtime to Rust agentically
devops. Task: Microsoft used AI agents (GPT-5.6 Sol, Claude Opus 4.8) to port the GitHub Copilot runtime from TypeScript to Rust. The process involved module-by-module replacement, spawning child sessions for complex files, and iterative investigation rather than direct code generation. Developers supervised the process, managing regressions and ensuring the port remained simple without structural optimization.; tools: custom-built or unnamed agent. Human role: The original post does not say how a person stayed involved.; limitations: AI agents struggle with Rust's semantic correctness, often producing code that compiles but behaves incorrectly. The process requires significant human oversight to identify regressions and validate logic. Token costs are high ($120k for this scale), and the approach is not suitable for optimizing existing structures, only for direct porting. Human developers must treat the compiler as a teacher, not an oracle..
Tools
custom-built or unnamed agent
Published on Agent Work Atlas:
Overview
devops
Workflow
Microsoft used AI agents (GPT-5.6 Sol, Claude Opus 4.8) to port the GitHub Copilot runtime from TypeScript to Rust. The process involved module-by-module replacement, spawning child sessions for complex files, and iterative investigation rather than direct code generation. Developers supervised the process, managing regressions and ensuring the port remained simple without structural optimization.
Reported result
The migration converted 430,000 lines of TypeScript into 800,000 lines of Rust, costing ~$120,000 in tokens and 3 weeks of developer time. Performance benchmarks showed a 15.9x speedup in session lifecycles and significant memory reduction (126 MB vs 1,383 MB). However, dozens of regressions occurred, requiring human intervention to fix semantic errors that the compiler did not catch.
Human role
The original post does not say how a person stayed involved.
Limitations
AI agents struggle with Rust's semantic correctness, often producing code that compiles but behaves incorrectly. The process requires significant human oversight to identify regressions and validate logic. Token costs are high ($120k for this scale), and the approach is not suitable for optimizing existing structures, only for direct porting. Human developers must treat the compiler as a teacher, not an oracle.
Try it
I need to port a legacy TypeScript module to Rust using AI agents. Here is the source file: [replaceable input: path to TypeScript file]. Use GPT-5.6 Sol for initial analysis and Claude Opus 4.8 for code generation. First, perform a dry run on a small subset of functions. Generate the Rust code, then run the existing test suite against the new implementation. If tests fail, analyze the error logs and iterate. Do not deploy or merge any changes until I manually review the diff and confirm all tests pass. Report the token usage and any regressions found during the dry run.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.