Developer reportedly rewrote 65k lines of Go in Rust using Fable for $400

The rewrite leveraged Fable, an LLM-based tool designed to treat code as structured data rather than text. By extracting representations like state machines or ontologies, the tool could refactor logic before generating Rust code. This approach contrasts with traditional line-by-line translation, which often produces brittle or unidiomatic output. The $400 cost suggests the tool handled the bulk of the work, though the developer likely performed cleanup or validation afterward.

The method’s viability hinges on the quality of the intermediate representation. Fable’s ability to trace data flow precisely allowed it to avoid common pitfalls like shared mutable state or hidden dependencies, which are frequent sources of bugs in manual rewrites. However, the tool’s effectiveness may degrade in codebases with complex control flow or domain-specific logic that resists abstraction. The developer’s prior Rust knowledge also likely influenced the outcome, as they could guide the tool toward idiomatic patterns.

While the cost and speed are compelling, the rewrite’s long-term impact is unclear. The target project, rune, is a terminal editor with a focused feature set, which may not stress-test the tool’s limits. Larger or more complex systems could expose gaps in Fable’s ability to handle edge cases, such as concurrency models or platform-specific APIs. Additionally, the rewrite’s correctness relies on the tool’s output, which may introduce subtle bugs not caught by initial testing.

The approach aligns with broader trends in using LLMs for code transformation, but its scalability remains unproven. The developer’s workflow, extract, operate, regenerate, mirrors techniques used in Bun’s Zig-to-Rust rewrite, suggesting a repeatable pattern. However, the lack of independent verification or benchmarking means the results should be treated as anecdotal. Teams considering similar migrations would need to evaluate Fable’s compatibility with their codebase’s architecture and domain.