Author(s): Udaykiran Estari

Originally published on Towards AI.

The 11-day rewrite that sparked an industry debate holds a much deeper lesson: test suite readiness as a hard gate.

Bun ported 535,000 lines of Zig into a million lines of Rust in 11 days for $165,000 — and the Zig language’s own creator called the result “unreviewed slop.” The gap between these two realities reveals a hard truth about agentic rewrites: the safety net wasn’t the 64 parallel Claude agents, but a test suite the team spent years building before a single line of Rust existed. If your team is eyeing a similar rewrite because the headline numbers look irresistible, this playbook decodes the mechanics that made it possible and the institutional readiness required to survive it.

After the introduction, the article explains how the “11-day” outcome depended on long-prepared organizational assets rather than speed from AI alone: an existing language-independent, high-assertion test suite that acted as a behavioral oracle, plus a structured migration pipeline. It lays out four reusable mechanisms—mechanical porting before idiomatic refactoring, using a shared PORTING.md contract, precomputing lifetimes/ownership in LIFETIMES.tsv with adversarial validation, and isolating implementers from diff-only adversarial reviewers—then details what actually broke during the project (semantic mismatches, operational fragility from shared workspace state, and environment-driven failures) and what the postmortem numbers showed for ROI. The author also takes Andrew Kelley’s rebuttal seriously, arguing that tests can prove regressions against known behavior but may miss unknown risks, especially in security- or integration-heavy domains, and frames the approach as viable only when the “behavioral surface area” is sufficiently deterministic and the organization already has the institutional artifacts to support agent-driven work. The piece ends with a readiness checklist and a conclusion emphasizing that AI agents are the fast part; the hard part is building the test safety net first.

Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI