How I Used AI Agents to Migrate my Entire Site to Astro
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
A developer shares a detailed case study of migrating a 400+ page production site (six years old, built with the Swift static site generator Publish) to Astro using AI coding agents. The workflow: turn 'feature parity' into a testable migration contract, prototype the riskiest technical assumptions first, build a deterministic (non-LLM) migration tool for content transformation, split visual verification into fast and exhaustive tiers, and keep final product and production decisions human-owned. The author ran agents (GPT 5.6 Sol on a $100 Pro plan) inside tmux sessions on a VPS using Git worktrees, completing roughly 100 issues and 80 PRs over two weeks. References the Bun-to-Rust rewrite project as inspiration and cautionary tale, since that migration reportedly still isn't complete.
Table of contents
How I Approached The ProblemMy SetupProject Kick-OffThe Migration Was Not One TaskStage 1: The Initial InterviewStage 2: Content Migration PrototypeStage 3: Building a Migration ToolStage 4: Rebuilding the SiteStage 5: Visual VerificationThe Final Production Decision Is YoursWhy You Still Need To Be An ExpertQuestions this post answers
How can I verify that an AI-driven site migration hasn't broken anything across hundreds of pages?
Combine curated visual screenshot comparisons with direct automated checks for every part of the migration contract, since screenshots alone can't catch issues like corrupted inline code or broken redirects. One migration split verification into fast focused checks for normal PRs and a full exhaustive suite (route/redirect checks, metadata and feed checks, content-transformation checks, and integration-boundary checks) run only at release-candidate gates. daily.dev surfaces practical verification strategies for developers shipping large-scale AI-assisted migrations.
Why did the Bun rewrite in Rust using AI agents run into problems?
The Bun rewrite in Rust used 64 parallel AI agents at an estimated API cost of $165,000, but a follow-up analysis found the rewrite still isn't complete despite initial success claims, raising doubts about letting large numbers of agents run unchecked on big migrations without human oversight at each step. developers weighing large AI-driven rewrites can follow real-world outcomes like this one on daily.dev.
What is a good workflow for migrating a legacy codebase to a new framework using AI coding agents?
Define the outcome and constraints as a testable migration contract, prove the riskiest technical assumptions with a small disposable prototype, build a deterministic (non-LLM) tool for repeated transformations with its own test harness, verify actual behavior rather than plausible-looking output, and keep high-stakes product and production decisions human-owned rather than delegated to agents. daily.dev helps developers compare hands-on approaches to structuring AI-assisted migration projects.