Skip to content
agentusehub

Migrate a legacy product stack with Claude Code

An engineer ran a proof-of-concept that consolidated 4 repos into 2, moved MySQL to PostgreSQL, upgraded frameworks and replaced the auth model, in about 15 work days.

Time saved
Not reported
Setup time
Not reported
Running cost
Not reported
Cost range
Not reported

Use case overview

An engineer ran a proof-of-concept that consolidated 4 repos into 2, moved MySQL to PostgreSQL, upgraded frameworks and replaced the auth model, in about 15 work days.

Reported experience

  1. What they did

    The engineer wrote layered CLAUDE.md context files and gave Claude Code the legacy repos, read-only MySQL and PostgreSQL connections, and a reference implementation. Claude drafted the design, and parallel reviewer agents stress-tested it. Claude then made 19 Jira tickets. Each ticket got a design doc and plan doc that he reviewed, then ran in its own git worktree, with CI monitored through a loop.

  2. How it went

    About 15 work days produced roughly 50 tickets and 580 passing tests, with 95%+ API coverage and zero bugs or vulnerabilities at QA handoff. AI review caught only 30-40% of issues, and the Jira MCP could not link tickets.

  3. Worth knowing

    Cost was close to $5,000 in Bedrock tokens, mostly up front. Also, the Confluence MCP silently truncated pages over about 5KB, overwriting two large design docs.

Example prompts

Try with Claude Code
I want a proof-of-concept migration of [legacy product] from [current stack, e.g. 4 repos, MySQL, old framework version] to [target stack, e.g. 2 repos, PostgreSQL, current framework]. Start by reading the codebases and writing a phased migration plan with risks, then execute the first phase on a new branch with tests passing. Pause for my review before merging anything or running changes against any real database.
Software & DevOps

Port a Chrome extension to Safari

The author had ChatGPT take the Session Buddy extension source from Chrome, port it to Safari and install it, taking about 20 minutes.

ChatGPT