A password flow moved, slower than doing it by hand
Chris_Newton had Claude Code and Opus move a password flow in an old web app. It cost about as much as two months of Pro, took half a working day, and was slower than similar moves he had done by hand.
Done with Claude CodeAdvanced
What was done
On 8 October 2026 Chris_Newton writes on Hacker News about a trial with Claude Code and the latest Opus, paid with the API because it was work for his company. The task was to move a typical password-reset flow from legacy libraries to newer ones. The cost was roughly the same as two months of a Pro subscription, and the wall time about half a working day. The code was decent at a small scale, but the architecture was not always on point. The agent pulled in other changes. He had to sit with it the whole way, read everything, and clean up the result afterwards. Compared with similar moves he had already done by hand, it was slower. He writes that it would have made several subtle errors if he had not stopped them.
How it went
The move was done, with cleanup afterwards. It was slower than the hand work he compares it with, and cost about two months of Pro. The post does not say the errors reached production.
Worth knowing
This is one trial of one flow. He sat with it the whole time. The cost is his estimate, not an invoice reproduced here.
How to do it
- Describe one flow, not the whole app. Ask for a plan before code.
- Read each change. Reject what does not belong to the task.
- Run the tests for the password flow yourself before anything is merged.
Try it with Claude Code
Move only [password flow] in [repo] from [old library] to [new]. Make a plan first. Do not change anything else. Show the diff. Do not merge. I will run the tests.
Replace text in brackets with your own.
Done when. The diff contains only that flow, and you have run the tests.
Safety rule. No merge. No changes outside that one flow. You read the diff.
Read the original
Read the original · Chris_Newton on Hacker News ·