Claude Code migrates a password-reset flow, with costly review
A developer reports Claude Code produced small-scale password-reset migration code, but review, architecture drift and cleanup made the task slower than manual work.
news.ycombinator.com
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Migrate only the password-reset flow described in this plan. First list the affected files, compatibility risks and acceptance checks. Keep unrelated refactors out of scope, make small changes and report any uncertainty before continuing.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source poster / publisher
- Chris_Newton
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- The author reports usable code at small scale, but says close supervision and cleanup made the half-day task slower than comparable manual work.
- Human involvement
- The developer says they stayed closely involved, reviewed every proposal for subtle errors and scope drift, and cleaned up the final output. No production deployment is reported.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Plan the password-reset migration in the agent harness before changing the legacy application.
- Have Claude Code update the flow while supervising each proposed change closely.
- Review architecture and scope to catch subtle errors and unrelated modifications.
A Hacker News commenter using the handle Chris_Newton describes testing Claude Code on a routine company maintenance task: moving a password-reset flow in a long-lived web application from legacy libraries and architecture to newer equivalents. The developer first prepared a plan with the agent harness, then supervised the coding work closely. The run used company API tokens and took about half a working day in wall-clock time.
The author says Claude produced code of decent quality at small scale, but its architectural choices were uneven and it tried to pull unrelated changes into the task. They had to monitor proposals to prevent subtle errors and scope drift, then spend substantial time cleaning the final result. Compared with similarly complex migrations the author had completed manually, this attempt was slower overall. The account is a useful counterexample to treating rapid code generation as equivalent to faster delivery.
The source does not name the libraries, publish the patch or provide an exact API bill; it compares the token cost with roughly two months of a Pro subscription. The practical result is a completed experiment with a mixed quality assessment, not a production migration independently verified by the site. For this author, the plan and close review were necessary, and they erased the expected time savings.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Hacker News · Chris_Newton
Frequently asked questions
What part of the application was being migrated?
The author describes a password-reset flow in a long-lived web app, moving away from legacy libraries and architecture.
Did Claude Code make the task faster?
No. The author says supervision and cleanup made this attempt slower than similar migrations done manually.
What problems did the developer report?
The code drifted architecturally and the agent tried to include unrelated changes; the developer says close review caught subtle errors.
More details
Remaining steps
- Clean up the output and compare the elapsed effort with similar manual migrations.
About this account
Claude Code generated changes for the planned password-reset migration and proposed adjacent modifications during the supervised session.
Hacker News comment opened through the official Firebase item API on 2026-10-18; author handle and exact timestamp verified there. The account concerns company work and states that API tokens were used. It does not publish a patch or exact cost, and does not report a production deployment. Suggested prompt written by Agents In Use.
Credit: Chris_Newton (Hacker News)
Source published
Added to Agents In Use
Source checked
Tools and topics
Claude Code · Claude Opus · Company API access · Legacy web application
Code migration · Scope control · Human review
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.