Use review loops to turn a spec into working code

A developer fed a spec into an implementation engine, then repeated verification and review passes until the code matched the plan.

Done with Custom agent· Advanced· Oct 2026

What they did
The developer first pinned down the intended behavior and the technical approach, then wrote that up as a spec. They sent the spec to a durable implementation engine, which produced code and then ran automated checks. Different agents, using different models, reviewed the output, grouped problems into clusters, and screened out false alarms before fixes were applied. The developer’s own review stayed limited to code layout and structure.
How it went
The process kept cycling until the implementation aligned with the spec. The commenter did not give timings or counts, but argued that one-shot builds, even for large projects, tend to leave too many defects.
Tips
This method depends on repeated passes, not a single generation. Keep the spec explicit, and separate must-fix issues from optional ones before making changes.

Try it yourself with Codex

I’ve put my spec at [spec file path]. Read it, implement the required behavior in this repository, and stop before any irreversible action such as deleting files, submitting changes, or making external calls. After the first pass, run automated checks and then do a review round that compares the result to the spec. Group findings into must-fix, should-fix, and ignore; double-check anything that may be a false positive; and fix only the must-fix items. Repeat the build-and-review cycle until all spec items are satisfied and the tests pass, or until you reach [number of rounds]. Then summarize what changed, what remains open, and any spec items that were ambiguous.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on news.ycombinator.com ↗

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.