Using AI · Growth Library
Use AI to Scaffold Code, Then Make Tests Decide
Accelerate the first implementation without treating generated code as reviewed code
Try this
Define the behavior and constraints first. Ask AI for a small implementation. Read the diff, run the existing test suite, add tests for edge cases and failure paths, and inspect security or data-handling implications. Keep only code you can explain and maintain.
Check-in: Did AI shorten implementation time without increasing code I do not understand, test, or want to own?
Evidence: Reviewed · source · Peng et al. (2023) — The Impact of AI on Developer Productivity: Evidence from GitHub Copilot
Review note: Reviewed against Peng et al. (2023). A controlled coding experiment supports a speed signal on one bounded task. The source is not treated as proof of universal software quality, security or maintainability gains; tests and diff review are explicit gates.
Generated code can be fast to obtain and expensive to misunderstand. Use the speed; do not outsource the acceptance criteria.
1. Define behavior before code
State inputs, outputs, failure behavior, constraints, and the tests that would convince you the task is complete. This gives the model less room to solve a different problem elegantly.
2. Generate a small change
Prefer a reviewable diff over a repository-wide rewrite. Small generated changes are easier to test, explain, and revert.
3. Let executable checks disagree
Run the existing suite and add edge cases the generated code may have missed. When tests and the explanation conflict, investigate instead of asking the model to sound more confident.
4. Keep only maintainable code
Reject code you cannot explain, secure, or support. AI assistance changes how the code was drafted; it does not change who inherits the bug.
Questions to consider
What did the coding experiment show?
In a controlled experiment on a JavaScript HTTP-server task, developers with GitHub Copilot completed the task faster on average. It was a specific task and the result should not be generalized to every language, repository, or quality dimension.
What is the minimum review?
Read the diff, run relevant tests, add missing edge cases, inspect dependencies and secrets, and understand the code path well enough to debug it later.
Sources
Evidence state: practical. Sources are provided so the underlying material can be inspected directly.
Maintenance record
Review history
Current status: Reviewed. An editorial or evidence review is recorded and no later event changes that conclusion.
2026-09-15 · reviewed · Brali editorial agent Source: review-registry-using-ai-starters.json
Keep the productivity signal narrow and pair it with ordinary software QA so generated code is treated as an implementation candidate, not a correctness proof.
Browse the review ledger · Challenge or update this hack
Implementation observations can trigger a review, but they cannot change an evidence conclusion by themselves. Commercial relationships do not control review status or outcomes.
Continue from here
Choose the next useful path.
Article versions
Brali keeps substantive article history visible. Later reviews may refresh wording, sources, examples, or boundaries without silently replacing the record.
- Version 2026-09-15: first recorded publication in the migrated Brali corpus.
- Evidence review 2026-09-15: reviewed by Brali editorial agent; evidence state is
reviewed.
Canonical source record · Evidence state: reviewed.