AI-generated code can turn an idea into a working artifact quickly, but a working result may hide defects, security risks, and implementation choices nobody understands. Vibe engineering keeps the speed of exploration while making intent, verification, and ownership explicit. Unlike intuition-first vibe coding, it treats generated code as a candidate to be checked, not evidence that the feature is correct. section 1.3 section 1.8
The central artifact is a human-authored executable specification: a contract such as a test suite that states the intended behavior and can be run against the implementation. Goals, constraints, edge cases, error handling, and non-functional requirements must be written down rather than left in someone’s head. section 1.8 section 1.0
The engineering loop carries a feature from exploration to owned code:
- Vibe: explore the problem with a simple prototype; use it to discover rules, data shapes, and edge cases rather than assuming the prototype is production-ready. section 1.5
- Specify and plan: turn human intent into executable behavior, then state architectural and operational constraints. Ask the AI for candidate plans and overlooked cases, but retain human judgment. section 1.5
- Task and verify: generate focused increments against tests and contracts, and check them before merging. Mechanical checks do not replace judging whether the result matches the business intent. section 1.8
- Refactor and own: improve the implementation and leave tested code the team can maintain. Prompts, outputs, and specifications also need owners. section 1.8 section 1.4
A 30-day refund policy
A product note such as “refunds within 30 days” leaves boundary times, money arithmetic, and rejection reasons open. An executable specification can distinguish an approved refund from an expired request or a partial refund exceeding the remaining amount. A focused prompt can then request a pure policy with clock and money ports, plus property tests around the 30-day boundary; the pipeline can run the resulting checks. section 1.5
Feature: Refund policy
Rule: Time-bounded, single-order refunds
Example: Within 30 days is allowed
Given an order "O1" purchased at 2025-08-01T10:00:00Z
When a refund of 25.00 USD is requested at 2025-08-28T09:59:00Z
Then the refund is approved and remaining refundable equals 75.00 USD
Example: After 30 days is rejected
Given an order "O2" purchased at 2025-08-01T10:00:00Z
When a refund is requested at 2025-08-31T10:00:01Z
Then the request is rejected with reason "expired"
Example: Partials must not exceed the original
Given prior refunds of 90.00 USD on a 100.00 USD order
When a refund of 20.00 USD is requested
Then the request is rejected with reason "exceeds-remaining"The refund examples make approval, expiry, and remaining-balance rules concrete before implementation. section 1.5
An ISBN validator
Two generations from the same ISBN-13 prompt produced different implementations; one rejected valid hyphenated inputs and referenced an undefined variable. Tests for hyphens, checksums, length, and nonnumeric characters make those expectations visible before accepting generated code. section 1.4
Unverified output creates trust debt: the immediate generation saving becomes deferred work for someone who must understand and make the code safe. Reduce it with honest tests, clear dependency boundaries, fast feedback, and review that scales. Increase agent autonomy deliberately, with verification and governance matched to the risk. section 1.4 section 10.8.4 section 1.8
