Prelint
Overview

Product drift

Business logic#

AI changes pricing, discount, or billing rules without understanding the financial impact.

An agent switches pricing from vendor currency to customer currency. FX exposure shifts silently. Revenue leaks. No test fails.

Compliance#

AI stores data, skips consent flows, or breaks retention rules it was never told about.

An agent adds a user activity log but stores IP addresses without a retention policy. You are violating GDPR and CCPA and will not know until the audit.

Tooling and infrastructure#

AI introduces new vendors, new SaaS, and new dependencies because it can.

An agent adds Twilio. You already standardized on AWS SNS. Now you pay twice and split your infrastructure.

Domain language#

AI uses generic industry terms instead of your established internal language, splitting concepts that should be unified.

The codebase uses "merchant." AI introduces "seller," "vendor," and "partner" across new services. Same entity, four names, three confused teams.

Scope creep#

AI builds features nobody asked for.

An agent adds a full i18n framework with locale files for 12 languages. The product serves English-speaking users only.

Strategic drift#

AI builds off-roadmap features, public APIs for internal tools, or premature abstractions.

The spec says internal dashboard. The agent builds a public REST API with versioning, rate limiting, and OAuth. Three sprints on something not on the roadmap.

Why review catches what tests do not#

AI does not slow down when it is wrong. It keeps building, confidently, in the wrong direction. Code review constrains technical error. Product review constrains spec drift. Without the second layer, each merged pull request inherits the drift of the last one and the error compounds.

With a product review step, the agent gets corrective feedback before merge and the error stays bounded.

Powered by Docsbook