liveBook · Manning

Code review

As explained in Vibe Engineering

The manual examination of proposed source-code changes to identify defects, risks, and conformance or maintainability problems before integration.

Code review examines proposed changes before integration, using human judgment to check logic, intent, conventions, and architectural alignment. Passing tests is evidence, not a substitute for understanding the change: a professionally written diff can still hide a wrong design choice or subtle coupling. section 10.5 section 8.1.6

Change size is a practical control. A large migration may extract more work from an AI tool, but a giant pull request makes defects harder for a reviewer to catch. Split work into increments large enough to justify prompting overhead and small enough to review and test reliably. section 2.5

Use review as one gate in a verification pipeline:

  • Before merge, read for logic and intent, check team conventions and architectural fit, and run integration tests for affected components. section 8.8.4 section 10.5
  • On every change, use type checking, linting, security analysis, unit tests, and CI to catch mechanically detectable problems. Automated checks and human review catch different failures. section 8.8.4 section 8.1.6
  • Scale further verification to criticality: standard production features warrant review and integration tests, while security, financial, or safety code can justify more intensive techniques. section 8.8.2

The review queue becomes the bottleneck

If six developers produce eight pull requests a day, they create 48 daily reviews. Two reviewers clearing ten each leave 28 queued per day. Under that pressure, skimming a payment change like a variable rename creates review theater: approval appears to provide oversight without the necessary judgment. Count merged pull requests and how many a human actually read to establish a baseline. section 10.5 section 10.8.4

Automation can release review capacity, but an auto-merge is a bet that a change is safe without a human read. Start conservatively, measure which patterns are approved without modification, promote those patterns for automation, and recalibrate. Track escapes and what the released capacity actually produces; avoided review time is not automatically a budget saving. section 10.5.3

For unsupervised agents, replace continuous human supervision with validation, anomaly detection, audit logs, circuit breakers, and regular review of sampled outputs. Governance may also require evidence of meaningful human review; repeatedly clicking approval is not necessarily sufficient proof. section 5.3.4 section 5.5

How Vibe Engineering covers it

How deep it goes
Explains review criteria, verification gates, queue capacity, and conservative auto-merge calibration, without a detailed line-by-line review procedure.
Examples
Uses a legacy application migration and a 48-pull-request daily queue to illustrate change-size and reviewer-capacity tradeoffs.
What you can do after
Practice sizing changes for review, checking intent and architecture, and measuring the proportion of merged pull requests read by humans.
How it connects
Connects code review to continuous integration, integration testing, static analysis, observability, and compliance.

Also explained in

Compare how 12 books explain code review →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Vibe Engineering. The section links point to the parts of the book it draws on.