Methods & how-tos
Practical, tool-agnostic methods for verifying AI-generated code: spec-vs-implementation checks, checkable task definitions, measurable verification debt, and review workflows that survive AI speed.
How to use this collection
Use methods & how-tos when you need a bounded answer rather than a generic promise about AI coding. The 8 articles separate observed evidence, working assumptions, product boundaries, and decisions that still belong to a human reviewer. Each guide is dated and keeps its source limitations visible.
Start with “AI Coding Verification” for the broadest entry point, then use “AI Session Handoffs” when you need the collection's more specific edge. The cards below state what each article covers, so you can choose by task instead of reading a manufactured sequence. Related links inside each guide connect methods, risks, evidence, and next actions without treating one check as universal proof.
Check each article's publication date, cited source, and stated scope before applying it to a live repository. Examples explain a method; they do not replace your project rules, threat model, tests, or accountable reviewer. If two guides appear to conflict, compare their assumptions and evidence rather than selecting the more confident wording.
Method
AI Coding Verification
How teams check AI-generated code against explicit task intent, validation plans, tests, and evidence before a change is accepted - the verification loop, step by step.
Updated: July 17, 2026Read article →
Method
Spec-vs-Implementation Check
Verify AI-generated code against a written statement of intent instead of your memory of the prompt - the five steps, the circularity problem it solves, and where the method ends.
Updated: August 15, 2026Read article →
Method
Code Review vs. Verification
Review judges quality, verification checks a change against written intent - why AI speed broke review-only workflows, what the data shows, and the division of labor that works.
Updated: August 6, 2026Read article →
Method
Machine-Checkable Specifications
Turn prompts into verifiable tasks: goal, boundaries, yes/no acceptance criteria, validation plan - the four building blocks and the rules that make criteria checkable.
Updated: July 17, 2026Read article →
Method
Measuring Verification Debt
Four metrics computable from git and PR data - generation-to-verification ratio, review depth, unverified-merge rate, two-week churn - with formulas, starting thresholds, and a worked example.
Updated: August 15, 2026Read article →
Method
Spec-Driven Development
Spec first, then code: how SDD works with AI agents, what Spec Kit and Kiro actually deliver, and the honest limits practitioners report - including why specs still need verification.
Updated: July 17, 2026Read article →
Show 2 more articles
- Two-Pass Review WorkflowMachine pre-check first, human architecture review second: what belongs in each pass, how to keep the machine gate high-precision, and why the human always makes the merge call.
- AI Session HandoffsSessions forget - compaction drops details, new sessions start cold. Write state, decisions, and open verification points into a persistent handoff artifact: the method, a template, its limits.
Want to see what your last agent run would have looked like?
Request access