← AI-Driven Development

Comparison

Not a replacement for existing practices. A different scope.

Established Methodologies

TDD - complementary. TDD tests code units. Journeys test end-to-end arcs. TDD has automated regression, journeys currently do not. Use both.

BDD - closest ancestor. BDD defines behavior in Given-When-Then format and automates as acceptance tests. This methodology borrows “define behavior from the user’s perspective” but diverges in execution: BDD verifies the response, this also verifies the internal execution path via traces.

ATDD - adjacent. ATDD defines acceptance criteria before implementation. Evaluation criteria here serve the same purpose, but are checklists applied during walks with trace analysis as an additional verification dimension.

Exploratory Testing - spiritual ancestor. A skilled tester explores the system using judgment. This methodology is exploratory testing where the explorer is an AI agent, the exploration follows a defined journey, and the judgment is replaced by human-defined criteria.

Adjacent Approaches

SDD (Spec-Driven Development) - opposite direction. SDD workflow: Spec to Code. This methodology: Code to Walk to Verify. SDD is a code generation discipline. This is a code verification discipline. They can coexist.

Vibe Coding - opposite end of the spectrum. AI generates code, developer accepts without thorough review. This methodology exists because vibe coding works, and that is the problem. These are not competing approaches: vibe coding describes how code gets written, this describes how code gets verified.

The research supports the need: CodeRabbit found AI co-authored code has 1.7x more major issues. Veracode found AI-generated code security has not improved despite functionality gains. METR found experienced developers are 19% slower with AI tools while believing they are faster. A verification discipline is not optional.

Positioning

DimensionTDDBDDSDDVibeThis
VerifiesCode unitsUser behaviorN/AN/AEnd-to-end arcs
OracleAssertionsAcceptance testsN/AHuman intuitionOTel traces + criteria
Executed byDeveloperTest runnerAI agentAI agentAI agent
DirectionTest to CodeBehavior to CodeSpec to CodePrompt to CodeCode to Walk to Verify
AutomatedYesYesN/AN/ANot yet

Closest compound description: AI-driven acceptance testing with distributed trace verification through the product’s callable surface.


Full comparison with detailed analysis →