ARCHITECTURE · AI-ASSISTED ENGINEERING
SDLC of AI-Assisted Development
Controlled Software Development Lifecycle
A specification-driven approach to engineering work performed with AI coding assistants and agents.
An AI coding assistant may propose a change, but proposal generation is not evidence that the change was incorporated, tested, or accepted. This architectural work addresses the gap between suggested implementation and verified engineering state.
Distinct engineering states
Requirement or defect identified; solution proposed; change approved; code actually incorporated into authoritative source; regression tests executed; resulting evidence reviewed; engineering acceptance recorded.
Runners, harnesses, and controls
A runner executes and coordinates a workflow. A harness establishes workflow constraints, checks, recovery paths, and supporting context. The governing requirements and acceptance criteria must remain independently identifiable.
A governance compilation model can constrain the proposals an AI system may make, while deterministic execution controls can independently govern consequential commits.
Failure and recovery
Relevant failure modes include partial fixes, context loss, mismatched specifications, unverified claims of completion, stale evidence, and regression after apparently successful changes.
Scope
This is an architectural and engineering methodology; the page does not claim a generally available production product or a specific benchmark result.