Shift-Up framework proposes SE guardrails for AI-native development
A new research framework called Shift-Up reinterprets established software engineering practices — BDD, C4 modeling, and ADRs — as structural guardrails to reduce architectural drift and improve maintainability in GenAI-native, agent-driven development.
Score breakdown
Teams building with AI coding agents can use Shift-Up's approach of embedding BDD specs, C4 diagrams, and ADRs as machine-readable inputs to reduce agent drift and maintain architectural control without abandoning the speed benefits of agentic development.
- 01Shift-Up is a framework applying design science research (DSR) methodology to GenAI-native software development.
- 02Vibe coding is identified as suffering from architectural drift, limited traceability, and reduced maintainability.
- 03The framework repurposes BDD (executable requirements), C4 (architectural modeling), and ADRs (architecture decision records) as structural guardrails for AI agents.
Shift-Up is a proposed framework that responds to a growing tension in AI-native software development: while agent-driven "vibe coding" enables rapid prototyping, it tends to produce architectural drift, poor traceability, and code that is difficult to maintain. The paper applies design science research (DSR) methodology to reframe well-established software engineering practices as structural control mechanisms suited to GenAI workflows. Specifically, it positions executable requirements via Behavior-Driven Development (BDD), architectural modeling via the C4 model, and architecture decision records (ADRs) as machine-readable guardrails that constrain and guide agent behavior.
An exploratory evaluation compared three development approaches — unstructured vibe coding, structured prompt engineering, and the full Shift-Up framework — in the context of building a web application.
An exploratory evaluation compared three development approaches — unstructured vibe coding, structured prompt engineering, and the full Shift-Up framework — in the context of building a web application. Initial findings indicate that embedding machine-readable requirements and architectural artifacts reduces implementation drift and stabilizes agent outputs. Crucially, the framework also repositions human effort away from low-level coding tasks and toward higher-level design and validation activities, suggesting a meaningful division of labor between engineers and AI agents. The authors characterize these results as preliminary but indicative that traditional SE artifacts can serve as effective control mechanisms in AI-assisted development.
Key facts
- 01Shift-Up is a framework applying design science research (DSR) methodology to GenAI-native software development.
- 02Vibe coding is identified as suffering from architectural drift, limited traceability, and reduced maintainability.
- 03The framework repurposes BDD (executable requirements), C4 (architectural modeling), and ADRs (architecture decision records) as structural guardrails for AI agents.
- 04An exploratory evaluation compared unstructured vibe coding, structured prompt engineering, and the Shift-Up approach on a web application.
- 05Preliminary findings show that machine-readable requirements and architectural artifacts stabilize agent behavior and reduce implementation drift.
- 06The framework shifts human effort toward higher-level design and validation rather than manual coding.