Generative AI

3 min read Last updated Tue Jun 09 2026 03:05:56 GMT+0000 (Coordinated Universal Time)

Automates significant portions of SE phases traditionally driven by human cognitive effort. The engineer’s role shifts from producer to supervisor and validator.

Impact by Phase

Requirements

AI can process stakeholder inputs at scale and generate candidate requirement sets. Accelerates elicitation and reduces omissions.

Requirements must still be precisely specified as AI input.

Design

AI-assisted tools generate architectural alternatives from high-level specifications, augmenting model-driven approaches. Generated architectures must still be validated for non-functional properties.

xUML Integration

Executable UML is constrained to formal model inputs. Natural language specifications cannot be processed directly. Generative AI addresses this by translating natural language or semi-formal specifications into well-formed xUML models.

3-layer architecture:

  • AI-Mediated Requirements and Modelling Layer
    Generative AI accepts natural language requirements and produces xUML models. Handles requirement interpretation, ambiguity resolution, and OCL constraint generation.
  • xUML Model Validation and Transformation Layer
    Generated models pass through a validation engine checking formal consistency, completeness, and constraint satisfaction. A model-to-code transformer applies PSM translations to produce deployable code.
  • Automated Build, Test, and Deployment Layer
    A CI/CD pipeline executes AI-generated test suites derived from the use cases in the original xUML models and deploys the validated system. No human intervention required. Governance checkpoints are enforced programmatically.

Extends MDA by replacing the human modeller with a generative AI agent, enabling end-to-end automation from natural language intent to deployed software.

Implementation

Code-generation tools drastically reduce manual coding effort. Generated code still requires correctness verification.

Testing

AI generates test cases from specifications, improving coverage without proportional human effort. Risk of subtle defects in AI-generated code makes testing discipline more critical, not less.

Repositioning of SE Methods

Generative AI does not eliminate SE methods. It repositions them.

New responsibilities introduced:

  • Validation of AI-generated artefacts
  • Prompt engineering as a formal skill
  • Governance of AI-assisted processes
  • Quality gates suitable for AI-produced outputs

AI in the Spiral Model

AI augments each spiral iteration across all four sectors.

Applications:

  • Risk prediction
    Machine learning models identify high-risk areas from historical project data during risk assessment.
  • Effort estimation
    AI-assisted estimation during objective setting improves forecast accuracy.
  • Code and test generation
    Generative tools accelerate the development and validation sector.
  • Defect detection
    Static analysis and ML-based reviewers catch defects before formal testing.
  • Decision support
    AI summarizes iteration outcomes and recommends planning adjustments for the next spiral.

Benefits:

  • Reduced risk exposure per iteration.
  • Improved planning accuracy over successive spirals.
  • Faster development cycles.

Future Practice

Hybrid model: generative AI handles high-volume, pattern-based generation. Human engineers exercise judgment, ethical oversight, and correctness assurance.

SE methods that fail to integrate AI tooling become inefficient. SE methods that do integrate AI must develop new validation disciplines.

Was this helpful?