From Prompt to Process: a Process Taxonomy and Comparative Assessment of Frameworks Supporting AI Software Development Agents
Researchers conducted a comparative study of six AI software development frameworks—GitHub Spec Kit, OpenSpec, BMAD Method, GSD, Spec Kitty, and Reversa—revealing a structural trade-off between process depth and portability. The analysis identified a taxonomy across six dimensions (specification, context, roles, execution, validation, portability) and found that successful frameworks increasingly rely on persistent artifacts, work contracts, and human review rather than isolated prompts.
The academic research addresses a critical gap in understanding how AI development tools operate at the process level rather than merely as isolated capabilities. As AI agents become integral to software engineering workflows, the study reveals that frameworks are converging around common mechanisms: specification-driven development, persistent artifact management, and mandatory human verification checkpoints. This shift from single-prompt interactions to structured development processes reflects industry maturation.
The six-dimension taxonomy provides the first systematic framework for evaluating AI development agents across comparable metrics. The research identifies a fundamental constraint: no existing framework adequately covers all dimensions, forcing practitioners to choose between operational richness and cross-platform compatibility. This has direct implications for tool adoption—teams must either accept incomplete coverage or accept vendor lock-in through platform-specific implementations.
The recurring risks identified—specification drift, over-reliance on generated artifacts, fragility of extensions, and platform dependence—create enterprise adoption barriers. Without standardized benchmarks for complete processes, organizations struggle to predict outcomes and validate quality. The lack of intermediate-quality metrics means teams cannot distinguish between frameworks objectively. These limitations delay enterprise adoption of AI development agents beyond individual use cases into production pipelines.
The research agenda emphasizing context governance, installation security, and reproducibility signals that AI development infrastructure requires governance layers comparable to traditional software deployment. This emerging complexity may drive demand for specialized platforms and consulting services focused on process standardization and risk mitigation.
- →Successful AI development frameworks increasingly prioritize persistent artifacts and human review over isolated prompt interactions
- →Fundamental trade-off exists between process depth and cross-agent portability with no framework achieving optimal coverage across all six dimensions
- →Specification drift, excessive trust in generated code, and platform dependence represent recurring structural risks in current frameworks
- →Enterprise adoption requires standardized benchmarks and intermediate-quality metrics that do not yet exist
- →Future AI development infrastructure will require governance layers for context management, security, and reproducibility