From Prompt to Process: a Process Taxonomy and Comparative Assessment of Frameworks Supporting AI Software Development Agents
Summary
AI software development is evolving beyond simple autocomplete to structured frameworks that integrate process, roles, artifacts, and verification. A new study, published June 3, 2026, introduces a six-dimension process taxonomy to assess these frameworks, including GitHub Spec Kit, OpenSpec, BMAD Method, Get Shit Done (GSD), Spec Kitty, and Reversa. The taxonomy covers specification, context, roles, execution, validation, and portability. Key findings indicate a convergence among frameworks towards persistent artifacts, work contracts, traceability, and human review to reduce ambiguity. However, a structural trade-off exists between process depth and portability across agents, with no single framework strongly covering all six dimensions. The research also identifies recurring risks such as specification-code drift, over-reliance on generated artifacts, and platform dependence, proposing an agenda for empirical evaluation.
Key takeaway
For AI Engineers evaluating AI software development frameworks, prioritize solutions that integrate structured processes, persistent artifacts, and human review. You should assess frameworks based on their ability to manage context, define roles, and provide robust validation mechanisms beyond final tests. Be aware of the trade-off between process depth and portability, and consider the supply chain risks of community extensions. Your focus should shift from mere code generation to comprehensive process governance.
Key insights
AI software development frameworks shift from isolated prompts to structured processes with artifacts, roles, and validation, balancing autonomy and governance.
Principles
- Process depth often trades off with portability.
- Context is a critical engineering asset.
- Intermediate validation is crucial for agentic flows.
Method
A comparative study using a six-dimension process taxonomy (specification, context, roles, execution, validation, portability) and a scoring rubric, applied to six selected frameworks and an out-of-sample case.
In practice
- Implement persistent artifacts and work contracts.
- Integrate human review points into agentic workflows.
- Prioritize context assembly as an explicit task.
Topics
- AI Software Development
- AI Agents
- Agentic Frameworks
- Process Taxonomy
- Specification-Driven Development
- Software Engineering
Code references
Best for: Research Scientist, AI Scientist, AI Engineer, MLOps Engineer
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by cs.SE updates on arXiv.org.