New Protocol Language Defines Human-Agent Boundaries in AI Software Development

Ylli Prifti's paper, 'Specifying AI-SDLC Processes: A Protocol Language for Human-Agent Boundaries,' addresses a critical gap: as AI agents become first-class team members in software development, no specification language exists for expressing human-agent responsibility boundaries, approval gates, and governance constraints. Existing approaches rely on fragile agent prompts or target adjacent domains (workflow management, access control), but none provide a unified formal framework. Prifti proposes a domain-specific language (DSL) with formal syntax, well-formedness conditions, operational semantics, and enforcement invariants. The language explicitly distinguishes policy (declared intent) from mechanism (structural enforcement), enabling implementations to bound process non-determinism using primitives like validation tokens and capability boundaries.

Three key results emerge from the formal analysis. First, a failure rate analysis shows that structural enforcement bounds system failure rates at a weighted product of agent and validator rates, whereas behavioral compliance permits cumulative or near-saturating growth—a significant safety advantage. Second, the 2+N team pattern formalizes classical Separation of Duties for AI-SDLC, specifying two human-in-control roles plus N specialized agent members. Third, Kleene closure of orchestration loops and reflexive protocol-adherence validation emerge as design properties rather than special-case constructs. Prifti positions this against multi-agent frameworks (MetaGPT), workflow specifications (FlowAgent, BPMN extensions), and capability-based security (SAGA), noting the novelty lies in their specific integration. A working implementation demonstrates feasibility, with empirical evaluation planned as future work.