An Empirical Comparison of Multi-Agent LLM Architectural Patterns for Automated Unit Test Generation

Sammanfattning

Software engineers are increasingly adopting multi-agent large language model (LLM)systems, yet the controlled empirical evidence comparing the underlying architectural patterns’ trade offs on software engineering tasks is still lacking. This thesisapplies a two phase mixed-methods design. The first phase conducts a structuredreview of 31 articles published between 2023 and 2026, resulting in a taxonomy of10 coordination patterns for multi-agent LLM systems. In the second phase, an empirical experiment comparing three patterns a Single-Agent Baseline, a Sequentialarchitecture, and a Hierarchical architecture is carried out on automated unit testgeneration using the TestEval benchmark. Five dependent variables are evaluated(success rate, error handling, latency, cost, and agent communication turns) running each architecture independently on 30 tasks, all using Claude Sonnet 4, for 270task executions in total. The Sequential pattern recorded the highest success rate(92.2%), the lowest latency variance (295.56 s2), and a competitive cost (0.089 USDper task). The Single-Agent Baseline reached a moderate success rate (81.1%) atthe lowest cost (0.083 USD per task), while the Hierarchical pattern recorded thelowest success rate (54.4%), the highest median latency (142.37 s), and the highestmean cost (0.308 USD per task). The hierarchical pattern suffered from supervisorinformation bottleneck according to the qualitative results confirmed by 100% ofthe failing tasks of the architecture. These results come from a single LLM, a singlebenchmark, and non-optimized prompts per architecture, which means replicationis needed before these findings are treated as general architectural principles.