Can AI Agents Generate Microservices? How Far are We?

A new research paper accepted at the 2026 IEEE International Conference on Software Architecture (ICSA 2026) reveals significant progress in AI-generated microservices. The study by Bassam Adnan, Matteo Esposito, Davide Taibi, and Karthik Vaidhyanathan tested 144 generated microservices across 3 different AI agents, 4 projects, and 2 prompting strategies. The research examined both incremental generation (working within existing systems) and clean-state generation (starting from requirements alone), evaluating them with unit tests and integration tests respectively.

Clean-state generation proved surprisingly effective, achieving 81-98% integration test pass rates, indicating strong adherence to API contracts. However, incremental generation performed less consistently with 50-76% unit test pass rates. The researchers found that minimal prompts actually outperformed detailed ones in incremental scenarios. While generated code showed lower complexity than human baselines, generation times varied widely from 6-16 minutes per service, and the study concluded that fully autonomous microservice generation remains out of reach due to inconsistent correctness and the need for human oversight.