MOSAICO

Generative AI based on Large Language Models (LLMs) is increasingly being applied to software engineering (SE) tasks, with very promising results.1 As assistants in software development, such models are very fast (after training) and relatively cheap, but they can also be highly unreliable and possibly biassed. The lack of reliability hampers the general applicability of generative AI to (possibly critical) SE. Current research is mainly focused on improving the reliability of the AI assistance by traditional software verification and validation methods. On the other hand, with the landscape of AI quickly evolving, we witness a rapid increase in the variety of high-quality accessible models (e.g., more than 40 independent LLMs are available at present). We anticipate that the cost of accessing such models will further decrease in time, as open and self-hosted solutions improve and become widespread. At the same time, the emergence of Small Language Models2 is reducing the energy footprint of generative AI in several tasks.

MOSAICO’s vision is for the future of SE is a reliable application of generative AI to SE tasks, enabled by the coordinated and supervised collaboration of – a possibly large number of – different LLM-based AI agents, that we call a Community of AIs (or AI-agent community). This cooperation will need precise communication among AI agents in all phases of the SE process. Software modelling languages are precise and uniform descriptions of software in all its life-cycle, and have been historically designed as communication tools among software engineers. They are the most natural candidates for exchanging artefacts in the communication among AI agents for SE, and between agents and human engineers.

MOSAICO aims at producing a holistic methodology and a set of solutions for the engineering and operation of communities of AIs across the SE life-cycle. The solutions will be composed into an integrated MOSAICO platform, handling communication, orchestration, governance, quality assessment, benchmarking and reuse of AI agents. MOSAICO will be integrated with existing development environments, to present the results to software engineers, and allow expert users to intervene in the AI decisions.

Grant Number: 101189664

This project is running from Jan 2025 - December 2027.

Consortium

• Institut Mines-Telecom - FR (coordinator)
• Luxembourg Institute of Science and Technology - LU
• University of York - UK
• Universita Degli Studi Dell’Aquila - IT
• Eclipse Foundation - DE
• NetCompany-Intrasoft SA - LU
• Immersion - FR
• National Bank of Greece SA - EL
• Advanced Laboratory on Embedded Systems SRL - IT
• Unparallel Innovation LDA - PT
• Codium LTD - IL
• F6S EU Tech Innovation Network Designated Activity Company - IE
• … 12 partners