Ake Sjursen Hauge
• Eirik Varnes
abstract
Our goal is to improve AI-driven development in large codebases—where
a single agent often loses context and hallucinates. We are solving this with
a multi-agent system that automates generation, review, and testing. To
achieve this, we will compare two state-of-the-art tools for building multi-
agent systems: the OpenAI Agent SDK and Google’s ADK. We will evaluate
both platforms in terms of capabilities, features, and ease of integration, then
choose the one best suited for our goals. We will use Python for scripts, JSON
for message schemas, virtual environments (venv) with pip for dependency
management, and Docker for deploying and testing our agents.
outcomes