This repository contains code implementations from the paper "Barbarians at the Gate: How AI is Upending Systems Research" (arXiv:2510.06189).

Some examples are still work in progress (WIP) and code may be private — we're actively updating and releasing them. Stay tuned!

Make sure you have uv installed.

uv sync
export OPENAI_API_KEY="..."
export GEMINI_API_KEY="..."

This repository showcases AI-driven research examples across multiple frameworks and domains:

• Location: openevolve/examples/ADRS directory
• Collection of AI-driven systems tasks including:
  • MoE expert placement
  • Global model scheduling (PRISM)
  • Transaction scheduling
  • Telemetry repair
  • LLM-SQL optimization
  • Spot instance scheduling for single region
  • Spot instance scheduling for multi-region
  • Multi-region data transfer (Cloudcast)
  • Sparse attention design
  • Multi-agent system design
  • HP Quantization

Note: Check out the README inside each folder for more details about setup and usage.

python -m openevolve.cli \
  openevolve/examples/ADRS/<case>/initial_program.py \
  openevolve/examples/ADRS/<case>/evaluator.py \
  --config openevolve/examples/ADRS/<case>/config.yaml \
  --iterations <N> \
  --output openevolve/examples/ADRS/<case>/output

• Location: cursor/ directory

• Location: gepa/ directory