NVIDIA's NOOA paper: agents as plain Python objects, combining six model-facing ideas in one surface
omarsar0 · x · 2026-10-21
NVIDIA's new paper presents NOOA (Object-Oriented Agents), a model-agnostic Python framework built on one idea: an agent is a Python object. Methods are the actions the model can take, fields hold state, docstrings are prompts, and type annotations act as contracts.
Key points:
- The boundary lives in the source: a method whose body is "..." is completed at runtime by a validated LLM loop, while normally-bodied methods stay deterministic Python — the probabilistic/deterministic line is readable right in the code
- The paper claims the first combination of six model-facing ideas: typed input/output, pass-by-reference over live objects, code as action, programmable loop engineering, explicit object state, and model-callable harness APIs for context and events
- Evaluated on SWE-bench Verified, Terminal-Bench 2.0, and ARC-AGI-3; on ARC-AGI-3 it compresses a multi-agent world-model system into a single agent with a one-page skill while advancing the score-cost Pareto frontier
- Why it matters: agents and developers share one programming model, so agent behavior becomes testable, traceable, and refactorable with existing repo tooling — a more realistic reliability path than another orchestration DSL
Inspired by PyTorch: a powerful runtime can still present a simple Python programming model.
More from coding & agent
- EAU: Co-evolving user simulator, environment, and RL setup for training LLM agents — UAlberta · 2026-10-21
- Meta's Self-Credit: Dense Rewards from Sparse RLVR via Intra-group Rescaling — meta · 2026-10-21
- Notion details its Auto model router: Elo-based evals, 62-93% cheaper than flagship models — ivanhzhao · 2026-10-21
- Tamper-proof agent logs still can't prove the agent did the job: who reads the record back? — ZealousidealAct9661 · 2026-10-21
- One prompt: Claude builds an animated mech spider in Blender via MCP, first try — sidahuj · 2026-10-21
- OpenAI Details Coordinated Disclosure of Critical Vulnerabilities Found by Agents — deliciousmark12 · 2026-10-21