AOrchestra: Automating Sub-Agent Creation for Agentic Orchestration
Jianhao Ruan, Zhihao Xu, Yiran Peng, Fashen Ren, Zhaoyang Yu, Xinbing Liang, Jinyu Xiang, Bang Liu, Chenglin Wu, Yuyu Luo, Jiayi Zhang
2026-02-04
Summary
This paper introduces a new system called AOrchestra designed to make AI agents better at handling complex tasks that require multiple steps, like completing projects with many different parts.
What's the problem?
Current AI agents that use smaller 'sub-agents' as tools to solve problems aren't very flexible. They struggle to adapt when faced with new situations because they don't have a good way to understand and change how those sub-agents work. Essentially, they treat these sub-agents as fixed entities instead of being able to dynamically create specialized ones as needed.
What's the solution?
The researchers created a way to describe any AI agent using four key components: what it's *told* to do (Instruction), the information it has access to (Context), the tools it can use (Tools), and the underlying AI model powering it (Model). AOrchestra uses this description to automatically build and launch the best sub-agent for each specific part of a larger task, changing them as the task evolves. It's like having a flexible toolkit where you can assemble the right tool for each job on the fly.
Why it matters?
This work is important because it reduces the amount of manual work needed to create effective AI systems. Instead of painstakingly designing each agent, AOrchestra can adapt and create them automatically. It also allows for a balance between how well the system performs and how much it costs to run, making it more practical for real-world applications. The system showed significant improvements in performance on several challenging tests, demonstrating its potential.
Abstract
Language agents have shown strong promise for task automation. Realizing this promise for increasingly complex, long-horizon tasks has driven the rise of a sub-agent-as-tools paradigm for multi-turn task solving. However, existing designs still lack a dynamic abstraction view of sub-agents, thereby hurting adaptability. We address this challenge with a unified, framework-agnostic agent abstraction that models any agent as a tuple Instruction, Context, Tools, Model. This tuple acts as a compositional recipe for capabilities, enabling the system to spawn specialized executors for each task on demand. Building on this abstraction, we introduce an agentic system AOrchestra, where the central orchestrator concretizes the tuple at each step: it curates task-relevant context, selects tools and models, and delegates execution via on-the-fly automatic agent creation. Such designs enable reducing human engineering efforts, and remain framework-agnostic with plug-and-play support for diverse agents as task executors. It also enables a controllable performance-cost trade-off, allowing the system to approach Pareto-efficient. Across three challenging benchmarks (GAIA, SWE-Bench, Terminal-Bench), AOrchestra achieves 16.28% relative improvement against the strongest baseline when paired with Gemini-3-Flash. The code is available at: https://github.com/FoundationAgents/AOrchestra