AOrchestra: Automating Sub-Agent Creation for Agentic Orchestration

Language agents have shown strong promise for task automation. Realizing this promise for increasingly complex, long-horizon tasks has driven the rise of a subagent-as-tools paradigm for multi-turn task solving. However, existing designs still lack a dynamic abstraction view of sub-agents, thereby hurting adaptability: sub-agents are either context-isolated threads that lack specialization, or static roles that require human-engineering. We address this challenge with a unified, framework-agnostic agent abstraction that models any agent as a tuple (Model, Task, Tools, Context). 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 and environments (GAIA, SWE-Bench, Terminal-Bench), AOrchestra achieves 16.28% relative improvement against the strongest baseline when paired with Gemini-3-Flash.

Modern AI assistants are increasingly asked to tackle long, complicated tasks, such as fixing a software bug, searching across websites and files, or using tools to answer a multi-step question. These tasks often strain a single all-purpose agent. A common solution is to let a “manager” AI delegate work to “worker” AIs, but existing systems usually make those workers either predefined roles that are hard to adapt, or generic isolated threads that do not get the right expertise for each new subtask. We propose AOrchestra, a system where the manager creates a tailored worker whenever a new subtask arises. Each worker is specified using four ingredients: the instruction, the relevant background, the tools it can use, and the model that will run it. This lets the system create specialized workers on demand, without requiring users to manually design a fixed set of worker roles. On three challenging benchmarks for web, terminal, and software-engineering tasks, AOrchestra achieves about a 16% relative improvement over the strongest baseline when paired with Gemini-3-Flash. The system can also be further optimized from past interaction traces to improve success and choose cheaper model combinations, pointing toward more dependable AI assistants for long, real-world workflows.