The paper presents Qwen-Planner-Agent, a mobile planning agent developed through a closed-loop AI-for-AI framework. Specialized AI agents generate training data, train the model via supervised learning and reinforcement learning, and coordinate a runtime harness with memory and skills, with execution feedback feeding back into model and harness improvements. On MobilePA-Bench the agent achieved the best overall performance among all evaluated models and systems, improving over its base model across tool use, memory, skills, and sub-agent coordination, while also improving on non-mobile agentic benchmarks and largely preserving general capabilities.
Closed-loop development improves mobile agents across planning and tool use
The authors connect agent-generated data, reinforcement learning, and runtime feedback to jointly improve a mobile planning model and its execution harness.
Independent
Tingyu Qu · Weigao Sun · Yuecheng Liu · Yucheng Zhao · Yi Zhu · Yifeng Ding · +20 more
Research Digest··2 min read
Thread:Agent Harness Optimization
In one line
Qwen-Planner-Agent uses AI agents to build and improve a mobile planner agent, outperforming prior systems on MobilePA-Bench.
Before you read this
Three ideas this paper builds onagentic reinforcement learningTraining approach that lets LLM-based agents learn multi-step tool use and planning via reinforcement learning, often with dense or per-tool rewards.120 papers →recursive self-improvementAn AI system repeatedly uses its own outputs to enhance its capabilities, creating a feedback loop for autonomous improvement.61 papers →harness evolutionIteratively adapting an agent's environment, prompts, or evaluation setup to improve task performance and generalization.66 papers →
Read it as
Qu et al. developed Qwen-Planner-Agent using a human-gated pipeline in which AI systems help create training data, optimize the planner, and refine the runtime harness.
The authors built a closed-loop development framework around a shared action–feedback–verification contract.
How it works
before and afterThe Daily Edition
The front page in your inbox each weekday morning: the lead, the picks, and why each one earned its place.