A new AI workflow framework automatically selects, coordinates and optimises multiple models, reducing computing costs while improving response speed and maintaining output quality.
Researchers at the Harvard John A. Paulson School of Engineering and Applied Sciences (SEAS) have developed Orla, an artificial intelligence workflow framework that simplifies the development and execution of complex AI applications. Instead of requiring engineers to manually choose and coordinate multiple AI models, Orla automatically determines the most suitable execution strategy while optimising workflows for cost, accuracy and response time. The framework is designed to make advanced AI systems easier to build, deploy and scale.
Modern AI applications increasingly rely on multiple specialised models working together rather than a single large language model. Selecting the right model for each task and coordinating them efficiently has become a major engineering challenge. Orla addresses this by allowing developers to describe the desired workflow while the framework automatically assigns tasks, manages execution and balances performance.
The research builds on earlier work in AI scheduling, load balancing and memory management. According to the team, the framework applies these optimisation techniques to complete AI workflows instead of focusing on individual models. This enables AI agents to collaborate more efficiently while reducing unnecessary computational overhead.
During evaluation, Orla demonstrated lower computing costs and faster response times without compromising answer quality. The researchers believe the framework could make sophisticated AI systems more practical for enterprise deployments, particularly as organisations adopt increasingly complex multi-agent applications.
The team expects automation frameworks such as Orla to play an important role in the next generation of AI software by reducing engineering complexity and improving operational efficiency. The findings were presented at the ACM Conference on AI and Agentic Systems (CAIS 2026), highlighting a shift towards intelligent workflow orchestration that can simplify the deployment of large-scale AI applications while maintaining reliable performance.
