Editorial illustration for AgentCo-op imports and refines searched workflows via component grounding
AgentCo-op imports and refines searched workflows via...
Forget building complex AI workflows from scratch. A team of researchers has a better idea: start with a blueprint. Their new system, AgentCo-op, first searches for an existing workflow structure to use as a skeleton.
Then it gets practical, hunting down the specific tools, data, or specialized agents needed to flesh out each step, fixing problems along the way. This hybrid approach—retrieval-based synthesis—topped the charts. In tests across six standard benchmarks for coding, math, and question-answering, it delivered the best results on four of them.
It cut costs, too.
AgentCo-op can also import a searched workflow as a structural prior and improve it by grounding nodes with retrieved components and applying local repair, showing that synthesis and search are complementary. On six coding, math, and question-answering benchmarks, AgentCo-op achieves the best result on four benchmarks and the best average score under a unified backbone setting, while consistently reducing per-task cost relative to multi-agent baselines. Together, these results suggest that retrieval-based synthesis can extend automated agentic workflow design beyond benchmark-optimized agent graphs to open-world workflows built from existing agents, tools, and typed artifacts.
The implications are laid out in a paper on arXiv. Benchmarks are one thing; real utility is another. The researchers argue their technique is a necessary shift toward practical systems that pull from a living library of existing agents and tools. That's the open-world goal.
Related Reading
AI Tools & Apps Claude gains shared context in Excel, PowerPoint; Microsoft adds Copilot Cowork
AI Tools & Apps Windows Copilot AI unable to pinpoint image source in user test
AI Tools & Apps LG's recent webOS update adds Microsoft Copilot app, now removable
From Archives Google tests SaaSpocalypse: Gemini to suggest apps and 450K titles
From Archives AI Agent in EOS Automates Validation, Error Correction for Full Lab Cycle
Common Questions Answered
How does AgentCo-op's retrieval-based synthesis approach differ from building AI workflows from scratch?
AgentCo-op uses an existing workflow structure as a skeleton blueprint rather than constructing workflows entirely from scratch. The system then searches for and grounds specific tools, data, and specialized agents needed for each step, while automatically fixing problems that arise during implementation. This hybrid approach proved superior across six standard benchmarks compared to traditional workflow creation methods.
What specific benchmarks did AgentCo-op perform well on according to the research?
AgentCo-op was tested across six standard benchmarks covering coding, math, and question-answering tasks. The system topped the charts on these benchmarks, demonstrating its effectiveness across diverse problem domains that require different types of specialized agents and tools.
What is the component grounding process in AgentCo-op?
Component grounding in AgentCo-op refers to the process of hunting down and integrating the specific tools, data, or specialized agents needed to flesh out each step of a workflow template. This process includes identifying the right components for each workflow stage and fixing any problems that arise during integration to ensure the workflow functions properly.
Why do the researchers argue that AgentCo-op represents a necessary shift toward practical AI systems?
The researchers contend that AgentCo-op's technique is necessary because it enables practical systems that can pull from a living library of existing agents and tools rather than requiring custom development for each workflow. This open-world goal makes AI workflow creation more efficient and accessible by leveraging previously developed components and solutions.
Further Reading
- Retrieval-Based Synthesis of Interoperable Multi-Agent Workflows — arXiv
- Grounding with Search - Agent Development Kit (ADK) — Agent Development Kit
- Grounding with Agent Search | Gemini Enterprise Agent Platform — Google Cloud Docs
- RAG and grounding on Vertex AI — Google Cloud Blog
- Agentic workflows: Getting started with AI Agents — Weights & Biases
