AI BriefingCORPGEN Advances AI Agent Technology for Real-World Work Environments
Key point
Microsoft has unveiled **CORPGEN**, an AI agent framework capable of performing complex multitasking work.
Details
Existing AI agent benchmarks focus on single-task testing, failing to reflect real work environments where multiple tasks progress with interdependencies. To address this, Microsoft researchers built Multi-Horizon Task Environments (MHTEs), environments that require managing multiple complex tasks simultaneously.
Test results showed that existing leading agents have a limitation where task completion rates sharply drop from 16.7% to 8.7% as workload increases. The main causes identified were memory overload, interference between tasks, complex dependencies, and difficulty in continuously re-prioritizing.
CORPGEN introduces the concept of digital employees to solve these problems. They are equipped with hierarchical planning, memory isolation, and experiential learning capabilities, and operate through the following core structures.
- Hierarchical planning: Breaks down goals into daily objectives and detailed decisions to support systematic execution.
- Memory isolation: Allows sub-agents to work in independent contexts, preventing interference between tasks.
- Tiered memory system: Selectively recalls only necessary information to prevent memory overload.
CORPGEN adopts an architecture-agnostic and modular design that is not tied to a specific model. Through this, it achieved a task completion rate up to 3.5x higher than existing base models, and is structured so that performance can further improve as base models advance.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.