CORPGEN: Simulating Corporate Environments with Autonomous Digital Employees in Multi-Horizon Task Environments

2026-02-15

AI Summary

This work addresses the challenge of sustaining efficient task completion in realistic organizational environments characterized by long-horizon, concurrently executed tasks with dynamically evolving dependencies and priorities. To this end, we introduce Multi-cycle Task Environments (MHTEs), which simulate digital employees endowed with persistent identities and authentic schedules to collaboratively manage dozens of interwoven long-duration tasks. Our key innovations include hierarchical planning to align multi-cycle objectives, sub-agent isolation to prevent cross-task interference, and a three-tier memory architecture—comprising working, structured, and semantic memory—augmented with adaptive summarization to effectively mitigate context saturation, memory interference, and rescheduling overhead. Evaluated in the OSWorld Office environment, our approach achieves a task completion rate up to 3.5× higher than the baseline (15.2% vs. 4.3%), maintains stable performance under increasing workload, and ablation studies reveal that experience-based learning contributes most significantly to this improvement.

Technology Category

• Multiagent Systems: Multiagent Planning
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Cognitive Modeling & Cognitive Systems: Agent Architectures

Application Category

• Economics, Online Markets and Human Computation: Architectures and workflows that use LLMs for crowd work
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Agentic search

Abstract

Long-horizon reasoning is a key challenge for autonomous agents, yet existing benchmarks evaluate agents on single tasks in isolation. Real organizational work requires managing many concurrent long-horizon tasks with interleaving, dependencies, and reprioritization. We introduce Multi-Horizon Task Environments (MHTEs): a distinct problem class requiring coherent execution across dozens of interleaved tasks (45+, 500-1500+ steps) within persistent execution contexts spanning hours. We identify four failure modes that cause baseline CUAs to degrade from 16.7% to 8.7% completion as load scales 25% to 100%, a pattern consistent across three independent implementations. These failure modes are context saturation (O(N) vs O(1) growth), memory interference, dependency complexity (DAGs vs. chains), and reprioritization overhead. We present CorpGen, an architecture-agnostic framework addressing these failures via hierarchical planning for multi-horizon goal alignment, sub-agent isolation preventing cross-task contamination, tiered memory (working, structured, semantic), and adaptive summarization. CorpGen simulates corporate environments through digital employees with persistent identities and realistic schedules. Across three CUA backends (UFO2, OpenAI CUA, hierarchical) on OSWorld Office, CorpGen achieves up to 3.5x improvement over baselines (15.2% vs 4.3%) with stable performance under increasing load, confirming that gains stem from architectural mechanisms rather than specific CUA implementations. Ablation studies show experiential learning provides the largest gains.

Problem

• long-horizon reasoning
• multi-horizon task environments
• autonomous agents
• task interleaving
• dependency management

Innovation

• Multi-Horizon Task Environments
• Hierarchical Planning
• Sub-agent Isolation
• Tiered Memory
• Adaptive Summarization