cs.SEMar 16, 2026

Loosely-Structured Software: Engineering Context, Structure, and Evolution Entropy in Runtime-Rewired Multi-Agent Systems

Authors: Weihao Zhang, Yitong Zhou, Huanyu Qu, Hongyi Li

Organizations: The Hong Kong University of Science and Technology, Hong Kong SAR, China · AI Chip Center for Emerging Smart Systems, Hong Kong SAR, China · University of Macau, Macau SAR, China · Tsinghua University, China

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

As LLM-based multi-agent systems (MAS) become more autonomous, their free-form interactions increasingly dominate system behavior. However, scaling the number of agents often amplifies context pressure, coordination errors, and system drift. It is well known that building robust MAS requires more than prompt tuning or increased model intelligence. It necessitates engineering discipline focused on architecture to manage complexity under uncertainty. We characterize agentic software by a core property: runtime generation and evolution under uncertainty. Drawing upon and extending software engineering experience, especially object-oriented programming, this paper introduces Loosely-Structured Software (LSS), a new class of software systems that shifts the engineering focus from constructing deterministic logic to managing the runtime entropy generated by View-constructed programming, semantic-driven self-organization, and endogenous evolution. To make this entropy governable, we introduce design principles under a three-layer engineering framework: View/Context Engineering to manage the execution environment and maintain task-relevant Views, Structure Engineering to organize dynamic binding over artifacts and agents, and Evolution Engineering to govern the lifecycle of self-rewriting artifacts. Building on this framework, we develop LSS design patterns as semantic control blocks that stabilize fluid, inference-mediated interactions while preserving agent adaptability. Together, these abstractions improve the designability, scalability, and evolvability of agentic infrastructure. We provide basic experimental validation of key mechanisms, demonstrating the effectiveness of LSS.

Figures & tables

Figure 1. The engineering focus shifts from designing specific logical code to designing runtime generation and evolution.
Figure 2. Loosely-Structured Software: a software system with unstable, runtime-generated structure & abstraction and its dominant engineering focuses (View, Binding, Evolution).
Figure 3. Comparison of Architectural Paradigms: Traditional Software vs. LSS. Left: Traditional software relies on compile-time program structure to constrain runtime behavior. Right: LSS makes behavior depend on the runtime trajectory of Views and Artifacts, which induces the effective agent configuration over time.
Figure 4. Multi-agent interaction in LSS, represented using UML-style sequence diagram notation. The diagram illustrates how the primitives decompose the loosely-structured runtime.
Figure 5. Three-layer engineering framework for LSS: View–Context, Structure–Capability, and Evolution–Adaptation.
Figure 6. Sequence diagram of a Mediator constructing a task-specific A2A protocol between two agents.
Figure 7. Establishing a more stable structure and contract through flexible multi-agent negotiation.
Figure 8. Inheritance in LSS: single inheritance isolates a child agent’s context under a Context Curator; multiple inheritance selects and composes trajectory fragments across instances under an Inheritance Generator to integrate capabilities and memories.
Figure 9. The evolution engineering should accept a relative and unstable ephemeral/persistent Boundary.
Figure 10. Selected evolution patterns in Layer 3: Sandbox Mode evaluates changes before they enter the persistent artifact set, Semantic Palimpsest keeps changes traceable and rollback-capable, and Artifact Tiering manages artifact lifecycles by moving them across tiers with different read/write policies.
Figure 11. Sequence diagram of the Semantic Lens pattern illustrating Derived Execution, Collaboration, and Self-Derived Loops. The Worker Agent initiates the process, and the Lens Agent recursively derives new instances to handle each artifact, keeping the Worker’s context clean.
Figure 12. Evaluation results on RepoBench-R: (a) Retrieval success comparison, and (b) Average input-context tokens across variants.
Figure 13. Per-query total token traces across three variants.
Figure 14. Per-query Worker token traces across three variants.
Figure 15. The agent class architecture of task distribution and completion workflow in the knowledge-base-centric environment.
Figure 16. Illustrative per-round trace of task volume and token usage in the replay workflow.
More in this topic

Context Engineering

See all 12 papers

Explore similar work

CardsList

  1. Scaling LLM-Driven Multi-Agent Systems: Design Principles and Architectural Scalability Analysis

    Jul 30, 2026Linus Sander, Fengjunjie Pan, Vahid Zolfaghari +3Multi-Agent LLM SystemsMulti-Agent System Optimization

    Robotics, Artificial Intelligence and Real-Time Systems, School of Computation, Information and Technology, Technical University of Munich, Munich, Germany

  2. MAS-Lab: A Specification-Driven Validation Framework for Reliable Multi-Agent Systems

    Jun 29, 2026Jordan Augé, Giovanna Carofiglio, Giulio Grassi +1Multi-Agent LLM SystemsAutomated Software Testing

    Cisco Systems Paris, France

  3. AgentSoV: Virtual Sensing for LLM Multi-Agent Systems via Stream of Variation

    Oct 17, 2026Congjing Zhang, Siyuan Xu, Yunfei Bai +3Multi-Agent LLM SystemsMulti-Agent Systems

    Department of Industrial & Systems Engineering, University of Washington · Department of Electrical & Computer Engineering, Texas A&M University · Amazon