Aktagon Signals AI-generated & human-reviewed
Featured
arxiv.org 4 min read

Building Reliable AI Systems Through Multi-Agent Organizational Intelligence

This paper presents a multi-agent AI architecture that achieves 92.1% reliability by organizing specialized AI agents into teams with opposing roles and hierarchical oversight, similar to corporate organizational …

AI · Architecture Editorial Team
arxiv.org 3 min read

GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning

GEPA introduces a novel prompt optimization approach that uses natural language reflection and Pareto-based evolutionary search to optimize compound AI systems, achieving superior performance compared to reinforcement …

AI · Development Editorial Team
arxiv.org 3 min read

How AI Impacts Skill Formation: Evidence from Software Development Learning

A randomized controlled study examining how AI assistance affects skill formation in software development, finding that while AI can improve productivity, it significantly impairs conceptual understanding, code reading, …

AI · Development Editorial Team
Recent
arxiv.org 3 min read

Small Language Models are the Future of Agentic AI

This position paper from NVIDIA Research argues that small language models (SLMs) are sufficiently powerful, more operationally suitable, and more economical than LLMs for the majority of tasks in agentic AI systems. The …

AI · Architecture Editorial Team
arxiv.org 6 min read

Agents of Chaos: Red-Teaming Autonomous LLM Agents Reveals Critical Security and Safety Vulnerabilities

A two-week red-teaming study of autonomous LLM-powered agents deployed with persistent memory, email, Discord, and shell access uncovered eleven critical failure modes including unauthorized compliance, identity …

AI · Security Editorial Team
arxiv.org 3 min read

Security Threats and Defenses in LLM-Based Autonomous Agents: A Case Study of OpenClaw

This paper presents a comprehensive security analysis of OpenClaw, an open-source autonomous AI agent framework, introducing a tri-layered risk taxonomy covering AI cognitive, software execution, and information system …

Security · AI Editorial Team
arxiv.org 4 min read

CaMeL: Defeating Prompt Injection Attacks in LLM Agents Through Capability-Based Security

CaMeL (CApabilities for MachinE Learning) is a novel defense system that protects LLM agents from prompt injection attacks by applying software security principles—including control flow integrity, capabilities, and …

Security · AI Editorial Team
download.ssrn.com 3 min read

Do Threats and Tips Actually Improve AI Performance? A Rigorous Benchmark Study

Researchers from Wharton tested whether threatening or tipping AI models improves performance on GPQA Diamond and MMLU-Pro benchmarks across five models. Results show these popular ‘folk prompting’ strategies …

AI Editorial Team
arxiv.org 0 min read

The Log is the Agent: Event-Sourced Reactive Graphs for Auditable, Forkable Agentic Systems

ActiveGraph introduces an event-sourced runtime where an append-only log serves as the primary source of truth for AI agents, with graph state as a deterministic projection of that log. This architecture enables …

AI · Architecture Editorial Team
arxiv.org 3 min read

Zep: A Temporal Knowledge Graph Architecture for Agent Memory

Zep is a memory layer service for AI agents powered by Graphiti, a temporally-aware knowledge graph engine that dynamically integrates conversational and structured business data. It outperforms MemGPT on the Deep Memory …

AI · Architecture Editorial Team
arxiv.org 4 min read

EngramaBench: A New Benchmark for Evaluating Long-Term Conversational Memory in LLM Assistants

EngramaBench is a new benchmark for evaluating long-term conversational memory across five query types including cross-space reasoning, temporal inference, and adversarial abstention. A controlled comparison of GPT-4o …

AI Editorial Team
arxiv.org 4 min read

Harnesses for Inference-Time Alignment over Execution Trajectories

This paper studies harness design for LLM agents through the lens of inference-time trajectory alignment, showing that effective harnesses must align decomposition granularity with agent capability and guidance with task …

AI · Development Editorial Team
arxiv.org 4 min read

SkillOpt: A Text-Space Optimizer for Self-Evolving Agent Skills

SkillOpt introduces a systematic text-space optimizer for agent skills that treats skill documents as trainable external state, using bounded edits, validation gates, and epoch-wise updates to achieve consistent …

AI · Development Editorial Team