🛰 AI Brief — Nov 20, 2026
How to read
prioand sources
prio Nis the radar’s practical-relevance score for this item (higher runs first; items at or below the noise threshold are filtered out as noise). Under each signal: Concepts / Entities are graph links; Source / N sources list every outbound link for that story.
🥇 GrepMem: Lossless, Index-Free Memory for Long-Horizon LLM Agents ·
prio 13For builders designing long-horizon AI agents, this research challenges complex embedding-based vector memory architectures by showing that a simple append-only text log paired with regex search can achieve higher recall accuracy with lower infrastructure overhead. It directly addresses community knowledge gaps around agent memory and context engineering by offering a practical, lossless alternative to lossy summarization. Concepts: Agent Memory Agents Context Engineering Tool Use Entities: Hugging Face CORE Connected Papers Litmaps DagsHub Qwen3.5-9B 2 sources: arxiv.org, arxiv.org
🥈 CompAct-Bench: Evaluating Working-Context Compaction in Long-Horizon Agent Tasks ·
prio 11For builders running long-horizon agents or coding assistants that compress conversation history to manage token limits, this paper highlights that context compaction frequently degrades downstream task accuracy and distorts task state even in frontier models. Understanding these failure modes helps teams design more reliable context-management and verification mechanisms instead of relying naively on LLM summarization. Concepts: Context Engineering Agents LLM Evals Entities: Claude GPT Source: arxiv.org
🥉 alphaXiv Releases OpenResearch, an Open-Source Local Harness for Research Agents ·
prio 11OpenResearch provides a local workspace and execution harness directly compatible with the community’s primary coding agents (Claude Code, Cursor, Codex, and Google Antigravity). It demonstrates a practical pattern for agent memory by persisting experiment history, logs, and artifacts in a local SQL database so agents can build on previous trials. Concepts: Agents Agent Memory Code Agents Entities: alphaXiv Google Hugging Face Modal Source: github.com
4️⃣ DriftGuard: Online Memory-Drift Detection via Causal Attribution for LLM Agents ·
prio 10For builders creating multi-step agents that rely on persistent memory and tool calling, memory poisoning or drift poses a critical reliability challenge. DriftGuard provides a concrete causal method to detect and filter harmful retrieved memories at serving time without incurring prohibitive rollout latency. Concepts: Agent Memory Agents Tool Use Source: anonymous.4open.science
5️⃣ AutoRAGBench: A Benchmark for LLM Agents Automating RAG System Development ·
prio 9AutoRAGBench evaluates whether coding agents can automate end-to-end RAG architecture design and iterative optimization, directly connecting to the community’s interest in autonomous coding agents and weak area in RAG development. The findings demonstrate that agents like GPT-6.1 Sol paired with Codex CLI can iterate on and surpass human-designed baseline RAG pipelines. Concepts: RAG Agents Code Agents LLM Evals Entities: arXiv GPT-6.1 Sol Source: arxiv.org
Knowledge Gaps
Topics the AI stream keeps raising that the knowledge base hasn’t sufficiently covered yet — candidates for what to learn next. Agent Memory · Context Engineering
🚀 Models & Releases (2)
prio 7Hugging Face Introduces Inkling 2 Mini Looped MoE Model for Local AI Concepts: Long Context Open Source LLMs Entities: Hugging Face Thinking Machines Lab Inkling 2 Mini Inkling 2 Small Source: huggingface.coprio 7Inkling 2 Mini Looped Open-Weights Transformer Released for Local Agentic Coding Concepts: Open Source LLMs Code Agents Tool Use Entities: Hacker News Stanford NVIDIA Tinker Source: thinkingmachines.ai
🧪 Research Papers (29)
prio 9The ROI of Small Examples: Optimizing In-Context Learning Under Token Budgets Concepts: Context Engineering Entities: arXiv arXivLabs Connected Papers Litmaps Source: arxiv.orgprio 9Beyond Weight Updates: Evolving Agent Harnesses for Domain Specialization Concepts: Agents Code Agents Tool Use LLM Evals Source: arxiv.orgprio 8ChronoDossier: Benchmarking Long-Form Personas as Temporal Memory Representations for Agents Concepts: Agent Memory Agents LLM Evals Entities: arXiv Hugging Face GPT-5.1 Sonnet 4.6 Source: arxiv.orgprio 8FinVeriQA: Multi-Agent Architecture for Question Answering with MCP Tool Use and Independent Evidence Verification Concepts: Agents Tool Use MCP Hybrid Search RAG Entities: SEC AMD NVIDIA Alphabet Source: arxiv.orgprio 8MemMend: A Selective Revision Framework for Memory Question Answering Concepts: Agent Memory Entities: Hugging Face Connected Papers Litmaps DagsHub Source: arxiv.orgprio 8AppSci-Bench: Evaluating How LLMs Build and Improve AI Agents Concepts: Agents LLM Evals Code Agents Entities: Anthropic OpenAI Claude Opus 5.0 openai/gpt-5.2 Source: github.comprio 8MKA Benchmark Evaluates Multimodal Knowledge Agents on End-to-End Visual Tasks Concepts: Agents LLM Evals Entities: arXiv Gemini 4 Carbon Claude Fable 5.2 Claude Opus 5.5 Source: arxiv.orgprio 8EverydayBench: Evaluating Satisfactory Completion of Live User-Facing Digital Work Concepts: LLM Evals Agents Entities: Claude Opus 5.5 Claude Fable 5.1 GPT-6 Sol GPT-5.6 Sol Source: arxiv.orgprio 7AltRA-Test: Statistical Test for Justifying LLM Judges Without Shared Annotators Concepts: LLM Evals Entities: Hugging Face Source: arxiv.orgprio 7Coverage-Guided Self-Evolution of Agent Skills with SkillPlumber Concepts: Agents Entities: arXiv arXivLabs Hugging Face GPT 5.5 Source: arxiv.orgprio 7Study Finds RL Inherits Categorical Blind Spots from LLM Judges Rather Than Specific Knowledge Concepts: LLM Evals Entities: arXiv Hugging Face DagsHub Connected Papers Source: arxiv.orgprio 7CANDID: Candidate-Driven Intervention for Disciplined Execution in Local Computer-Use Agents Concepts: Agents Source: arxiv.orgprio 7Study Finds Feedback-Free Self-Evolution Degrades Agent Skills Toward Model Priors Concepts: Agents LLM Evals Entities: GPT-5.6 Luna Source: github.comprio 7LoopRoute: Loop-Triggered Routing for System 1 Decision Agents Concepts: Agents Entities: Hugging Face Jev Source: arxiv.orgprio 7BuggyMLE: Scaling Machine Learning Engineering Difficulty with Bug Injection Concepts: LLM Evals Code Agents Entities: arXivLabs DagsHub Hugging Face Qwen3.5-9B Source: arxiv.orgprio 6TableRoute: Capability-Aware Routing for Scalable Table Question Answering Concepts: Context Engineering Entities: Hugging Face Source: arxiv.orgprio 6Rubric Semantics or Numerical Priors? Auditing Scale-Direction Bias in LLM Judges Concepts: LLM Evals Entities: arXiv Hugging Face Source: arxiv.orgprio 6The Coverage Selection Gap: Ranking Limits Document-Grounded LLM Rubric Construction Concepts: LLM Evals Entities: arXiv Hugging Face Source: arxiv.orgprio 6RoCoKV: Accelerating LLM Inference Serving via Uncertainty and Redundancy Aware KV Cache Compression Concepts: Open Source LLMs Entities: Hugging Face Connected Papers Litmaps scite Source: arxiv.orgprio 6LLMs Are Already Confidence-Aware Zero-Shot Classifiers Concepts: Open Source LLMs LLM Evals Entities: arXiv Gemma 4 31B Qwen 3.5 9B Source: arxiv.orgprio 6Scaling Study Shows Canon Layers Aid Pretraining but Reveals Benchmark Blindspots in In-Context Retrieval Concepts: LLM Evals Entities: arXiv Mamba2 GLA5 GDN2 Source: arxiv.orgprio 6TETHER: Training-Free ToC-Anchored Hierarchy Extraction for Long Documents Concepts: Chunking Entities: arXiv arXivLabs alphaXiv CatalyzeX Source: arxiv.orgprio 6Measuring Hallucination in Large Language Models for Cyber Threat Intelligence Concepts: LLM Evals Entities: arXiv Source: arxiv.orgprio 6Diagnosis Is Not a Policy: Auditing Capability-Conditioned Test-Time Inference Concepts: LLM Evals Context Engineering Entities: arXiv Source: arxiv.orgprio 6Google Research Introduces Pro² Agent with Problem State Tracking for MLE-Bench Concepts: Agents LLM Evals Entities: Google DAIR.AI Source: arxiv.orgprio 6Neurosymbolic Multimodal Framework for 30-Day Diabetic Readmission Prediction Using Hybrid RAG and Clinical Embeddings Concepts: RAG Hybrid Search Reranking Embeddings Entities: Hugging Face arXiv Bio_ClinicalBERT Source: arxiv.orgprio 6Parametric User Memory Management via LoRA Decay and Policy Control Concepts: Agent Memory RAG Entities: arXiv Hugging Face Connected Papers Litmaps Source: arxiv.orgprio 6Learning to Read Events for Exact State Tracking with Frozen Language Models Entities: Hugging Face Sonnet 5.5 GPT-6.1 Sol Source: arxiv.orgprio 6Rate-Distortion Framework for Hierarchical Summarization and Retrieval-Augmented Generation Concepts: RAG Long Context LLM Evals Source: arxiv.org
🛠 Tools & Frameworks (9)
prio 9Hugging Face Introduces Harness Adapters to Tune Models for Agent Frameworks via RL Concepts: Agents Code Agents Tool Use LLM Evals Entities: Hugging Face Qwen3.8-27B Source: huggingface.coprio 8How to Securely Connect Internal Cloud Run MCP Servers to Gemini Enterprise Concepts: MCP Agents Tool Use Entities: Google GitHub Gemini Enterprise Source: developers.googleblog.comprio 8OpenComputer: Open-Source Self-Hostable Devin Alternative Concepts: Code Agents Agents Entities: Digger Cloudflare Fly Source: github.comprio 7Agnes Harness Opens Source Dual-System Decision Framework and Self-Evolution for Agents Concepts: Agents Agent Memory MCP Entities: QbitAI Stanford University Agnes Sys1.0 Source: qbitai.comprio 7Handling User, Connector, and Agent Authentication in Gemini Enterprise with a Custom UI Concepts: Tool Use Agents Entities: Google Microsoft Okta Jira Source: developers.googleblog.comprio 7Google Introduces Semantic Governance Policies for AI Agents in Preview Concepts: Agents Tool Use Entities: Google 2 sources: [developers.googleblog.com](https://developers.googleblog.com/check-what-the community’s-ai-agents-do-before-they-do-it-with-semantic-governance-policies/), developers.googleblog.comprio 6Gemini 4 Neon Leads the Models Engineering Index Rankings Concepts: LLM Evals Entities: Google Anthropic OpenAI Meta Source: git-aarvin.github.ioprio 6Jujutsu Primer: An Introduction to the Jujutsu Version Control System Source: ethanlowman.comprio 6OpenAI Expands Business Agent Publishing in ChatGPT to Non-U.S. Ecommerce and Online Services Concepts: Agents Entities: OpenAI Source: x.com
🏢 Industry / Business (1)
prio 7Epic Games Shifts Unreal Engine Architecture to Support AI Coding Tools Concepts: Code Agents MCP Entities: Epic Games Seeker Entertainment AWS Anthropic Source: spectrum.ieee.org
💬 Opinions (4)
prio 9Why llms.txt and MCP Servers Failed to Make a Package AI-Ready Concepts: MCP Tool Use Code Agents Entities: Claude Source: backpackforlaravel.comprio 8Integrating AI Agents into Requirements Engineering Pipelines via VS Code and Skills Concepts: Agents Open Source LLMs Entities: Habr GitHub Azure OpenAI Source: habr.comprio 7Token Fraud and Free-Trial Abuse in AI Agent Applications Concepts: Agents Entities: ChatPRD Roblox Fortnite OpenRouter Source: youtube.comprio 7Developers Transitioning into Software Architects in the Era of AI Coding Concepts: Code Agents Entities: Google Hacker News Source: manidoraisamy.com
FAQ
What is in the 2026-11-20 AI brief?
The 2026-11-20 brief selected 50 signal items for AI builders and filtered 199 items as noise, using the radar’s community-relevance scoring.