Section · 42 stories
Research
Papers, benchmarks, and what we'd build differently after reading them.
Research · Jun 19 · 9 min
68 Agent Papers Map Production LLM Engineering
Research · Jun 19 · 6 min
CUA-Gym Generates 32K Verified RLVR Tuples for Computer-Use Agents
Research · Jun 15 · 7 min
AgingBench Measures How Deployed AI Agents Degrade
Research · Jun 15 · 7 min
LAP Protocol Targets Agent-to-Instrument Gap in Labs
Research · Jun 9 · 8 min
Arxiv Preprints Map Agent Evaluation Gaps in 2025
Research · Jun 9 · 5 min
GroupTravelBench Tests LLM Agents on Group Travel Planning
Research · Jun 9 · 6 min
ProCodeBench Exposes Gap Between Simulated and Real Developer Traces
Research · Jun 9 · 6 min
TFlow and LCGuard Push Agent Comms Beyond Text
Research · Jun 8 · 6 min
Three arXiv Papers Map Where LLM Agents Break Down
Research · Jun 7 · 6 min
Framework Maps Reliability Gaps in Computer-Use Agents
Research · Jun 7 · 7 min
LLM-Wiki Reframes RAG as Reasoning for Agent Pipelines
Research · Jun 7 · 6 min
New Benchmark Tracks Instrumental Convergence in LLM Agents
Research · Jun 7 · 5 min
stable-worldmodel Targets Reproducibility Gap in World Modeling
Research · Jun 6 · 6 min
RL Framework Targets Fairness Under Selective Labels
Research · Jun 5 · 5 min
SafeHarbor Cuts LLM Agent Over-Refusal With Memory Rules
Research · Jun 4 · 6 min
OTora Framework Exposes R-DoS Attacks on LLM Agents
Research · Jun 4 · 5 min
Switchcraft Routes Tool Calls to Cut Inference Cost 84%
Research · Jun 4 · 5 min
SynAE Framework Benchmarks Synthetic Agent Datasets
Research · Jun 4 · 5 min
Two arXiv Papers Map Distinct LLM Agent Memory Failures
Research · Jun 3 · 6 min
Tool-Calling Decisions Are Linearly Steerable in LLMs
Research · Jun 2 · 5 min
BeliefMem Brings Probabilistic Memory to LLM Agents
Research · Jun 1 · 5 min
BioResearcher Brings Scenario-Guided Agents to Translational Medicine
Research · May 31 · 6 min
Memory-R2 and MEMDRIFT Target Agent Memory Failures
Research · May 31 · 5 min
SIRA Collapses Multi-Round Search Into One BM25 Call
Research · May 29 · 8 min
Five arxiv Papers Advance RLVR Post-Training for LLMs
Research · May 29 · 5 min
Maestro RL Framework Routes Tasks Across Frozen Expert Models
Research · May 29 · 6 min
MOSS Lets Agents Rewrite Their Own Source Code in Production
Research · May 27 · 5 min
Curiosity Agent Pairs 3D Reconstruction With Sequence Memory
Research · May 27 · 8 min
May 2025 LLM Research Targets Reasoning, Perception, Culture Gaps
Research · May 25 · 6 min
Paper Argues Agentic AI Is Necessary Fix for OOD Limits
Research · May 24 · 5 min
Skill-R1 Trains a Skill Generator to Guide Frozen LLMs
Research · May 21 · 6 min
Agentic Success Rate Exposes Hidden LLM Payment Shortcuts
Research · May 21 · 7 min
Nautilus Compass Detects LLM Agent Persona Drift Without Model Weights
Research · May 19 · 6 min
Cross-Component Interference Undermines LLM Agent Stacking
Research · May 18 · 5 min
LATTE Framework Uses Task Graphs to Coordinate LLM Teams
Research · May 18 · 6 min
NeuroAgent Automates Multimodal Neuroimaging Pipelines
Research · May 14 · 7 min
CIVeX Verifier Targets Causal Gaps in Tool-Using Agents
Research · May 14 · 6 min
RAO Trains Agents to Delegate Sub-Tasks to Themselves
Research · May 14 · 5 min
Safactory Offers Closed-Loop Pipeline for Agent Training
Research · May 14 · 6 min
SkillOS Trains LLM Agents to Curate Reusable Skills
[research · screenshot]
Research · May 11 · 5 min
Google AI Co-Mathematician Scores 48% on FrontierMath Tier 4
[research · screenshot]
Research · May 11 · 11 min