Agentic AI Digest: February 2026
All 28 daily agentic AI research digests from February 2026: agent planning, coordination, evaluation and deployment, newest first.
Feb 28, 2026
Bedrock Stateful Runtime reports CORPGEN concurrency gains, Phoenix Open Agent
Feb 28, 2026: Amazon Bedrock Stateful Runtime Environment adds persistent state for agent workflows; Observability for Agentic System Reliability.
Feb 27, 2026
MobilityBench, EMPO², AgentDropoutV2, Exgentic, OmniGAIA benchmarks
Feb 27, 2026: MobilityBench benchmarks LLM route agents and exposes preference-constrained routing failures. Plus 4 more items.
Feb 26, 2026
ACAR routing failures, Pancake memory indexing, AgenticTyper typing verification
Feb 26, 2026: ACAR adaptive ensembling routes on three probes but collapses on unanimous wrong answers; AgenticTyper Automates Typing for Legacy Codebases.
Feb 25, 2026
HieraMAS credit attribution, Trace-Free+ tool schemas, LongCLI-Bench 16.7%
Feb 25, 2026: HieraMAS hierarchical reward attribution improves multi-agent LLM credit assignment; Trace-Free Curriculum for Tool Interface Optimization.
Feb 24, 2026
NutriOrion, EDU-MATRIX, UMA and ACAL on agent state and coordination
Feb 24, 2026: NutriOrion hierarchical nutrition agents enforce guideline and drug–food constraints; LLM Analysts Reveal Analytic Instability.
Feb 23, 2026
MultiVer recall/FPR trade-offs and El Agente Gráfico typed execution
Feb 23, 2026: Belief Divergence-Aware Communication for 6G agents triggers messages on predicted misalignment. Plus 4 more items.
Feb 22, 2026
APPO-DTDE implicit cooperation, Agent Builder memory, EVMbench
Feb 22, 2026: Implicit Cooperation: APPO-DTDE reaches 91.7% coordination. Plus 4 more items. What each means for teams building AI agents.
Feb 21, 2026
Proxy State Evaluation, Graphon MARL, and CUWM UI World Models
Feb 21, 2026: Proxy state LLM agent evaluation benchmarks multi-turn tool use without deterministic backends. Plus 4 more items.
Feb 20, 2026
GUI-Owl-1.5 crop refinement, AlphaEvolve MARL variants, Calibrate-Then-Act priors
Feb 20, 2026: GUI-Owl-1.5 crop refinement lifts ScreenSpot-Pro accuracy to 80.3; AlphaEvolve evolves CFR and PSRO variants with LLM-guided search.
Feb 19, 2026
Pythia prompt optimization failures, LLM social dilemmas, PAHF memory
Feb 19, 2026: Pythia iterative prompt optimization undermines rare symptom detection. Pythia exhibited prevalence-dependent optimization instability.
Feb 18, 2026
ACP latency results, ResearchGym outcomes, and Colosseum collusion audits
Feb 18, 2026: Agent Communication Protocol (ACP) standardizes decentralized agent discovery and federated orchestration.
Feb 17, 2026
Adversarial Self-Critique and Agent Mars coordination benchmarks
Feb 17, 2026: Adversarial Self-Critique in Regulated Underwriting Agents reduces hallucinations from 11.3% to 3.8%. Plus 4 more items.
Feb 16, 2026
Skill Integration Security, GT-HarmBench, and Theory-of-Mind Coordination
Feb 16, 2026: Security risks in agent skill integration expand the attack surface of modular skill systems. Plus 4 more items.
Feb 15, 2026
Structural Context Model, BG-MCTS, and MILE-RefHumEval updates
Feb 15, 2026: SynthAgent simulates virtual obesity patients with mental health comorbidities. Plus 4 more items. Every item links its primary source.
Feb 14, 2026
SHARP credit assignment and VERIFY-RL verifiable decomposition gains
Feb 14, 2026: SHARP enables precise credit assignment in multi-agent RL; AnalyticsGPT implements an LLM workflow for scientometric question answering.
Feb 13, 2026
VIRF Logic Tutor, MemAdapter, SAGE-RL, Gemini synthesis eval
Feb 13, 2026: VIRF uses a Logic Tutor to causally diagnose and repair embodied-AI plans. Plus 3 more items. What each means for teams building AI agents.
Feb 12, 2026
PASB, AGENTWM, SCOUT-RAG, and regime leakage in evaluation
Feb 12, 2026: PASB security evaluation finds OpenClaw vulnerabilities across prompts, tools, and memory. Plus 3 more items.
Feb 11, 2026
PRISM, AT-GRPO, and Reinforcement Inference thresholds in agents
Feb 11, 2026: PRISM decomposes multi-agent reasoning gains into exploration, information fidelity, and aggregation. Plus 4 more items.
Feb 10, 2026
RuleSmith, HMAGAT, and agent sandbox connection patterns
Feb 10, 2026: RuleSmith framework automates multi-agent game rule balancing with Bayesian optimization. Plus 4 more items.
Feb 09, 2026
GNN epistemic planning, Constella tool study, and long-horizon stability limits
Feb 9, 2026: GNN Heuristics for Multi-Agent Epistemic Planning reduces node expansions but not runtime. Plus 3 more items.
Feb 08, 2026
DeepRead structure-aware retrieval, reducible uncertainty, and MINT knowledge-gap queries
Feb 8, 2026: DeepRead structure-aware multi-turn search uses document hierarchy to localize and read evidence; Reducible Uncertainty for LLM Agents.
Feb 07, 2026
Structured multi-evaluator bias, SocialVeil barriers, and GAMMS simulation
Feb 7, 2026: Structured Multi-Evaluator Framework measures -0.33 to +0.77 judge bias in MCC risk rationales. Plus 4 more items.
Feb 06, 2026
AgentArk, Execution-Driven Iterative Reasoning and PlanBench
Feb 6, 2026: AgentArk distillation compresses multi-agent reasoning into a single LLM; Execution-Driven Iterative Reasoning in LLMs.
Feb 05, 2026
LangChain multi-agent patterns, OpenAI agent link safety, CrewAI Flows
Feb 5, 2026: Multi-Agent Architectures for Complex Tasks; OpenAI agent link safety mitigates URL exfiltration and prompt injection.
Feb 04, 2026
Deep Agents context isolation for multi-agent delegation
Feb 4, 2026. Deep Agents structures multi-agent applications by assigning specialized subagents to discrete subtasks and coordinating outputs.
Feb 03, 2026
CrewAI large-scale workflows and Arize observability-driven sandboxing
Feb 3, 2026: Arize observability-driven sandboxing gates agent side effects without changing planning. Plus 1 more item.
Feb 02, 2026
Deep Agents and Alyx hierarchical memory for agent context scaling
Feb 2, 2026: Deep Agents SDK filesystem offloading and summarization manage multi-turn context. Plus 3 more items. Every item links its primary source.
Feb 01, 2026
Arize agent failure tracing, OWASP ASI observability, PrivacyChecker CI-RL
Feb 1, 2026: Operational Failure Patterns in AI Agents; OWASP Agentic Application Risks maps ASI risks to Arize AX observability controls.