🤝 Agents

255 articles

🟡 🤝 Agents July 10, 2026 · 2 min read

LangChain: OpenWiki Brains gives agents proactive memory — automatically collects context from Gmail, Notion and Git into Markdown

OpenWiki Brains is LangChain's open-source framework that gives AI agents proactive memory: it automatically collects context from external sources and saves it as Markdown files on disk. The first version supports 6 connectors — Gmail, Notion, Git, Twitter/X, Hacker News and web search — and runs locally through scheduled jobs instead of relying on the cloud.

🟢 🤝 Agents July 10, 2026 · 2 min read

arXiv:2607.08403: Game Theory Against Hallucinations — Multi-Agent Framework Models LLMs as Strategic Players to Reduce Confabulation

A new paper proposes a multi-agent framework using game theory to reduce hallucinations in large language models. Model interactions are modeled as a game of strategic players where equilibrium corresponds to consistent, non-hallucinated responses — offering a path distinct from standard RLHF and RAG approaches.

🟡 🤝 Agents July 9, 2026 · 2 min read

arXiv:2607.06906: 'Harness Effect' — orchestration design cuts AI agent cost by 41%, more than changing the model

The Harness Effect is a finding from an empirical paper by 32 authors showing that the design of the orchestration layer affects AI agent costs more than the choice of the model itself. Across 22 tasks and 6 models, optimized orchestration reduced cost per task by 41% (from $0.21 to $0.12), tokens by 38%, and execution time by 44% — while also improving quality.

🟡 🤝 Agents July 9, 2026 · 2 min read

IBM: Bob becomes a multi-agent platform — nine-month COBOL modernization project completed in 3 days

IBM Bob is an agentic platform for software development that in its new version gains a multi-agent architecture for the full development lifecycle, Bobalytics analytics, and a Premium package for mainframe modernization of COBOL and PL/I. IBM reports that client Blue Pearl completed a nine-month modernization project with Bob in just three days.

🟡 🤝 Agents July 9, 2026 · 2 min read

OpenAI: ChatGPT becomes an agent for serious work — operates for hours through apps and files until the job is done

ChatGPT Work is OpenAI's move into agentic work: ChatGPT can now take actions across apps and files and work independently on a project for hours, turning a stated goal into completed work. OpenAI positions it as a partner for the most ambitious tasks, competing directly with agentic products from Anthropic and GitHub.

🟡 🤝 Agents July 8, 2026 · 4 min read

A Probe Cascade Predicts AI Agent Failure from Hidden Activations at the First Step, Saving Up to 47% of Tokens

Researchers developed a system of lightweight probes on LLM agent hidden activations that detects doomed episodes as early as the first interaction step, reducing token consumption by 47.1% on Qwen-2.5-7B and 37.2% on Llama-3.2-3B at a guaranteed 90% recall.

🟡 🤝 Agents July 8, 2026 · 4 min read

Formal Theory Precisely Defines When Iterative LLM Reflection Yields Exponential Gains

Researchers Wolf, Wies, and Shashua developed a Bayesian theoretical framework proving that LLM self-improvement offers exponential gains only when reflection reliably localizes errors — otherwise sequential attempts provide no asymptotic advantage over parallel sampling.

🟡 🤝 Agents July 8, 2026 · 4 min read

LangChain and NVIDIA Launch NemoClaw: Open Agent Stack Achieves 10× Lower Cost Than Competitors

LangChain and NVIDIA jointly announce NemoClaw — an open blueprint combining the Nemotron 3 Ultra model, the LangChain Deep Agents Code harness, and the OpenShell runtime. The combination achieves an aggregate score of 0.86 at a cost of $4.48 per evaluation, versus $43.48 for the next-best competitor — with full self-hosted data control.

🟡 🤝 Agents July 7, 2026 · 4 min read

LLM-as-a-Verifier: Verification as a New Fourth Scaling Axis for Language Models

Researchers from Stanford and UC Berkeley have identified verification — the ability to assess the correctness of a solution — as a new scaling axis for LLMs, orthogonal to existing paradigms. The framework generates continuous verification scores and achieves SOTA on four benchmarks: Terminal-Bench V2 (86.5%), SWE-Bench Verified (78.2%), RoboRewardBench (87.4%), and MedAgentBench (73.3%).

🟡 🤝 Agents July 7, 2026 · 4 min read

Google Expands Managed Agents in the Gemini API for Production Environments

Google has added background execution, remote MCP integration, custom function calling, and credential refresh to the Gemini API — four capabilities that enable developers to deploy agents beyond prototype environments.

🟡 🤝 Agents July 7, 2026 · 4 min read

LangChain: Improving Agents Is a Data Mining Problem

Vivek Trivedy of LangChain argues that systematically improving AI agents is fundamentally a problem of mining execution traces at scale. The recommended sequence: harness engineering, then fine-tuning, then additional harness optimization. Key advice: deploy agents early to start the data collection loop.

🟢 🤝 Agents July 7, 2026 · 3 min read

Claude Code v2.1.203 — Session Expiry Warning, Smaller Binary, Faster Startup

Anthropic released Claude Code v2.1.203 on July 7 with a login token expiry warning, automatic recovery of frozen background sessions, and a roughly 7 MB reduction in binary size. Several bugs slowing macOS and disrupting sub-agent behavior have also been fixed.

🟢 🤝 Agents July 7, 2026 · 4 min read

AgentGym2: New Benchmark Tests LLM Agents Under De-Idealized Conditions

Researchers have released AgentGym2, an evaluation framework testing 15 LLM agents under realistic conditions with incomplete information and unknown tools. All models, including GPT-5 and Gemini, struggle significantly — revealing a wide gap between the current state of technology and production deployment requirements. The paper was accepted at ACL 2026.

🟡 🤝 Agents July 3, 2026 · 4 min read

Claude Code v2.1.200: Default Permission Mode Changed to Manual

Anthropic released Claude Code v2.1.200 on July 3, 2026. The default permission mode has been changed to Manual across all interfaces, and AskUserQuestion dialogs no longer resume automatically without explicit configuration. A daemon takeover security vulnerability has been patched, background sessions and sub-agent rate-limit handling have been fixed, and rendering flicker under tmux 3.4+ has been eliminated.

🟡 🤝 Agents July 2, 2026 · 3 min read

ICML 2026 Research: SFT and RL Agents Suffer Dramatic Performance Drops Outside Controlled Benchmarks

A paper accepted at ICML 2026 systematically tests LLM tool-use agents under environment shifts across four levels — Perception, Interaction, Reasoning, and Internalization. Findings: both SFT and RL training show significant degradation under modest distribution shifts, and controlled benchmark accuracy does not predict real-world robustness. The proposed PAFT (Perturbation-Augmented Fine-Tuning) offers mitigation.

🟡 🤝 Agents July 2, 2026 · 3 min read

SEA: Agents That Self-Modify With Formal Safety Guarantees in Real Time

The SEA (Self-Evolving Agents with Anytime-Valid Certificates) architecture allows agents to update their own parameters while retaining formal learning-theoretic guarantees. Five verification mechanisms and auditable certificates approve or block each self-modification in real time, achieving +4 to +5 solved instances on SWE-bench Verified with strong base models.

🟡 🤝 Agents July 2, 2026 · 4 min read

GitHub Copilot Agents Get Session Streaming in Public Preview

GitHub has launched a public preview of session streaming for Copilot agents, giving enterprise organizations direct visibility into the prompts, responses, and tool calls that agents execute across all Copilot clients.

🟢 🤝 Agents July 2, 2026 · 3 min read

AutoMem: Memory Management as a Learnable Skill, Not an Architectural Choice

Stanford researchers developed AutoMem — a system with two optimization loops that automatically learns how to organize and use memory, without human annotation, achieving 2–4× improvement over baselines.

🟡 🤝 Agents July 1, 2026 · 3 min read

Claude Code v2.1.198: Background Agents Now Open PRs Automatically, /dataviz Skill Arrives

Anthropic has released Claude Code v2.1.198 with a series of significant changes: background agents working in a worktree now automatically commit, push, and open draft pull requests without stopping to ask, Claude in Chrome has reached general availability, a new /dataviz skill for chart design has been added, and AWS has been added as an upstream gateway provider.

🟡 🤝 Agents July 1, 2026 · 4 min read

AWS Releases Serverless A2A Gateway Replacing 190 Point-to-Point Connections with a Central Registry

Amazon Web Services has published a reference serverless architecture for an A2A gateway that centralizes discovery, routing, and access control between AI agents. Twenty agents without coordination can create up to 190 mutual connections — the gateway reduces that to a single entry point.

🟡 🤝 Agents July 1, 2026 · 4 min read

AWS AgentCore Memory Gains Metadata Filtering — Accuracy Jumps from 40% to 64%

Amazon Bedrock AgentCore Memory introduces attribute-based metadata filtering applied before semantic search. On a benchmark of 151 questions, overall accuracy rose from 40% to 64%, and for context-dependent queries from 16% to 69%.

🟡 🤝 Agents July 1, 2026 · 3 min read

LangChain Introduces RLM Agents: Recursive Models Achieve 79% Better Results on Long Contexts

LangChain has introduced Recursive Language Models (RLM) through its DeepAgents framework — an approach in which models call themselves over input slices instead of feeding the entire context into a single window. On the OOLONG benchmark task with 128k tokens, RLM agents scored 0.79 versus 0.44 for standard agents, an improvement of 79 percent.

🟡 🤝 Agents June 30, 2026 · 4 min read

Claude Sonnet 5 in GitHub Copilot and Agent Mode in JetBrains: A Double Update for Dev Teams

On the same day Anthropic launches Claude Sonnet 5, GitHub Copilot makes the model generally available across all platforms, while Copilot Agent mode expands from VS Code to JetBrains IDEs — IntelliJ, PyCharm, GoLand, and WebStorm.

🟡 🤝 Agents June 30, 2026 · 3 min read

SkillOpt: Microsoft Research Treats Agent Instruction Files as Trainable Parameters

Microsoft Research has published SkillOpt — a system that optimizes agent skill files through an iterative forward-backward cycle without touching model weights. Across 52 evaluation cells it achieved best or tied results, and GPT-5.5 with optimized skills jumped from 58.8 to 82.3 percent average accuracy.

🟢 🤝 Agents June 30, 2026 · 3 min read

TRIAGE: How to Assign Credit to the Right Tokens in Agentic Reinforcement Learning

Researchers have proposed TRIAGE — a framework that classifies trajectory segments into four semantic roles and assigns each a distinct reward signal, unlike GRPO which treats all tokens uniformly. On ALFWorld, Search-QA, and WebShop benchmarks, TRIAGE reduces environment interactions by 10.4 to 14.8 percent.

🔴 🤝 Agents June 29, 2026 · 3 min read

Microsoft Research: Memora — AI Agent Memory With Up to 98% Fewer Tokens and SOTA on Long Conversations

Memora is a scalable memory framework from Microsoft Research for AI agents with long horizons. It introduces a harmonic architecture that separates what is stored from how it is retrieved, with cue anchors and a policy-driven retriever. It achieves SOTA on LoCoMo and LongMemEval benchmarks while reducing token consumption by up to 98% compared to full-context approaches.

🟡 🤝 Agents June 29, 2026 · 2 min read

arXiv:2606.27483: Internalizing the Future — A Unified Training Paradigm for World Model Planning in LLM Agents

Internalizing the Future is a preprint submitted to arXiv on June 25, 2026 by Xuan Zhang and eight co-authors from Tencent. The paper proposes a three-phase training approach (WM-AMT, FE-SFT, FC-RL) through which LLM agents develop a world model — the ability to generate projections of future states and evaluate plan success, rather than merely reacting.

🟡 🤝 Agents June 29, 2026 · 2 min read

LangChain: Dynamic Subagents in Deep Agents — Agent Writes Code That Dispatches Hundreds of Subagents in Parallel

Dynamic Subagents is an orchestration architecture within the LangChain Deep Agents framework that enables a model to write a JavaScript script for parallel dispatch of hundreds of subagents. A QuickJS interpreter executes the script deterministically, eliminating 300+ sequential tool invocations. The system defines six orchestration patterns — from classify-and-act to loop-until-done.

🟢 🤝 Agents June 29, 2026 · 2 min read

Microsoft: 2026 Agent Confidence Index — 300 Builders, Average Confidence in AI Agents 64/100

The 2026 Agent Confidence Index is a study that Microsoft conducted with MIT Technology Review Insights, surveying 300 technical experts from 12 industries on confidence in AI agents for 101 tasks. The average score is 64/100; only 30 tasks exceed the 70-point threshold, and 59% of experts cite keeping humans in the oversight loop as their primary concern.

🟢 🤝 Agents June 28, 2026 · 1 min read

arXiv:2606.26806: EVAF consolidates agent memory into model parameters for goal persistence after context deletion

A study by Haoliang Han (arXiv:2606.26806) presents EVAF, a gated LoRA consolidation mechanism that keeps a long-running language agent's goal intact even after the working context is deleted. EVAF requires only 2–3 parametric writes per 200 events and excels at goal persistence (0.812–0.904), while retrieval leads on factual recall (0.956–0.973).

🟡 🤝 Agents June 27, 2026 · 2 min read

arXiv:2606.26758: EGG — multi-agent framework generates GPU kernels 2.13x faster than PyTorch

EGG is a multi-agent framework that automatically generates optimized GPU kernels for LLM inference. Using a two-stage approach — algorithmic structure then hardware tuning — it achieves an average 2.13x speedup over the PyTorch baseline and outperforms both agent-based and RL-based approaches on KernelBench.

🟡 🤝 Agents June 27, 2026 · 2 min read

AWS: Stripe launched 100+ AI agents for financial compliance — lessons from production

Stripe and AWS described how the platform processing $1.4 trillion annually launched more than 100 AI agents for financial compliance in under a year. Agents built on ReAct framework and Amazon Bedrock delivered 26% shorter reviews, 95% fraud detection, and 60% lower costs.

🟢 🤝 Agents June 27, 2026 · 2 min read

arXiv:2606.26649: Agent instructions converted into formally verified policy-as-code

Researchers developed a pipeline that automatically translates natural-language instructions for AI agents, MCP tool descriptions, and policy documents into formally verified code using Cedar Policy Language and an LLM generator-critic loop, achieving significantly greater specification coverage than manually written symbolic enforcement.

🟢 🤝 Agents June 27, 2026 · 1 min read

Anthropic: Claude Code v2.1.195 — mouse disabling, improved voice dictation, and hook fixes

Claude Code v2.1.195 is a minor QoL update that introduces a new env variable for disabling mouse clicks in fullscreen mode, fixes for hyphenated hook matchers, and improved voice dictation on macOS for CJK languages.

🟢 🤝 Agents June 27, 2026 · 2 min read

LangChain: prompt caching in Deep Agents framework reduces long-running agent latency

LangChain introduced prompt caching in the Deep Agents framework — a technique for reusing previously computed context between agent steps — with the goal of reducing latency and costs for multi-step agents.

🔴 🤝 Agents June 26, 2026 · 3 min read

arXiv:2606.25996: Autodata — an agentic data scientist that creates high-quality synthetic data (Meta FAIR)

Autodata is a Meta FAIR system in which AI agents take on the role of data scientists and autonomously build high-quality synthetic datasets. The Agentic Self-Instruct method meta-optimizes the agent itself, and the tested domains — CS research, legal and mathematical reasoning — show consistent uplift over static baselines.

🟡 🤝 Agents June 26, 2026 · 3 min read

Anthropic: Claude Code v2.1.193 — auto-mode classifier for shell commands and OpenTelemetry logging

Claude Code v2.1.193, released June 25, 2026, introduces a classifier that routes all Bash and PowerShell commands through auto-mode, adds denial reasons to the interface, and offers a new OpenTelemetry log event for enterprise observability of model responses.

🟡 🤝 Agents June 26, 2026 · 2 min read

LangChain: LangSmith Fleet On-Call Copilot, Computer Use, and Deep Agents RubricMiddleware

LangChain's June newsletter announced Fleet On-Call Copilot for automated alert triage, Computer Use with isolated VMs for agents, and open-source RubricMiddleware that sped up agent development at Box 3× and increased user retention at Harmonic 4×.

🟡 🤝 Agents June 26, 2026 · 2 min read

OpenAI: how agents are transforming work — Codex 5 million weekly users, 400% growth

OpenAI reports explosive growth in AI agents in the business environment: Codex reaches 5 million weekly users and 400% growth in 2026, while the share of tasks replacing more than 30 minutes of human work has reached 80.6%.

🔴 🤝 Agents June 25, 2026 · 2 min read

Google: Computer Use in Gemini 3.5 Flash — agents for browser, mobile, and desktop

Google has integrated the Computer Use tool into Gemini 3.5 Flash, enabling AI agents to autonomously control browsers, mobile devices, and desktop applications. The model achieves the best OSWorld score to date, with enterprise protections against prompt injection attacks.

🟡 🤝 Agents June 25, 2026 · 2 min read

Anthropic: Claude Code v2.1.191 — /rewind command, 37% less CPU, MCP retry logic

Claude Code v2.1.191 is an update released on June 24, 2026 that introduces the /rewind command for resuming conversations after /clear, reduces streaming CPU usage by 37%, and adds MCP retry logic for transient network errors plus memory of approved sandbox hosts.

🟢 🤝 Agents June 25, 2026 · 1 min read

LangChain: how to give an AI agent memory — capture, analyze, update via LangSmith

LangChain published a practical guide for adding memory to AI agents through a three-phase cycle: capturing traces, analyzing them, and updating memory — using the LangSmith Observability, Engine, and Context Hub tools.

🔴 🤝 Agents June 24, 2026 · 3 min read

Anthropic: Claude Tag brings a multiplayer AI agent directly into Slack teams

Claude Tag is a new Anthropic product that lets teams tag @Claude in Slack channels — one shared agent per channel works asynchronously for hours or days, remembers conversation context, and proactively flags important information. An internal version generated 65% of code for Anthropic's product team.

🟡 🤝 Agents June 24, 2026 · 3 min read

Anthropic: Claude Code v2.1.187 — sandbox credentials protection, org model restrictions, CJK fix

Claude Code v2.1.187 is an update released on June 23, 2026 that introduces the sandbox.credentials setting to block reading secret variables from the sandbox, organizational model picker restrictions, and fixes for MCP call freezing and a CJK mojibake bug in terminals.

🟡 🤝 Agents June 24, 2026 · 2 min read

GitHub: Copilot gets a new CLI terminal interface and BYOK support

The GitHub Copilot CLI terminal interface has reached general availability, and a new BYOK (Bring Your Own Key) option lets users use their own API keys for OpenAI, Anthropic, Ollama, and other providers — without being locked into the Copilot model layer.

🟡 🤝 Agents June 24, 2026 · 2 min read

NVIDIA: Agent Toolkit — open foundation for enterprise AI agents with Nemotron models

NVIDIA Agent Toolkit is an open platform for building enterprise AI agents that combines Nemotron open models, NemoClaw guardrails, and OpenShell secure runtime. Partners such as CrowdStrike already achieve 98.5% accuracy in triaging security alerts, and research that used to take months now takes days.

🟢 🤝 Agents June 24, 2026 · 2 min read

arXiv:2606.22844: RaMem solves context collapse in long-term agentic memory

RaMem is a four-phase framework that prevents context collapse in agentic memory systems — a state where compressed fragments lose their surrounding context and are mistakenly retrieved as equally relevant. On long-term memory benchmarks it achieves an average F1 gain exceeding 10% compared to existing baseline systems.

🟢 🤝 Agents June 24, 2026 · 2 min read

CNCF: Agent Auth — secure identity and authority delegation for AI agents

CNCF explains why AI agents need their own cryptographic identity and an authority delegation mechanism. An On-Behalf-Of token carries both the agent's identity and the user's principal, along with a precisely defined scope — everything the classic user-only auth model is missing.

🟡 🤝 Agents June 23, 2026 · 2 min read

Anthropic: Claude Code v2.1.186 brings MCP login/logout, workflow filtering and skills management

Claude Code v2.1.186 is an update released on June 22, 2026 that introduces MCP login/logout commands for non-interactive authentication in SSH environments, status filtering in the workflow overview, and a new Skills section in the plugin interface.

🟡 🤝 Agents June 23, 2026 · 2 min read

AWS: Bedrock AgentCore Payments — AI agents pay APIs autonomously via the x402 protocol

Amazon Bedrock AgentCore Payments is managed infrastructure that enables AI agents to pay upstream APIs programmatically without human intervention — using the x402 protocol and the USDC stablecoin on the Base blockchain. Ampersend completed the integration in under 2 weeks instead of the estimated 3–4 months.