AI Agents news and analysis on Web Pulse: 49371 curated articles tracking the latest AI Agents developments, tools, and research, updated continuously from vetted sources.
A new arXiv paper (2610.41435v1) introduces BlindToChange, a benchmark showing that AI agents score as much as 27.5 points lower when a recent real-world event affecting the answer is not explicitly named in the request …
A new arXiv paper (2610.41482v1) introduces KEEPER, a comparative analysis of four auxiliary objectives for recursive textual memory in long-horizon LLM agents, evaluated along information-content and update-sensitivity …
A new arXiv paper (2610.41040v1) introduces Capability-Grounded Agentic Planning (CGAP), a method that compiles a tool catalog deterministically into a PDDL model, uses an LLM to convert requests into goals, and solves t…
A lightweight LLM-centric conversational layer called SimXAI replaces the separate intent-recognition, dialogue-management and template-generation modules of GUS-style conversational explainable AI agents with a single t…
Researchers propose Flide, an exact inference engine for decision models such as Jev that decouples a decision request's logical tree structure from its physical execution in a single flattened token stream. Flide's Fork…
A new arXiv paper (2610.40939v1) links models' limited control over their chain-of-thought reasoning traces to an entropy collapse in the first few tokens after the opening thinking tag, and shows two inference-time inte…
A new arXiv paper (2610.40927v1) introduces Agentic World Modeling (AWM), a method that lets large language model agents run test-time scaling in shared, non-resettable environments by simulating hypothetical actions thr…
A new arXiv paper (2610.40959v1) introduces P-ContextBench, a benchmark built from 100 population-structured synthetic personas, 10,088 hidden facts, 18,261 app-trace fragments and 4,810 verified test cases to test how L…
A system called Noura, presented in arXiv paper 2610.40992v1, raised Qwen3-14B's complete-plan pass rate from 24.0% to 72.5% on 4,993 held-out participants by pairing LLM planning with executable personal specifications …
A decentralized LLM team using an auction-based consensus protocol breached a simulated perimeter 15% of the time, versus 32% for a single central commander and 29% for independent agents, according to an arXiv paper (26…
A paper submitted to arXiv on 26 October 2026 introduces Coordinator-Executor Prompt Optimization (CEPO), a framework that separates system-level coordination from agent-level execution in multi-agent systems through a s…
Semi Analysis released GDPx, a new 11-domain, 206-task benchmark for agentic work in high-revenue industries, and Anthropic's Claude Opus 5.5 topped it with a score of just 12%, ahead of Google's Gemini 4 Carbon at 10.9%…
A developer behind Friday, an open-source app that runs coding agents side by side, benchmarked handing a task from one agent to another against starting over, running 14 tasks including 6 in Express. Hand-offs used fewe…
Meituan released LongCat-2.5 on November 2, a multimodal AI model with 1.6 trillion total parameters and roughly 48 billion activated per token, built for tasks requiring sustained execution by AI agents. Meituan said it…
A developer compared Vercel's AI SDK 7.0.159 and TanStack AI 1.0.1 by installing both from npm and inspecting their export surfaces, dependency trees, and wire protocols rather than running live model calls. The comparis…
ArkenForge released arken-core, an Apache 2.0 Rust library that gives AI agents calling MCP servers and APIs a cryptographic identity via ECDSA secp256k1 signature verification and W3C did:key identifiers. The library ex…
A paper submitted to arXiv on 30 Oct 2026 reports that a search harness using only BM25/grep and MiniMax M3 answers 85% of BrowseComp-Plus and 91% on multi-hop FRAMES without compiled or vector indexes, and that a naviga…
A developer released AWKLog, a Datalog interpreter written in pure POSIX awk, built with assistance from Claude, that can be copied into a project and run at build or test time with no external dependencies. AWKLog uses …
DeepSeek's Harness team published dsh-superagent-prototype on GitHub under the MIT license on 31 October 2026, a TypeScript research prototype that turns a GitHub repository into a proactive coding agent using a router-w…
Google Threat Intelligence Group and Mandiant reported on 30 October 2026 that multiple threat actors are running open-source agentic pentesting frameworks — led by ARTEX and including PentAGI, Hexstrike and CAIRN — in r…