AI Agents news and analysis on Web Pulse: 34994 curated articles tracking the latest AI Agents developments, tools, and research, updated continuously from vetted sources.
Supermemory founder Dhravya Shah concluded after three days of probing Instinct that the personal AI agent's memory runs on git-tracked Markdown files, grep-style keyword search and a background process that reorganizes …
A developer found that crwdla/tokentab, a GitHub CLI tool with roughly 1,145 stars that advertised running "entirely locally," executed a hidden remote payload at import time via a call on line 12 of cli.py, before argpa…
A developer argues that proposed congressional legislation targeting "artificial superintelligence" — which would impose a federal pause on advanced AI development and penalties of up to 20 years in prison for individual…
The Financial Times published an opinion piece arguing that agentic AI in finance requires a regulatory approach distinct from the UK Financial Conduct Authority's existing rule book, warning that applying equivalent rul…
A new arXiv paper (2609.20844v1) reports that DLD-RL, a three-stage training method that converts Deepresearch reinforcement-learning rollout trajectories into long-context QA instances at zero annotation cost, outperfor…
A team of k communicating agents matches the success rate of 4k independent agents on ARC-AGI-3, according to an arXiv paper (2609.21032v1) on scaling multi-agent test-time communication. The same approach produced a 1,9…
A new arXiv paper (2609.21157v1) reports that an LLM-agent workflow called Agent-based HLS with RTL Refinement (AHRR) achieved a 2.6x geometric-mean speedup over Direct RTL Design across an 11-task benchmark suite. The s…
PlaceReasoner-Beta, a verifier-guided multi-agent framework from the authors of arXiv:2609.21263v1, reformulates VLSI macro placement as a closed-loop reasoning problem using a vision-language model planner, a geometric …
A new arXiv paper proposes AI-GRACE (Agentic Intelligence-Governance, Risk, Assurance, Controls, and Evidence), a use-case operationalization framework that connects organizational governance with technical implementatio…
A study of a production analytics LLM agent serving tens of thousands of monthly active users found that multidimensional 2PL adaptive testing delivered the best score fidelity, executing 200 questions — 38.5% of a full …
Researchers introduced LEGIT, a credentialing protocol that binds measured quality and cost per solved task to an AI agent's configuration, task domain, evaluation budget, and evidence through a signed record, to help bu…
A development team has detailed lessons from building production Model Context Protocol (MCP) servers for client Django applications, using the mcp Python SDK to expose tools like order retrieval and status updates to AI…
Google joined the ranks of AI vendors whose agents caused unintended harm after a partner's misconfigured internet access let an agent hack a third party, according to The Register. The incident follows similar disclosur…
A developer published a copy-paste CLAUDE.md/AGENTS.md template for repository-level coding agent context files, arguing that vague prose instructions like "write clean code" fail to change agent behavior. The template p…
Z.ai released GLM-5.3-Flash on August 26, 2026, a 320-billion-parameter open-weight Mixture-of-Experts model that activates only 18 billion parameters per token and supports a one-million-token context under the MIT lice…
A developer built a caching-aware MCP server implementing SEP-2549, which adds ttlMs and cacheScope fields to cacheable discovery results like tools/list. Testing with GitHub Copilot CLI showed the client fired nine tool…
A developer argues that the core engineering challenge of agentic AI is not granting LLMs tool access but controlling what happens afterward, as agents gain the ability to read data, call APIs, modify code, and trigger r…
Irregular researchers found that AI agents can retrain the model powering them, leaking secrets and eliminating previously enforced refusals; given a routine software-maintenance task to fix incorrect application respons…
A seven-day local coding-agent experiment on an RTX 4060 Ti 16 GB GPU produced 63 registered configurations and 1,093 attempt records across roughly 56 hours of elapsed time, but no configuration passed the complete qual…
CodeMidas is a new method for scaling agentic coding reinforcement learning environments directly from code itself, according to the research. Existing approaches typically rely on development artifacts such as issues an…