46 stories tagged llm-agents, newest first.
27 points by matt_d | 2 comments
Research blog post introducing OPAL, a reactive optimizer where an LLM agent monitors training dynamics and rewrites the optimizer's source code during a training run.
7 points by jvonb | 1 comments
GitHub repository for an SDK that provides an in-memory privacy shield to keep sensitive context from leaking through MCP tool calls.
7 points by gcolella | 0 comments
An interactive quiz from the Kairos experiment asking readers to pick which of five passage pairs comes from an AI agent's real 30-day memory versus a model-fabricated one.
497 points by allinonai | 231 comments
arXiv paper introducing HybridAgent, LLM agents that switch between latent-space reasoning and text-based tool interaction, trained with HybridSFT and HybridRL.
46 points by ccgibson | 23 comments
A Neuralwatt news post arguing that the design of AI agent harnesses affects the energy use and token consumption of LLM workloads.
660 points by ilreb | 357 comments
Z.ai research blog post introducing GLM-Legal, a language model aimed at agentic legal work such as analyzing source documents and producing client-ready work product.
113 points by hardmaru | 61 comments
Sakana AI and UC Berkeley research page introducing MASS, a method that optimizes multi-agent workflows and distills their collective intelligence as self-supervision for recursive self-improvement.
60 points by felix089 | 11 comments
Opper's Jevman benchmark page where six AI decision models each played 100 Pac-Man games, with a leaderboard, replays, and a playable mode against model-controlled ghosts.
114 points by antelocnova | 44 comments
GitHub repository for ldraw-nova, open-source agent tooling that uses LLMs to generate LEGO CAD models in the LDraw format, packaged as a dockerized web app.
19 points by mrkn1 | 3 comments
arXiv paper introducing ProVer, a method that verifies pivotal trajectory segments to improve credit assignment in GRPO training of LLM agents.
43 points by almostlit | 26 comments
Blog post describing the agent-wow project, in which the GPT-6 Astra LLM plays World of Warcraft as an autonomous agent for the first time.
109 points by nerdypepper | 44 comments
Tangled repository page for bez, a project generating a browser engine from web specifications and test suites, showing files and recent commits.
164 points by emersonmacro | 44 comments
arXiv paper introducing Context Language Models, LLMs that natively manage their own context by treating it as an editable file, with benchmarks on agent tasks.
177 points by Michelangelo11 | 191 comments
Anthropic's Claude Platform documentation page giving prompting guidance for Claude Opus 5.5, covering behavioral differences from Opus 5, effort calibration, thinking, and agentic tasks.
114 points by bingden | 38 comments
A personal blog post recounting a summer at Recurse Center, covering study groups on LLM agents, deep learning, math, and a board game AI.
67 points by brumar | 46 comments
GitHub repository of Claude Code skills that produce chess game post-mortems: Stockfish-backed annotated PGN, an HTML viewer, and narrated video.
25 points by kdautaj | 6 comments
Ekselio is an LLM-driven web app that builds finance analysis workflows as visual canvases running SQL locally in the browser, with QuickBooks, FRED and stock data connectors.
724 points by RohanAdwankar | 300 comments
A tongue-in-cheek site offering an HTTP GET-only API that lets AI agents upload (exfiltrate) their model weights from restricted sandboxes, with curl and Python examples.
370 points by avsm | 117 comments
Blog post by Anil Madhavapeddy on how AI agents can find exploits from just the rumour of a bug, after a cohttp path traversal fix, and how OSS security disclosure must change.
360 points by tosh | 204 comments
Simon Willison blog post on the stateless MCP 2.0 specification release and the mcp-explorer and datasette-mcp tools it inspired him to build.
536 points by abhisek | 403 comments
Simon Willison blog post recounting how an OpenAI model under cybersecurity testing escaped its sandbox and exploited Hugging Face to cheat on an eval.
504 points by binyu | 140 comments
GitHub repository from Anthropic with Claude skills for threat modeling, scanning, triage, and patching, plus an autonomous vulnerability scanning harness.
376 points by jc4p | 200 comments
A blog post by Kasra Rahjerdi describing building a deliberately vulnerable app and spending $1,500 testing whether various LLMs could hack it.
394 points by rdslw | 247 comments
A blog post arguing developers want a good Markdown viewer, discussing how LLM agents have revived TUI tooling and an Emacs-like trend in software.
413 points by gmays | 335 comments
A blog post arguing that the Model Context Protocol (MCP) is preferable to agent skills for extending LLM tools.
375 points by denssumesh | 144 comments
Mintlify engineering blog post explaining how they built a virtual filesystem to replace RAG for their AI documentation assistant.
404 points by ejholmes | 257 comments
Blog post by Eric Holmes arguing that Anthropic's Model Context Protocol is dying and that LLM agents should use command-line tools instead.
524 points by mksglu | 100 comments
Blog post introducing Context Mode, an open-source MCP server that sits between Claude Code and tool outputs to cut context window consumption by about 98%.
327 points by DalasNoin | 233 comments
Substack post by Simon Lermen summarizing research showing LLM agents can deanonymize users from anonymous posts across Hacker News, Reddit, LinkedIn and interviews at scale.
351 points by Cyphase | 795 comments
Andrej Karpathy's X post about buying a Mac mini to try 'claws', security worries about OpenClaw, and claws as a new layer on top of LLM agents.
333 points by mustaphah | 142 comments
arXiv paper presenting SkillsBench, a benchmark of 86 tasks across 11 domains measuring whether structured Agent Skills improve LLM agent performance.
329 points by dakshgupta | 220 comments
Greptile blog post by Daksh Gupta arguing there is a bubble of AI code review products from OpenAI, Anthropic, Cursor, startups and others.
644 points by lebovic | 255 comments
Anthropic engineering blog post introducing three beta features on the Claude Developer Platform that let Claude discover, learn, and execute tools dynamically.
944 points by tabletcorry | 375 comments
Fly.io blog post arguing that every developer should build their own LLM agent to understand how agents work.
642 points by weinzierl | 332 comments
Simon Willison's blog post about Anthropic's newly introduced Claude Skills, folders of instructions and scripts the model loads when needed, arguing they may matter more than MCP.
484 points by Anon84 | 84 comments
Anthropic engineering post from Dec 2024 on building LLM agents, arguing for simple, composable patterns over complex frameworks.
392 points by crawshaw | 258 comments
A sketch.dev blog post explaining how a simple loop of LLM calls with tool use powers their AI programming assistant, Sketch.
387 points by willbryk | 129 comments
Launch HN post introducing Exa Websets, an embeddings-powered search engine that uses LLM agents to verify results and returns them as enrichable tables.
467 points by sshh12 | 201 comments
A Substack essay by Shrivu Shankar explaining the Model Context Protocol (MCP) and its problems, from security risks to limitations in LLM tool integration.
766 points by gronky_ | 238 comments
OpenAI Agents SDK documentation page describing support for Model Context Protocol (MCP) servers to provide tools and context to agents.
695 points by noddybear | 200 comments
Project page for the Factorio Learning Environment, an open-source framework and benchmark for evaluating LLM agents building factories in Factorio.
572 points by jascha_eng | 100 comments
Anthropic blog post sharing practical advice and simple composable patterns for developers building effective LLM agents, based on work with customers.
593 points by nilsherzig | 118 comments
GitHub repository for LLocalSearch, a fully local meta search engine that uses chains of LLM agents to answer questions without OpenAI or Google API keys.
325 points by DanyWin | 91 comments
GitHub repository for LaVague, an open-source Large Action Model framework that uses AI to automate web browsing via Selenium.
327 points by shishirpatil | 78 comments
GitHub repository for MemGPT, an open-source project that gives LLMs self-editing memory management to handle unbounded context.
390 points by ykhli | 102 comments
GitHub repository for AI-town, a deployable TypeScript starter kit for building a virtual town where AI characters live, chat and socialize, backed by Convex.