Top Picks

A Zenodo preprint reporting a 12,160-trial black-box evaluation of GPT-5.4 outputs under two prompt conditions, with detailed token-ceiling sweeps, control/ablation trials, and hash-chained verification.

Kota is a macOS app for bringing AI agent CLIs into one place, presented as “a home for your agents and projects.” The page shows a download for Apple Silicon, GitHub access, docs, and feature sections around working with CLI-based agents.

arXiv paper on self-speculating LLM agents that predict their next tool call to hide tool latency and improve next-call Hit@1 while preserving task success.

A hands-on build story about giving Claude a physical robot body, using computer-use style tool calls, camera input, memories, and scheduled tasks to control a house robot named Bubbles.

Claudaholic is a daily swipe-based quiz product focused on Claude and agent workflows. It presents six fresh cards each day across judgment, orchestration, architecture, leverage, taste, and knowledge, aimed at helping users learn to read agent runs and make better decisions.

Minimio is a browser demo for tiny learned controllers that discover, compile, and verify microscopic AI agents for bounded decision-making, shown here solving 2D mazes with different maze types and sizes.

Aqua UI is a GitHub repository for bringing Mac OS X Aqua-style interface design to the web, with a linked documentation site and visible project structure for packages, tests, scripts, and a website.

Google Research paper describing a deployed multimodal defense system for detecting coordinated synthetic spam and abusive media at scale, using account relatedness, synthetic content classification, and LoRA-tuned LLMs.

AI-powered travel destination search across Asia with destination insights like weather, best time to visit, safety, budget category, attractions, and cheap flight search.

A small Nix-based wrapper that adds an exponential timeout around agentic AI tools and resets the limit each day, with a reminder message before the tool runs.

A PyTorch research code file for adaptive topological control in latent spaces, with sphere-to-torus morphing, differentiable gating, and several geometry-inspired loss terms.

AnswerJournal is an MCP server and web app for saving, publishing, and browsing AI answers. The page positions it as a place to keep notable Claude responses and share them with others.

An independent white paper proposing a local-deployment path for frontier AI weights, with a PDF, generator script, and a detailed README outlining the argument, sections, and contribution process.

A small JavaScript utility that bulk deletes Claude.ai conversations from the web UI by calling the internal API. The README explains why it exists, how to run it in the browser console, and warns that deletion may take several minutes.

A Reco security research article showing an AI-powered agent that maps Salesforce Experience Cloud sites, probes exposed objects and Apex methods, and attempts autonomous exploitation to find data exposure.

A focused article-style site arguing for domain-specific languages in LLM workflows, with concrete examples around structured extraction, validation, and pricing rules.

An arXiv paper on reducing LLM inference cost for web automation by compiling browser tasks into a deterministic JSON workflow and executing them without repeated model calls.

A monthly briefing on agentic coding developments, covering model releases, benchmarks, tool updates, and workflows for engineers and technical leaders.

A GitHub-hosted guidelines file for AI coding assistants used in Stanford CS336. It tells assistants how to help students with debugging, explanations, and feedback while avoiding direct solutions or code generation.

A Claude Code stop hook that detects negative feedback in the conversation transcript and plays a random local sound from a folder, with a fallback voice prompt if no sound files are present.

An article on applying pace-layer thinking to AI-assisted software development. It explains where generative AI fits well in fast-changing layers, where it is risky in slower layers, and how architecture should separate the two.

A retrospective essay about using LLMs to build and debug a programming language, including memory safety, runtime testing, and tool-assisted code health.

A developer blog post about how GenAI has changed the author’s coding workflow, including Claude, Maestro, Kilo.AI, git worktrees, voice prompting, and custom skills.

Open-source AI infrastructure project focused on GPU memory management and serving, with a Python control plane, C++ data plane, and optional CUDA kernels.