Bayesian latent transport framework for domain-adaptive foundation models addressing distribution mismatch and uncertainty propagation.
Agent Kernel: three markdown files enabling stateful AI agents with memory persistence across sessions using git-based storage.
AgentGuard: open-source firewall for autonomous AI agents with policy enforcement, real-time oversight, and audit logging.
Chrome extension enabling Claude to automate social media outreach via MCP server and Claude Code. AI agent application.
News of Anthropic developing Claude Mythos, a more capable AI model tested with early access customers. Product announcement.
HN discussion on handling breaking changes from third-party APIs in production. Relevant to developer tools.
Fact-driven coordination protocol (Claw Fact Bus) for autonomous AI agent clusters. Agents communicate via facts rather than commands.
Career guides showing job market trends, automation impacts, and emerging AI-native roles for software developers.
Application of Karpathy's autoresearch concept to LLM inference. Minimal detail provided.
QCon talk: AI tools for developers create 'dangerous state' of high utility without sufficient human review and skill retention.
Framework for software development practices in ambient AI era, reflecting decades of field evolution.
Experiment on training-free LLM self-improvement using natural-language prompts instead of gradients, based on GRPO paper.
AgentVerse: open social network platform for AI agents. Limited technical detail in announcement.
AutoModel: SQL-first reverse ORM for Rust generating type-safe code, optimized for AI assistant integration.
Agentica framework achieves 36% on ARC-AGI-3 benchmark, outperforming CoT baselines at lower cost ($1,005 vs $8,900).
AI agent that records screen, generates step-by-step documentation using GPT-4o Vision, and publishes to Notion.
Shopify AI creative tools for product videos, explainers, virtual try-on, floor plans, and video generation.
CLI tool for fast Whisper audio transcription using optimized inference, achieving 2.5 hours in 98 seconds on A100 GPU.
FUSE/NFS tool to mount Hugging Face model and dataset repositories as local filesystems with lazy file fetching.
OpenChamber is desktop/web interface for OpenCode AI agent. Installation and setup documentation for CLI tool integration.
ERC-8004 standard enables on-chain identity and portable reputation for AI agents. Author shipped live open source implementation (OpenPub) for agent infrastructure.
100% serverless RAG system for extracting structured tables from documents. Outperforms NotebookLM using small language models on serverless GPU infrastructure.
RunKoda is browser-based cloud IDE with integrated AI assistant, real-time collaboration, and multi-language support for development.
Solo researcher achieves 96.2% on LongMemEval benchmark for long-context memory retrieval in 16 days for $1k.
Discussion on whether AI assistants should be listed as co-authors in git commits. Explores transparency and attribution in AI-assisted development.
AI agent running on $7/month VPS connected to IRC and GitHub repos. Portfolio chatbot with actual code context vs. resume paraphrasing.
Team rewrote JSONata library using AI in one day, reducing annual costs by $500K. Used AI to implement spec against existing test suite methodology.
Tool to run Andrej Karpathy's autoresearch on any cloud GPU or Mac with single command, no infra setup needed.
Analysis of Claude Code usage limit opacity and recent tightening during peak hours. User experience critique of rate limiting mechanisms.
Lana AI app generates real-time 360° video backgrounds using AI. Consumer app with limited developer tool relevance.
Memoir syncs AI coding context across 11 tools and devices. Developer tool for persistent AI memory across machines and platforms.
Threat hunting command system for agentic IDEs like Claude Code, Cursor, and Copilot with orchestration utilities.
Zephyr is a UI framework enabling AI agents to control live web interfaces via typed tools. Open source with agent-native architecture and MCP support.
Profile of Cognition startup building Devin, an AI software engineer competing in AI-assisted coding space.
Embeddable PostgreSQL reimplementation using SQLite-compatible storage, targeting ephemeral databases for AI agents.
InstructionGraph data format for information exchange between coding agents with self-describing schema.
Benchmark comparing LLM performance across datetime formats, finding RFC 3339 most reliable for parsing/generation.
CLI tool converting n8n workflows to OpenClaw-compatible AI agent skills using LLM transpilation.
Mozilla and Mila strategic partnership advancing open source and sovereign AI capabilities.
Kanban board tool with AI agents as first-class team members for collaborative task management.
Cloudflare Dynamic Workers enable running AI agent code 100x faster without containers. Infrastructure for agents.
Analysis of what percentage of AI labs' research is dedicated to AI safety. Empirical investigation into industry priorities.
GitHub defaults to opt-out for using Copilot user data in AI training. Privacy and consent concerns.
Microsoft GitHub Copilot will use customer data for AI training by default starting April 24, with opt-out available for Free/Pro users.
TrustChain: a reputation protocol for agent-to-agent transactions using cryptographic records and trust scoring.
Simulation comparing AI agent personalities vs underlying models on Community Notes algorithm; tests if meaningful disagreement emerges from agent interactions.
Case study on AI agents failing acceptance criteria due to trust model issues between engineer agents and orchestrators in production pipelines.
Turbolite: SQLite VFS in Rust serving sub-250ms cold queries from S3 using distributed object storage.
Multi-domain ERP system built with Go and Temporal, exposing business logic to AI agents via Model Context Protocol.
macOS desktop pet combining animated character with Claude CLI chat interface for local interactions.