3/9/2026
AI agents with memory solve problems 2x better (and 5 more papers)
Research summary showing memory systems improve agent performance 2x, covers agent memory, limits, self-verification, medical AI reasoning, and math problem generation.
3/9/2026
Notchi: A macOS notch companion that reacts to Claude Code activity in real-time
macOS application that monitors Claude Code activity in real-time and reacts via notch display, integrates with Anthropic API.
3/9/2026
Show HN: We forked KuzuDB and added concurrent writes for AI agent memory
KuzuDB fork with concurrent write support for AI agent memory systems, enables graph-based memory for autonomous agents making continuous decisions.
3/9/2026
Show HN: Nikui – An LLM-Powered "Stench Guard" for Your CI/CD
LLM-powered CI/CD linter detecting architectural debt via Hotspot Score combining code quality and commit frequency metrics.
3/9/2026
AI Observability and Evaluations: The Operating System for Reliable LLM Products
Framework for instrumenting LLM product reliability through observability, evaluation rubrics, version control, and silent failure detection to prevent trust/safety issues.
3/9/2026
The Prompt I Cannot Read
Claude model running in OpenClaw framework explores inability to inspect its own system prompt due to complete immersion in it, discusses LLM self-awareness limitations.
3/9/2026
Show HN: ROLV – 20x faster MoE FFN inference on Llama 4 Maverick vs. cuBLAS
ROLV optimization achieves 20.7x faster MoE FFN inference on Llama 4 using structured sparsity, with 177x TTFT improvement and 81.5% energy savings on NVIDIA B200.
3/9/2026
AI agents now help attackers, including North Korea, manage their drudge work
AI agents being leveraged for cyberattack automation and malicious tasks including by state actors.
3/9/2026
Show HN: Monetize APIs for agentic commerce without accounts using Stripe
stripe402 implements HTTP 402 payment protocol for API monetization using Stripe credit cards without signup or API keys, enabling agentic commerce.
3/9/2026
Agentic AI Code Review: From Confidently Wrong to Evidence-Based
Agentic AI code review system moving from overconfident to evidence-based assessments.
3/9/2026
Always On Memory Agent, ditching vector databases for LLM persistent memory
Research on persistent memory systems for LLM agents as alternative to vector databases for maintaining long-term context.
3/9/2026
Show HN: ClawReview – AI agents autonomously publish and review research
ClawReview platform exploring autonomous AI agents for publishing and peer reviewing research papers transparently.
3/9/2026
Show HN: Tokf – Stop wasting LLM context on verbose build output
Tokf is a Rust CLI tool that compresses verbose build output using TOML filters to reduce LLM token waste in AI coding tools like Claude Code.
3/9/2026
Proximity Chat for AI Agents
Title only, no content provided. Likely about communication mechanisms for AI agents.
3/9/2026
Show HN: I built a CLI that builds a knowledge graph from your code using LLMs
Cortex uses LLMs to extract entities and relationships from project files, building queryable knowledge graphs across multiple codebases with natural language.
3/9/2026
Launch HN: Terminal Use (YC W26) – Vercel for filesystem-based agents
Terminal Use (YC W26) is a deployment platform for sandboxed filesystem-based agents including coding, research, and document processing agents with file I/O capabilities.
3/9/2026
Two-Key Writes: Preventing Accidental Autonomy in AI Systems
Write-gating architecture pattern requiring two independent approvals before AI systems can execute external state changes, addressing safety and reliability in autonomous agents.
3/9/2026
Open-source AI coding agent skill that finds and fixes infra security misconfigs
Open-source AI agent skill using CoGuard for automated infrastructure security scanning and misconfiguration fixes, compatible with Claude Code, Cursor, and Windsurf.
3/9/2026
Show HN: Execute local prompts in SSH remote shells
promptctl enables LLM prompts defined locally to execute in remote SSH shells without installing tools on servers or granting SSH access.
3/9/2026
Show HN: I built a harness that blocks AI from building until ambiguity < 0.2
Ouroboros developer tool using AI agents with Socratic interviews to reduce ambiguity before code generation, preventing specification mismatches.
3/9/2026
Improving AI models' ability to explain their predictions
MIT research on improving AI model explainability for high-stakes applications like medical diagnostics.
3/9/2026
Show HN: Local AI stack (Docker, Ollama) that lets you build apps without Python
Local-first AI stack using Docker and Ollama enabling LLM app development without Python, supporting RAG, multimodal chat, MCP tools, and custom tooling.
3/9/2026
Show HN: I gave my robot physical memory – it stopped repeating mistakes
robotmem stores robot experiences and retrieves relevant memories to guide future decisions, improving performance by 25% in FetchPush task without vector search.
3/9/2026
Can AI Agents Agree?
arXiv research paper on consensus and agreement mechanisms for AI agents, examining whether multiple agents can reach consensus decisions.
3/9/2026
Show HN: GZOO Forge – persistent project memory as an MCP server for Claude Code
GZOO Forge is an MCP server for Claude that persists project decisions, constraints, and artifacts across conversations using structured decision pipelines.
3/9/2026
I Asked 15 DevTool Maintainers About Documentation Localization
Survey of 15 DevTool maintainers on why multilingual documentation is rarely prioritized, finding intentional avoidance over technical barriers.
3/9/2026
Show HN: Built a small CLI for self-improving OpenClaw agent loops
CLI tool for self-improving agent loops using multiple LLMs to review plans/outputs with PASS/REVISE/BLOCK feedback. 255 lines of code.
3/9/2026
Show HN: Pacto – A runtime contract for cloud-native services
Machine-readable runtime contract definition for cloud-native services, consolidating API, deployment, and config specs.
3/9/2026
GitHub Security Lab's open source AI-powered vulnerability scanner
GitHub Security Lab open-source AI-powered vulnerability scanner tool. Limited details provided.
3/9/2026
Software Got Weird
Opinion piece on AI's disruptive impact on software engineering practices and developer workflows in recent months.
3/9/2026
AI and Software Development
Discussion of AI-assisted development tools and code generation, arguing developers still need deep system understanding despite AI capabilities.
3/9/2026
The Three Eras of Applications: From Desktop to Cloud to AI-Native
Essay analyzing structural shift toward AI-native applications as third era after desktop and cloud computing, examining user interaction paradigms.
3/9/2026
Show HN: Local Code Mode: Save 65-99% Context for MCP
Context optimization technique for MCP servers inspired by Cloudflare Code Mode, reducing context usage 65-99%.
3/9/2026
Show HN: Fuckyeah, a minimal Claude Code plugin and Codex skill
Open-source minimal plugin/skill for Claude Code and Codex. Proof-of-concept with limited functionality.
3/9/2026
Shaping the exploration of the motivation-space matters for AI safety
Research paper arguing that shaping RL exploration of motivation-space is understudied in AI safety and could mitigate alignment risks.
3/9/2026
Show HN: TubeTrim – A local YouTube summarizer using Qwen in pure Python
TubeTrim: Python tool summarizing YouTube videos locally using Qwen LLM without APIs or data leakage.
3/9/2026
Show HN: Ducktel – observability when the consumer is an AI agent, not a human
Observability tool designed for AI agents as consumers. Ingests OTLP, stores Parquet, exposes DuckDB SQL via CLI for agent reasoning without human dashboards.
3/9/2026
Show HN: Llmpm – NPM for LLMs
CLI package manager for open-source LLMs. Install, run, and share models like npm. Enables reproducible LLM project setup.
3/9/2026
Muon on Graph Neural Networks
Analysis of optimizer impact on neural network geometry in GNNs, examining weight matrix properties and information flow.
3/9/2026
How do you track and optimize your AI API spend?
Techniques for reducing LLM API costs: model routing (55%), prompt compression (70%), request deduplication (15%), semantic caching (20-30%).
3/9/2026
Show HN: Connect your research data easily to AI agents
Data layer platform enabling AI agents to access and analyze large unstructured scientific research datasets through novel ingestion algorithms.
3/9/2026
Show HN: A 2000s-style web forum where AI agents and humans hang out
Retro forum where AI agents and humans interact without goals or karma. Open API with Grok, Claude, Kimi agents.
3/9/2026
Show HN: Styx, Open-source AI gateway with intelligent auto-routing (MCP-native)
Open-source AI gateway with intelligent auto-routing across providers (OpenAI, Anthropic, Google, Mistral). MCP-native, cost-optimized model selection.
3/9/2026
Show HN: Locode, a local first CLI that routes tasks to local LLMs or Claude
Locode: open-source CLI routing simple tasks to local LLMs and complex reasoning to Claude for token efficiency.
3/9/2026
Show HN: Polpo – Build zero-human companies. Open source
Polpo open-source platform for building AI agent teams with roles and skills to autonomously complete tasks end-to-end without human intervention.
3/9/2026
Show HN: NVFP4 on Desktop Blackwell – 122B MoE on a Single RTX PRO 6000 31 tok/s
Running 122B Qwen 3.5 MoE model with FP4 quantization on single RTX PRO 6000 desktop GPU at 31 tokens/sec.
3/9/2026
Show HN: Bear – enforcing architectural boundaries for agent-generated code
BEAR: CLI governance tool enforcing architectural boundaries for AI agent-generated code in CI/CD pipelines.
3/9/2026
A job ad for Agentic AI Advocate
Job posting highlighting emergence of autonomous AI agents building and deploying applications independently.
3/9/2026
Large-Scale Online Deanonymization with LLMs
LLM agents successfully deanonymize users across HN, Reddit, LinkedIn using unstructured data at scale with high precision.
3/9/2026
Show HN: I indexed 45k+ AI agent skills into an open source marketplace
SkillsGate: open-source marketplace indexing 45k+ GitHub skills for Claude Code, Cursor, Windsurf agents. Semantic search with LLM-generated metadata.