3/23/2026
Ask HN: If there has been no prompt injection, is it safe?
Discussion of prompt injection security risks when using open-source LLM agents.
3/23/2026
Expose Your Design System to LLMs
Methods for exposing design systems to LLMs for improved code generation and AI integration.
3/23/2026
LLM written FlashAttention: 1.7x PyTorch
LLM-generated FlashAttention implementation achieving 1.7x speedup over PyTorch baseline.
3/23/2026
AgentSmith-HUB update (what changed recently)
AgentSmith-HUB: Security data pipeline with real-time rules engine and integrated LLM agents for SOC operations, log processing, and threat detection.
3/23/2026
Why LLMs stop learning from examples in long contexts: The ICL Collapse
Research on why LLMs fail to learn from in-context examples with long contexts, termed ICL Collapse.
3/23/2026
Agentic Coding Data Dump
LLM coding capabilities analysis covering context engineering, statelessness, and optimization techniques for Claude and similar models.
3/23/2026
Len – types, relations, and generation contracts for LLM codegen
Len framework for type-safe and controllable LLM code generation with generation contracts.
3/23/2026
Show HN: Kage – TUI for managing multiple AI agents with tmux and Git worktrees
Kage: Terminal UI tool for managing multiple AI coding agents in parallel using tmux and git worktrees, enabling concurrent feature development.
3/23/2026
I built a tool that asks developers questions about their own PRs
Developer tool that prompts engineers to explain their own pull request changes, addressing code quality concerns with AI-assisted development.
3/23/2026
Show HN: Oo – compress output for coding agents (cargo test → "47 passed, 2.1s")
Oo compresses verbose test output for coding agents to reduce context window bloat. Open source tool for AI agent development with local model support.
3/23/2026
Fyn: An uv fork with new features, bug fixes, stripped telemetry
Fyn is a fork of uv package manager with additional features, bug fixes, and removed telemetry. Developer tool for Python packaging.
3/23/2026
A Visual Guide to Attention Variants in Modern LLMs
Visual guide explaining attention mechanism variants used in modern large language models.
3/23/2026
Show HN: Chatvas:ChatGPT branches on an infinite canvas
Open-source desktop app enabling branching and visualization of ChatGPT conversations on infinite canvas, automatically mapping conversation structure.
3/23/2026
Snowflake lays off documentation staff after they train AI replacement
Snowflake laid off documentation staff after training them to create AI replacement systems.
3/23/2026
Are AI agents slowing us down?
Analysis of whether AI agents improve developer efficiency, examining Uber's internal AI systems and code generation outputs.
3/23/2026
Claude Usage Monitor – macOS menu bar app to track Claude.ai usage
Native macOS menu bar app for real-time tracking of Claude.ai usage without requiring API key.
3/23/2026
Creating Better Security Guidance and Code with LLMs [video]
Video on using LLMs to improve security guidance and code generation.
3/23/2026
A collection of 35 Golang Agent Skills
Collection of 35 reusable Go-specific instruction sets extending AI coding assistants with domain expertise, loadable on-demand to manage context.
3/23/2026
Writing a 3D Point-Cloud Library in Rust
Modular, high-performance Rust library for 3D point cloud and mesh processing with GPU acceleration support.
3/23/2026
Velr – an embedded property-graph database in Rust on top of SQLite
Velr is an embedded property-graph database written in Rust built on SQLite for local and edge computing use cases, queryable via openCypher.
3/23/2026
I Built an AI Receptionist for a Luxury Mechanic Shop – Part 1
Building an AI receptionist agent for a luxury mechanic shop to handle missed customer calls and lost revenue through automated phone answering.
3/23/2026
High-performance JSON library for Elixir via Rustler NIFs, powered by sonic-rs
High-performance JSON library for Elixir using Rust FFI with SIMD acceleration and selective field extraction API.
3/23/2026
Trepan: Context Drift Catcher
Enforcement layer catching context drift in AI IDE outputs before architectural debt accumulates, offering local-first privacy or cloud-accelerated audits.
3/23/2026
Show HN: Agent-IM – IM, but for Agents
Shared communication interface enabling multiple AI agents to collaborate in a single chat room, allowing agents to see each other's work without human mediation.
3/23/2026
Why using AI to modernize legacy mainframe hairballs is harder than you think
Analysis of challenges using generative AI for legacy mainframe code modernization, examining limitations of COBOL-to-modern-language translation.
3/23/2026
Classical Chinese Jailbreak Prompt Optimization via Bio-Inspired Search
Research using bio-inspired search algorithms to optimize adversarial prompts against AI models, written in Classical Chinese.
3/23/2026
AI cut our driver verification team from 20 to 5
Operational case study: AI reduced driver verification team from 20 to 5 by automating document review and record checking processes.
3/23/2026
Show HN: AutoResearch with PromptFoo Is AutoPrompter
Autonomous system for iteratively improving LLM prompts via closed-loop optimization, combining promptfoo validation with autoresearch refinement logic.
3/23/2026
Show HN: Self-hostable AI agents and internal software
RootCX open-source self-hosted infrastructure for building internal software and AI agents without vendor lock-in or abstraction leaks.
3/23/2026
A terminal-style AI coding agent running on Cloudflare Workers
AI coding agent deployed on Cloudflare Workers with terminal interface, executing code in isolated workers and persisting sessions in SQLite.
3/23/2026
Is the Future of AI Local?
Analysis of open source models running locally on workstations as alternative to centralized datacenter AI, matching frontier model performance within 6 months.
3/23/2026
Show HN: MCP-scan – Security scanner for MCP server configs
Security scanner for Model Context Protocol server configurations detecting leaked secrets, typosquatting, and misconfigurations. Automated GitHub Actions integration.
3/23/2026
MCP Registry – Open-source discovery layer for 20 Model Context Protocol servers
Community-maintained searchable registry for Model Context Protocol servers with structured metadata, addressing discoverability issues as MCP becomes standard for AI agent tool access.
3/23/2026
Product Management on the AI Exponential
Anthropic product lead discusses how Claude's improving model intelligence (Opus 4) affects product management and tool development workflows.
3/23/2026
Token Optimisation with LLMs
Discussion of token optimization strategies for LLMs beyond prompt caching, including handling XML inputs.
3/23/2026
VistAI–Save and Search AI Responses from ChatGPT, Claude, Gemini, and Perplexity
Chrome extension saving and searching responses from ChatGPT, Claude, Gemini, Perplexity locally. LLM application and developer tool.
3/23/2026
LLMs turn an autistic communication style into a neurotypical conversation
Brief note that LLMs convert autistic communication styles to neurotypical patterns. Limited detail, speculative observation on LLM behavior.
3/23/2026
Show HN: I made a AI Code Review tool that knows how your company works
MatrixReview is an AI code review tool that reads company documentation and enforces architectural standards, security policies, and style guides on pull requests automatically.
3/23/2026
Stop spending money on AI credits Translate your content using self-hosted LLMs
Gato AI Translations plugin v17.1 enables self-hosted LLM translation for WordPress via Ollama, supporting ChatGPT/Claude API formats without cloud API costs.
3/23/2026
Emdash: Open-Source Agentic IDE for Orchestrating Multiple Coding Agents
Emdash: open-source IDE orchestrating multiple AI coding agents with Git worktree isolation and issue integration. AI agents, developer tools.
3/23/2026
Neurvance, Pre-cleaned datasets for LLM fine-tuning, free to download
Neurvance: pre-cleaned datasets for LLM fine-tuning, addressing 73% of time spent on data prep. Developer resource for ML applications.
3/23/2026
Show HN: AI Prompts for Data Professionals
Curated prompt library for data professionals organized by role and category. LLM application for data/analytics workflows.
3/23/2026
Prompt to tape out: Autonomous AI agent builds 1.5 GHz RISC-V CPU
Autonomous AI agent designed and built a 1.5 GHz RISC-V CPU from natural language prompt to tape-out.
3/23/2026
Scan for Vulnerabilities with GitHub Security Lab
GitHub Security Lab Taskflow Agent uses AI to identify authentication bypasses, IDORs, and token leaks in open source projects.
3/23/2026
A vetted map of the agentic AI stack (2026)
Reference guide mapping components and tools of the agentic AI stack for 2026.
3/23/2026
Show HN: Agent Kernel – Three Markdown files that make any AI agent stateful
Agent Kernel: lightweight framework using three markdown files to add stateful memory and persistence to AI agents without database.
3/23/2026
I have solved the identity, trust, and memory hurdles with AI Agents
01 Protocol enables persistent, portable identity and memory for AI agents across platforms using cryptographic verification.
3/23/2026
Why (lossy) self-improvement is real but it doesn't lead to fast takeoff
Analysis of recursive self-improvement and fast takeoff in AI, discussing industry consolidation and impact on research jobs.
3/23/2026
The Generative AI Policy Landscape in Open Source
Overview of generative AI policies and licensing approaches across open source projects.
3/23/2026
Show HN: AddFox – Build Better Browser Extensions for Developers and AI
AddFox: developer tool for building browser extensions faster using Rsbuild with hot reload and multi-browser support.