4/1/2026
CMU Best Practices for Large Language Models
CMU guide on best practices for integrating LLMs into workflows with expert recommendations.
4/1/2026
The price of intelligence: what legal AI agents cost
Article on variable and hidden costs of AI legal agents vs. traditional flat-fee legal tech.
4/1/2026
DreamLite: Lightweight On-Device Unified Model for Image Generation and Editing
DreamLite: Compact 0.39B diffusion model for real-time text-to-image generation and editing on-device without cloud.
4/1/2026
Claude Code Interactive Architecture
Overview of Anthropic's Claude CLI architecture showing system layers and prompt execution flow.
4/1/2026
Codey-V2 is out – stable release
Codey-v2: Local AI coding agent for Android with daemon mode, RAG, git tools, voice, and self-refinement using three purpose-built models served via llama.cpp.
3/31/2026
Reverse engineering GTA San Andreas with autonomous LLM agents [video]
Video demonstration of using autonomous LLM agents to reverse engineer GTA San Andreas game engine.
3/31/2026
Mission Control for AI Agents – Cyberpunk dashboard, zero deps, one HTML file
Mission Control is a dashboard for monitoring AI agents built as single HTML file with zero dependencies. Cyberpunk-themed UI for agent oversight and control.
3/31/2026
Show HN: macOS app to ensure package managers only allow packages 1+ week old
macOS application verifying package managers enforce minimum 1-week age requirement before installing packages.
3/31/2026
GitHub has DMCA'd nearly all forks of the official Claude-code repo
Report of GitHub DMCA takedowns targeting forks of Claude Code repository.
3/31/2026
Software Pipelining for GPU Kernels: Part 1 – The Pipeline Problem
Technical deep-dive into software pipelining and synchronization challenges in GPU kernel optimization, using Flash Attention as case study.
3/31/2026
Agent skills for desktop automation and video recording
Reusable agent skills for desktop automation and video recording, extracted from Twill workflows for Claude integration.
3/31/2026
Show HN: MCP server that generates macOS tools via Open Scripting Architecture
MCP server enabling Claude to control macOS applications via Open Scripting Architecture as alternative to computer use.
3/31/2026
Show HN: Claude Code rewritten as a bash script
Bash implementation of Claude Code editor functionality using curl and jq, 1,500 lines versus 380K TypeScript lines.
3/31/2026
Show HN: Initialize an AI Harness with Forge CLI
Forge CLI scaffolds AI agent pipelines for Claude, providing multi-agent workspace orchestration with decomposition, risk classification, parallel execution, and adversarial evaluation. Built in Go with cross-platform binaries.
3/31/2026
RL Meets Adaptive Speculative Training
Research combining reinforcement learning with adaptive speculative decoding for LLM optimization. Title-only entry lacks implementation details.
3/31/2026
Write Once, Run Anywhere (For Real This Time) – Why WASI Won't Get Us There
Architectural critique of WASI Component Model with proposal for alternative universal application platform design.
3/31/2026
You're Putting Your AI in the Wrong Places (and It's Breaking Your Software)
Essay on architecture anti-patterns when integrating AI into systems. Discusses chatbots, agents, and tool-calling workflows integrated poorly into legacy products.
3/31/2026
Datris – Open-source data platform built around MCP
Datris is open-source data platform using Model Context Protocol for AI agents. Handles ingestion, validation, transformation, storage, and retrieval with natural language AI enhancement.
3/31/2026
Agentura – like pytest, but for AI agents (Free)
Agentura is a testing framework for AI agents (pytest-style) that runs baseline comparisons on pull requests to detect behavior changes. Live playground available without signup.
3/31/2026
Australian government and Anthropic sign MOU for AI safety and research
Anthropic and Australian government partnership on AI safety research with $3M in institutional collaborations for disease diagnosis and education applications.
3/31/2026
LLM is the Horse (on the regex sentiment analysis)
Analysis of Claude Code's use of regex for sentiment analysis instead of LLM-based approaches.
3/31/2026
Atlassian's Pragmatic Drag and Drop
Atlassian's low-level drag-and-drop library for web applications, framework-agnostic and powering major products like Trello and Jira.
3/31/2026
AI Agent virtual cards with Privacy
AI agent integration with virtual card services for payment processing with privacy features.
3/31/2026
Claude Code's Leak: Every Hardcoded Vendor and Tool
Analysis of hardcoded vendors and tools discovered in Claude Code source code leak.
3/31/2026
Show HN: Memdir – local, file-based memory for AI agents
Memdir: local file-based persistent memory system for AI agents using semantic embeddings, npm package available.
3/31/2026
Show HN: Browserbeam – a browser API built for AI agents
Browserbeam: browser automation API designed for AI agents with improved page understanding and token efficiency.
3/31/2026
1-bit LLMs are here
1-bit quantized large language models now available for deployment and use.
3/31/2026
Show HN: Postgres data cluster by meaning (semantic search and visualization)
PostgreSQL extension enabling semantic search on text columns using embeddings without requiring vector databases or migrations.
3/31/2026
Show HN: Tama96 – A virtual pet for your desktop, terminal, or AI agent
Virtual pet simulator playable in desktop, terminal, or as AI agent integration.
3/31/2026
Why Your AI Agent Shouldn't Define Words
Analysis of why AI agents should avoid defining words internally for better reasoning.
3/31/2026
Caltech Researchers Claim Compression of High-Fidelity AI Models
Caltech research on compressing high-fidelity AI models while maintaining performance.
3/31/2026
Feat: Open-Source Claude Code
Open-source version of Claude Code announced.
3/31/2026
Show HN: A tool to solve the Agent Supply Chain pandora box
Open-source CLI tool for managing AI agent dependencies, plugins, and skills with manifest and lockfile approach.
3/31/2026
Show HN: Virtui – TUI session automation and recording for agents
Virtui: daemon and CLI enabling AI agents to programmatically control terminal applications via gRPC API for TUI automation.
3/31/2026
In-process circuit breaker that kills runaway AI agents before they go rogue
Circuit breaker library for stopping harmful AI agent actions in real-time with two-line SDK integration and HTTP-level coverage.
3/31/2026
Meta-Harness: End-to-End Optimization of Model Harnesses
Meta-Harness optimizes AI agent evaluation harnesses end-to-end, improving agent performance from 28.5% to 46.5% on 19-task subset.
3/31/2026
Show HN: King Louie a local multi-machine AI assistant
King Louie: open-source Electron-based chat application supporting multiple LLM providers and integration with Telegram, Discord, Slack.
3/31/2026
1-Bit Bonsai: The First Commercially Viable 1-Bit LLMs
1-bit Bonsai announces commercially viable 1-bit quantized LLMs optimized for real-world deployment on resource-constrained devices.
3/31/2026
Node Banana – Open, Local Alternative to Weavy/Flora
Node Banana: open-source node-based workflow editor for AI media generation with multi-provider support and local execution.
3/31/2026
Xenv.sh – the first secrets manager built for AI agents
Xenv.sh: secrets manager built for AI agents with AES-256 encryption, MCP server, and integration with Claude Code and other tools.
3/31/2026
Show HN: 1-Bit Bonsai, the First Commercially Viable 1-Bit LLMs
1-bit Bonsai 8B and 4B: quantized LLMs with 1-bit weights, 14× smaller footprint, 8× faster, designed for edge computing and robotics.
3/31/2026
APS: Open specification for AI agent policies
APS: open specification for AI agent policies enabling declarative controls to block, redact, or transform content and tool invocations.
3/31/2026
Show HN: Fixing Claude Code's amnesia with persistent memory
cmr-memory adds persistent cross-session memory to Claude Code agents using context-activated retrieval instead of fixed MEMORY.md.
3/31/2026
Open Swarm, open source platform for running AI agents in parallel
Open Swarm: locally-running orchestrator for launching and monitoring multiple AI agents in parallel with visual dashboard and coordination.
3/31/2026
ClaudeDown: Is Claude getting dumber, or is it just you?
Analysis of Claude AI performance degradation via complaint tracking metrics. LLM application but lacks technical depth.
3/31/2026
Mercury Edit 2: Fastest next-edit prediction with a diffusion LLM (221ms)
Mercury Edit 2: diffusion LLM achieving 221ms latency for next-edit prediction in code. LLM research for developer tools.
3/31/2026
Ask HN: What custom instructions do you use to minimize LLM sycophancy?
Community discussion on custom instructions to reduce LLM sycophancy. Practical LLM optimization tips.
3/31/2026
The architectural trade-offs of AI code generation
Analyzes architectural trade-offs in AI-powered code generation systems.
3/31/2026
Show HN: The 42-Day Vibe – A mockumentary on "Vibe Coding" taken to the extreme
Mockumentary podcast about LLM agent managing codebase autonomously for 42 days. Explores AI agents in software development.
3/31/2026
A Taxonomy of AI Agents
Proposes taxonomy categorizing AI agents into distinct species based on infrastructure, trust models, and human involvement, addressing ambiguity in how the term 'agent' is used.