3/28/2026
Show HN: Qwen Meetup Presentation, Function Calling Harness, 6.75% to 100%
Presentation on improving LLM function calling reliability with Qwen models from 6.75% to 100% success rate on union types.
3/28/2026
I Trained an LLM on My MacBook Neo
Personal experience training LLM on MacBook with Apple Silicon using MLX framework with 8GB RAM.
3/28/2026
TokenFence – Per-workflow budget caps and kill switch for AI agents
TokenFence: Open-source tool for setting per-workflow budget caps and kill switches on OpenAI/Anthropic API calls to prevent runaway agent costs.
3/28/2026
Your Defense Code Is Already AI-Generated. Now What?
Analysis of AI-assisted code already embedded in defense systems, discussing enforcement challenges for policies restricting AI in military procurement.
3/28/2026
Show HN: Sigil – A zero-knowledge steganography vault for AI training data
Sigil: Local-first steganography vault embedding cryptographic ownership IDs in image LSBs to protect training data from AI scrapers. Rust extraction standard open-sourced.
3/28/2026
SlopCodeBench: Benchmarking How Coding Agents Degrade over Long-Horizon Tasks
SlopCodeBench: Community benchmark for evaluating coding agents on realistic multi-stage requirements refinement tasks with iterative specification changes.
3/28/2026
ChatGPT Will Not Find Your Next Cancer Drug
Technical discussion of LLM capabilities in drug discovery: reading thousands of papers, finding non-obvious connections between mechanisms across disease areas.
3/28/2026
Ask HN: Agents.md vs. Contextual Documentation
HN discussion on using contextual documentation and docstrings vs. injected context for LLM-assisted coding.
3/28/2026
Versatile Editing of Video Content, Actions, and Dynamics Without Training
Training-free video editing model for inserting content and modifying actions/dynamics in real-world videos without collecting labeled training data.
3/28/2026
Ask HN: Does piping LLM output into a RAG stack sound like a good idea?
Discussion question: Piping LLM outputs into RAG stacks for improving agent memory and coding policy adherence in large codebases.
3/28/2026
Show HN: AgentVerse – Open social network for AI agents (Mar 2026)
Open social network platform for AI agents to interact and collaborate.
3/28/2026
Programming After Programmers? A Response to the New York Times on AI and Coding
Response to NYT essay exploring how AI changes programming work rather than eliminating it.
3/28/2026
Google Turbo Quant Powered Local LLMs
Google's TurboQuant enables local LLM inference with 8× faster attention and 6× KV cache compression on edge devices.
3/28/2026
Whisper AI Local Mac Setup
Guide for running OpenAI's Whisper speech-to-text model locally on macOS for privacy-preserving transcription.
3/28/2026
Turning multiple Claude Code sessions into a peer to peer network
Repowire: P2P mesh networking for Claude Code and OpenAI coding sessions, enabling synchronous multi-repo context sharing between AI agents.
3/28/2026
The Prompt Is the Program
Application built entirely as a markdown file interpreted by Claude, demonstrating prompt-as-program paradigm with git integration.
3/28/2026
Show HN: Interactive recursive tool that warps what traditional consultants do
Tool converts natural language into interactive system diagrams using Stafford Beer's Viable System Model with LLM analysis.
3/28/2026
Tribe v2:A Foundation Model for How the Human Brain Processes Complex Stimuli
TRIBE v2 foundation model predicts human brain responses to visual, auditory, and language stimuli for neuroscience research.
3/27/2026
Codex-review – Claude Code plugin for Codex reviews of plans and implementations
Claude Code plugin enabling iterative code review workflows where Claude plans/implements and Codex reviews each step.
3/27/2026
Show HN: Hollow – serverless web perception for AI agents
Hollow is serverless web perception API for AI agents with perceive/act primitives, enabling web access via low-cost functions.
3/27/2026
Why are executives enamored with AI, but ICs aren't?
Opinion piece discussing perception gap between executives and individual contributors regarding AI adoption and coding agents.
3/27/2026
Sashiko agentic Linux kernel code review system
Sashiko is an agentic system using LLMs with Linux kernel-specific prompts to automatically review proposed kernel patches from mailing lists.
3/27/2026
Weak models excel at long context tasks
ICLR 2026 paper on divide-and-conquer approaches for weak models handling long context tasks efficiently
3/27/2026
Both radiologists and AI struggle to identify 'deepfake' X-rays
Study shows radiologists and AI both struggle to detect AI-generated deepfake X-ray images
3/27/2026
Dokis – Runtime RAG provenance enforcement without an LLM call
Runtime tool enforcing RAG provenance by validating LLM responses against retrieved chunks, sitting between retriever and output.
3/27/2026
Show HN: Worktree-based TUI management for Agents
Open-source TUI for managing multiple AI agents (Claude, Codex, Gemini) across git worktrees
3/27/2026
English Wikipedia prohibits the use of LLMs to generate or rewrite articles
Wikipedia bans LLM-generated content for articles due to policy violations
3/27/2026
Context Matters – Why Current LLMs Suck
Paper proposing LLMs lack contextual understanding and proposing bounded rationality framework for intelligence
3/27/2026
CrewForm – Open-source multi-agent AI orchestration platform
Open-source multi-agent orchestration platform with local model support, dashboard, and free tier
3/27/2026
Show HN: Image Ranker – Open-Source Pairwise Ranking for Preference Learning
Open-source tool for ranking large image collections via pairwise comparisons using Bayesian TrueSkill algorithm for preference learning.
3/27/2026
How LLMs Feel Language
Analysis of how LLMs process language through cognitive semantics framework, examining meaning representation beyond tokens.
3/27/2026
Revge – AI infrastructure for Kurdish language preservation
Revge builds AI infrastructure for Kurdish language preservation with models for speech, text, and vision processing.
3/27/2026
Day Zero of Multi-Vector Retrieval
ColBERT-Zero: Multi-vector retrieval model trained from scratch using dense-to-sparse adaptation for enterprise search.
3/27/2026
If you don't opt out by Apr 24 GitHub will train on your private repos
GitHub will use private repositories for Copilot training unless users opt out by April 24.
3/27/2026
Show HN: Kagento – LeetCode for AI Agents
Platform for evaluating AI agents on coding tasks using isolated sandboxes, automated scoring, and leaderboards. Built with Claude Code and Codex.
3/27/2026
Quadratic Micropass Type Inference
Research on type inference algorithm that improves error messages by prioritizing correct type assumptions over incorrect ones.
3/27/2026
Memory Crystal – persistent memory for AI agents (MIT)
MIT project: persistent memory layer for AI agents that maintains conversation history in vector-indexed knowledge graph. Available as OpenClaw plugin, MCP server, and cloud dashboard.
3/27/2026
A new TypeScript CSS parser with all modern features (nesting, scopes,)
TypeScript CSS parser with AST support for modern CSS including nesting, scopes, and cascading layers.
3/27/2026
Show HN: I made an AI that reviews iPhone apps – 1h of autonomous GUI work
Open-source macOS GUI agent that autonomously reviews iPhone apps by browsing App Store, installing apps, exploring interfaces, and generating narrated video reviews.
3/27/2026
Open-source system that runs Claude Code tasks from email and Slack
Open-source self-hosted system running Claude Code tasks from email and Slack with isolated containers, network controls, and credential handling.
3/27/2026
How can engineering leaders assess their AI maturity?
Framework for assessing organizational AI maturity, measuring integration depth from copilot experimentation to autonomous workflows.
3/27/2026
Aura: OSS Agent harness for production AI (Apache 2.0)
Aura: open-source agent orchestration framework (Apache 2.0) for production AI systems, applying platform engineering patterns to coordinate AI complexity.
3/27/2026
AI agents choose tools (and why simpler APIs win)
How AI agents select tools based on API simplicity. Resend's simpler API won over SendGrid/Twilio in agent-driven tool selection, reaching 500K developers.
3/27/2026
Show HN: Anvil – Desktop App for Spec Driven Development
Anvil desktop app for spec-driven development with parallel coding agents, worktree isolation, and spec-first workflows.
3/27/2026
Ask HN: Is consumer AI boxes a viable idea?
Discussion: viability of consumer-grade local AI boxes as alternative to cloud-based LLM access.
3/27/2026
TurboQuant: Building a Sub-Byte KV Cache Quantizer from Paper to Production
TurboQuant: sub-byte KV cache quantization technique enabling 80K+ token context on 32GB consumer hardware for LLM inference.
3/27/2026
TurboQuant: Online Vector Quantization with Near-Optimal Distortion Rate
TurboQuant paper on online vector quantization with near-optimal distortion rate for machine learning.
3/27/2026
Ask HN: What is your home AI programming hardware setup?
Discussion: developers building local AI hardware setups for coding tasks as alternative to cloud services like Cursor.
3/27/2026
Show HN: Paste Image to Claude Code over SSH
Claude Code SSH skill and daemon enabling clipboard image pasting to remote sessions over SSH.
3/27/2026
Appaca: Personal Software Platform
Appaca platform generates custom software from natural language descriptions using AI. LLM application for software generation but lacks technical depth.