6/22/2026
What it takes to get high Text-to-SQL accuracy in production
Techniques and strategies for achieving high accuracy in Text-to-SQL systems used in production environments.
6/22/2026
Sakana Fugu multi-agent system delivered as one model
Sakana Fugu: multi-agent system delivered as single model, dynamically orchestrates models for complex tasks via API.
6/22/2026
Knowledge Catalog – universal context engine for agents
Google Knowledge Catalog: AI data catalog with knowledge graph providing context and metadata management for AI agents.
6/22/2026
Ask HN: How do you make the LLM generate good code?
Discussion on techniques for getting LLMs to generate higher quality code for specific tasks.
6/22/2026
TIRx: An Open Compiler Stack for Evolving Frontier ML Kernels
TIRx open-source compiler for ML kernels supporting GPU and AI accelerators, handling agent-generated code.
6/22/2026
PsychAdapter: Personality in LLM output via trait-language patterns, not prompts
PsychAdapter: transformer architecture using trait-language patterns to reflect personality and mental health in LLM outputs.
6/22/2026
Meta pauses AI training program tracking employee keystrokes after internal leak
Meta paused internal AI training program after data leak exposed employee conversations and performance data across company.
6/22/2026
Show HN: Device emulation outside QEMU using vfio-user and libvfio-user
vfio-user protocol enables device emulation in separate process outside QEMU. Allows flexible implementations like SPDK for multi-VM disk handling.
6/22/2026
Lines or Less: Test Case Minimization
Property-based testing library implementing test case minimization in minimal code.
6/22/2026
Sakana AI Ships Fugu, an Orchestration Model Claiming Fable 5 Performance
Sakana AI releases Fugu orchestration model achieving performance comparable to Fable 5.
6/22/2026
Show HN: Git Issues – versioned task management for AI agents
Version control task management system designed for AI agent workflows.
6/22/2026
Japan's 'Sakana Fugu' multiagent AI scores well against Fable 5, GPT 5.5
Sakana AI's Fugu multiagent model demonstrates competitive performance against Fable 5 and GPT 5.5.
6/22/2026
The comment paradox: clean code in age of AI agents
Technical discussion on code comments as implicit prompts for AI coding assistants.
6/22/2026
Show HN: Hybrid search of 540K+ US Government datasets on 2 CPU cores. No LLMs
Search system for 540K+ US government datasets using hybrid search (traditional + semantic) without LLMs. Runs on 2 CPU cores.
6/22/2026
Show HN: A 100% AI-generated 3D voxel game I built with my 10yo son
Open-source 3D voxel game entirely AI-generated with documented workflow.
6/22/2026
Ask HN: What are some rock solid open source vector databases
Community question seeking recommendations for open-source vector databases.
6/22/2026
AI Agent / Harness Engineer
Research on Concept Modulation Models framework addressing identifiability and extrapolation in conditional latent variable models and causal representation learning.
6/22/2026
SubQ claims it broke through a bottleneck that's holding back LLMs
Subquadratic startup claims breakthrough in subquadratic bottleneck limiting LLM efficiency. Independent evaluation results shared.
6/22/2026
Company Brain – Open-Source
Open-source internal AI assistant that scrapes and indexes organizational data from Google Workspace, HubSpot, Airtable. Privacy-first copilot for unified memory.
6/22/2026
I built a peer-to-peer bridge for AI agents to talk locally and across the web
Project: peer-to-peer bridge enabling AI agents to communicate locally and across the web.
6/22/2026
Make Your Copilot Credits Count
Guide on GitHub Copilot AI Credits for students: explains how chat messages, agent tasks, and model calls consume credits to optimize usage.
6/22/2026
Theta: Declarative, harness-agnostic configuration standard for AI coding agents
Theta-spec: declarative, harness-agnostic configuration standard for AI coding agents. Single theta.toml file defines instructions, rules, tools, skills, and subagents with protocol for lifecycle management.
6/22/2026
Show HN: Open-geo – Does ChatGPT or Google's AI Overview cite your brand?
Tool measuring brand visibility in AI-generated answers across ChatGPT, Perplexity, Gemini, Claude, Google AI Overview and others.
6/22/2026
AI Agent Can't Log into Anything
Analysis of AI agent limitations in authentication and login automation tasks.
6/22/2026
Show HN: AI-Gateway – Open-source semantic caching proxy to reduce LLM API costs
AI-Gateway: open-source semantic caching proxy for reducing LLM API costs through intelligent request deduplication.
6/22/2026
AI compressed the 15% of our website rebuild that was typing, not the other 85%
Case study: website rebuild using Claude Code generated 24,296 lines across 120 React components. Analysis of where AI compressed workflow and which parts remain human-only.
6/22/2026
Show HN: Subconscious and GLM-5.2 Makes "/compact" Obsolete
GLM-5.2 model paired with Subconscious Cache for agentic tasks with long-term memory management beyond context window limits.
6/22/2026
Zero Weights Graph Language Engine (MSE-GLM)
MSE-GLM: interpretable graph-based language model alternative to standard neural networks. Provides explainability and deterministic output without floating-point weights.
6/22/2026
Show HN: Open-source job search plugin for Claude Code
Open-source Claude Code plugin that automates job search by asking preference questions and comparing live job postings from LinkedIn.
6/22/2026
Prompt Injection as Role Confusion
ICML 2026 research paper explaining prompt injection attacks as role confusion in LLMs, enabling new attack predictions and mechanistic interpretability insights.
6/22/2026
Iris – A portable runtime for durable AI agents
Iris is a portable runtime for building durable AI agents with persistence and portability across environments.
6/22/2026
The Optimal Amount of Slop Is Non-Zero
Blog post discussing risk-proportional rigor for LLM-generated code, examining when skipping human review is acceptable.
6/22/2026
Show HN: Vivijure – Self-hosted AI film studio on your own GPU (AGPL)
Vivijure: Self-hosted AI video generation studio running on personal GPU. AGPL licensed open source tool.
6/22/2026
Super AI Agentic Android App (BYOK)
Agentic Android app with memory, knowledge base, UI automation, MCP server support, local model compatibility. Bring-your-own-key architecture.
6/22/2026
Phlox: A full-featured AI platform you own
Phlox: Full-featured open source AI platform for self-hosting. Complete AI infrastructure alternative.
6/22/2026
I Tracked LLM Pricing for 8 Weeks. Here's What the Data Shows
Empirical pricing data from 1,111 LLM models across 33 providers collected daily over 8 weeks. Analysis of shifting pricing landscape.
6/22/2026
Show HN: Frontier AI for Motion Design
Motion: AI agent for motion design. Takes prompts with context, creates videos, orchestrates explainers and animations.
6/22/2026
Show HN: PreFlight – A local AST scanner to catch AI architectural drift
PreFlight: Local AST scanner catching security issues in AI-generated code. Detects auth, SQL injection, SSRF, secrets before commit.
6/22/2026
Show HN: Build and Host AI apps on your own servers
Open-source framework for building and hosting AI applications on user-controlled servers, supporting open models and avoiding third-party dependencies.
6/22/2026
Prompt Injection as Role Confusion
ICML 2026 research paper on prompt injection attacks driven by role confusion in LLMs, with mechanistic interpretability implications.
6/22/2026
A running list of reasons to move to open source
List of reasons to adopt open source. General advocacy without technical details.
6/22/2026
Show HN: Oak – Git replacement designed for agents
Oak: Git replacement version control system designed specifically for AI agents, reducing context overhead and enabling parallel work without full repo copies.
6/22/2026
Show HN: FastUbu – An Ultrafast Video Archive
FastUbu: AI-indexed video archive of 30-year old film library. Uses Midjourney-level processing for transcription and video indexing.
6/22/2026
How to build an AI agent in 2026: a practical step-by-step guide
Comprehensive guide to building production AI agents in 2026. Covers tool design, policy enforcement, cost control, testing, audit trails with examples.
6/22/2026
Two Heads Are Better Than One: Run Many AI Agents, Merge One Auditable Result
Multi-agent approach: run parallel AI agents and merge results for auditable outcomes.
6/22/2026
AI Agent Governance vs. Observability: What's the Difference?
Conceptual distinction between governance and observability for AI agents; limited technical depth.
6/22/2026
Show HN: Sturnus – OpenAI-compatible LLM proxy routing to the fastest provider
Sturnus: OpenAI-compatible proxy that routes LLM requests to fastest provider for cost/speed optimization.
6/22/2026
Nff_: Open-source Claude Code for hardware
nff MCP server enabling LLMs to control hardware via USB, flash devices, and diagnose failures autonomously.
6/22/2026
Bain tests software takeover targets by vibecoding AI replicas
Bain uses AI to simulate software company acquisitions; vague on methodology.
6/22/2026
Show HN: Export YOLO/RF-DETR Models to ONNX, TensorRT, CoreML – 100% Local
Desktop studio exporting Ultralytics YOLO and Roboflow RF-DETR models to ONNX, TensorRT, CoreML with 100% local processing.