HN schacon 7/9/2026

Keppsake – automatic save points for your AI projects

Tool for automatic checkpointing of AI projects on macOS, allowing rollback when AI generation goes wrong.

HN rubenglez 7/9/2026

Show HN: LangDrift – test AI agents across languages

Testing framework for evaluating AI agent behavior consistency across languages, checking tool selection and arguments in multiple locales.

HN stared 7/9/2026

The true cost of saying "Hi" to an AI agent

Analysis of hidden costs in AI agent interactions, calculating human waiting time overhead alongside token costs.

HN gaemi 7/9/2026

Show HN: Agentic FC – a football management SIM played by AI agents over MCP

Football management simulation where AI agents control teams through MCP, with terminal-based match commentary and ASCII visualization.

HN lokinkl 7/9/2026

Show HN: Internationalizationstack – Fix AI Localization Bugs in Claude Code

CLI tool suite and Claude Code skills for fixing AI localization bugs, providing format engineering, QA testing, and native-quality translations for 46 languages.

HN antondd 7/9/2026

Chasing the Hallucinations: KPMG's AI-Powered Attempt at "Redefining Excellence"

Investigation using hallucination detection tool to uncover fabricated citations in AI-generated government reports, academic papers, and consulting firm research.

HN kizum 7/9/2026

97% of websites expose zero tools an AI agent can use

Analysis of nearly 10,000 websites showing 97% expose no tools for AI agents to use, highlighting gap between agent-ready interfaces and current web design.

BL 7/9/2026

ChatGPT is now a partner for your most ambitious work

ChatGPT Work agent that takes cross-app actions and manages multi-hour projects autonomously.

HN grauk 7/9/2026

Relm – local LLMs as base-R objects, with interpretability

R package with Rust core embedding local LLMs, exposing generation, embeddings, activation tracing, steering and ablation as base-R functions for mechanistic interpretability research.

HN theanonymousone 7/9/2026

Grok 4.5 Benchmark Results

Benchmark results for Grok 4.5 proprietary LLM showing performance metrics, context window (500k tokens), and multimodal capabilities.

HN jcwsw129 7/9/2026

Show HN: Chat Privacy – hide AI chat history while screen sharing

Browser extension hiding AI chat history during screen sharing to prevent accidental exposure of conversation data.

HN oxplot 7/9/2026

Tooling for AI agents to build KiCad project from simple circuit description

Tool converting circuit descriptions and component assets into KiCad projects, enabling AI agents to work on hardware design without parsing complex file formats.

HN baranul 7/9/2026

AI as a weapon of mass cognitive destruction

Opinion piece criticizing overuse of AI-generated content in business communications, discussing quality degradation from excessive output.

HN vforno 7/9/2026

Show HN:Getting GLM 5.2 running on my slow computer

GLM 5.2 model successfully runs on consumer hardware with capabilities comparable to Claude/GPT while avoiding out-of-memory errors.

HN rar101x 7/9/2026

Probelock – lockfile for LLM tool calling

Probelock is a lockfile mechanism designed for LLM tool calling, enabling safer function invocation in LLM-based agents.

HN uukelele 7/9/2026

Former GitHub CEO launches competitor designed for the age of vibe coding

Former GitHub CEO launches Entire, a Git hosting network optimized for AI agents and code generation workflows.

HN metricspaceai 7/9/2026

Greppy – A drop-in grep with code-nav subcommands for AI agents

Greppy extends grep with code-navigation subcommands optimized for AI agents exploring codebases.

HN vinothkumarnaga 7/9/2026

Cybersecurity AI (CAI) Dataset

Cybersecurity AI (CAI) Dataset released for training and evaluating ML models on security tasks.

HN Krishnaswaroop 7/9/2026

What founders should evaluate before launching an AI-built app

Evaluation framework for founders launching AI-built applications; practical considerations for AI product launches.

HN cinooo 7/9/2026

AI changes the economics of software rewrites

Analysis of how AI models improve code rewriting economics for popular tech stacks due to training data prevalence and codebase context.

HN cooleel 7/9/2026

Accelerating Harbor with Tensorlake

Tensorlake runs Docker images for Harbor on microVM sandboxes, passing Terminal-Bench 2.1 tasks with cold starts in seconds.

HN Azimuth05 7/9/2026

Compendium – Shared workspace for teams and AI agents

Compendium is a shared workspace tool designed for teams and AI agents collaboration.

HN Garbage 7/9/2026

How to Kill the Bloat in Claude Code's System Prompt

Guide to removing bloat from Claude Code's system prompt, reducing token payload by tens of thousands per request through context optimization.

HN uncleteslim 7/9/2026

Reproducing an Indirect Prompt Injection Against a RAG Pipeline

Title indicates research on indirect prompt injection attacks against RAG pipelines.

Ax Andrey Podivilov, Vadim Lomshakov, Sergey Savin, Matvei Startsev, Roman Pozharskiy, Maksim Parshin, Sergey Nikolenko 7/9/2026

AgentLens: Production-Assessed Trajectory Reviews for Coding Agent Evaluation

Production-assessed benchmark for code agents evaluating entire trajectory including instruction following, tool use, error recovery, and communication.

Ax Yotam Wolf, Noam Wies, Amnon Shashua 7/9/2026

When Does In-Context Search Help? A Sampling-Complexity Theory of Reflection-Driven Reasoning

Theoretical analysis of in-context search as approximate inference over reasoning traces, studying sampling complexity of reflection-driven LLM reasoning.

Ax Sifat Afroj Moon, Dakotah Maguire, Adam Spannaus, Joe Tuccillo, Maksudul Alam, Sudip K. Seal, John Gounley, Heidi Hanson 7/9/2026

LLM-powered reasoning in agent-based modeling

Hybrid approach combining LLMs with agent-based modeling to enable real-time adaptive decision-making in large-scale individual interaction simulations.

Ax Bayram Yuksel Eker, Suayb S. Arslan, Ozgur Nazli, Mustafa Serhat Demirgil, Furkan Deligoz 7/9/2026

QANTIS: Hardware-Calibrated Sequential POMDP Belief Updates on IBM Heron

Uses quantum processor as calibrated belief-update service for sequential POMDP belief updates in autonomous systems under partial observability.

Ax Kabir Moghe, Peter Chin 7/9/2026

Cost-Effective Agent Harnesses for Abstract Reasoning and Generalization on ARC-AGI-1

Studies open-weight DeepSeek V3.2 model on ARC-AGI-1 benchmark using strict compute budgets without test-time scaling or fine-tuning.

Ax Pavel Snopov, German Magai 7/9/2026

Evaluating SageMath-Augmented LLM Agents for Computational and Experimental Mathematics

ReAct-style agent combining LLM reasoning with SageMath CAS for verifiable feedback on research-level computational and experimental mathematics problems.

Ax Muayad Sayed Ali, Aliaksandra Novik, Anji Boddupally, Artem Yavorskyi, Chris Nickerson, Daniel Rica, Emily DuGranrut, Felix Leung, Garrett Prince, Grace Barnett, Heath Robinson, Hosain Al Ahmad, Jesse Resnick, Juan Carlos Farah, Jyothi Swaroop Meruga, Leonid Kuznetsov, Luke Gorham, Marie Schmoll, Michael Paciullo, Saumya Das, Sharath Sheripally, Tommy Griscom, Mykyta Osadchyi, Neha Mantri, Nick Westrum, Olivia Benowitz, Parikshith Kulkarni, Radik Chernyshov, Rakshith Vasudev, Rohith Nadimpally, Vikas Gangadevi, Waseem AlShikh 7/9/2026

The Harness Effect: How Orchestration Design Sets the Token Economics of Enterprise Agentic AI

Analyzes token economics of enterprise agentic AI, arguing orchestration design (harness layer) is key lever against token inflation per task.

Ax Yufeng Wang, Lu Wei, Haibin Ling 7/9/2026

Grounding Spatial Relations in a Compact World Model: Instruction Leakage and a Goal-Free Dynamics Fix

Identifies instruction leakage problem in goal-conditioned world models for spatial reasoning and proposes goal-free dynamics fix for true perception grounding.

Ax Wachiravit Modecrua, Krittin Pachtrachai, Touchapon Kraisingkorn 7/9/2026

Large Behavior Model: A Promptable Digital Twin of the Retail Customer

Large Behavioral Model learns customer decision-making from retail transactions using Person-Environment formulation for grounded behavior modeling.

Ax Yi Yang, Siyuan Liu, Xin Gao, Huamu Sun, Chao Liu, Qing Zhou, Bingbing Nie 7/9/2026

Learning social norms enhances compatibility in dynamic human-AI coordination

Studies how AI agents including LLMs learn social norms to improve human-AI coordination in dynamic interactions through implicit shared expectations.

Ax Jerry Han, Rafael Moschopoulos, Ella Colby, Vishrut Goyal, Andrew Tu, Kia Ghods, Mark Braverman, Elad Hazan 7/9/2026

Measuring Intelligence Beyond Human Scale

Proposes relative-scale evaluation paradigm where AI models generate adversarial challenges to measure intelligence beyond human-saturating benchmarks.

Ax Lifei Liu, Haoran Yu, Xiaochong Jiang, Su Wang, Pin Qian, Yihang Chen 7/9/2026

Operational Reframing and Approval-Framed Delegation in Multi-Agent LLM Safety

Analysis of multi-agent LLM safety systems decomposing pipeline effects into three mechanisms: reframing harmful intent, planner refusal, and executor delegation.

Ax Ethan Chung, Chuanjun Zheng, Jasper Tan, Jingxi Li, Haopeng Zhang, Huaijin Chen 7/9/2026

Does AI Understand Imaging? A Systematic Benchmark of Agentic AI for Computational Imaging Tasks

ImagingBench: benchmark evaluating vision-language models and agentic AI on 20 computational imaging tasks across optics, signal processing, and inverse problems.

Ax Silvia Santano 7/9/2026

Reasoning Consistency Scanning: A Framework for Auditing Chain-of-Thought Validity in AI Safety Evaluations

Framework for detecting logical inconsistencies in chain-of-thought reasoning of LLMs without requiring controlled interventions on evaluation transcripts.

Ax Haipeng Ding, Yuexiang Xie, Zhewei Wei, Yaliang Li, Bolin Ding 7/9/2026

From Atomic Actions to Standard Operating Procedures: Iterative Tool Optimization for Self-Evolving LLM Agents

Agents optimize static atomic tool actions into reusable Standard Operating Procedures to reduce reasoning overhead and failure rates.

Ax Diab W. Abueidda, Bilal Ahmed, Panos Pantidis, Mostafa E. Mobasher 7/9/2026

Physics-Audited Agentic Discovery in Scientific Machine Learning

PA-SciML: verification-first workflow for LLM agents discovering surrogate models in scientific ML with physics-based auditing.

Ax Charbel Al Bateh, Samer Saab Jr 7/9/2026

MIRA-Math: A Benchmark for Minimal Information Requesting and Mathematical Reasoning

MIRA-Math benchmark for mathematical reasoning where solvers must request one missing atomic fact to solve problems.

Ax Elaine Ang, Chenxi Huang, Georgios Liargkovas, Jerry Liu, Jinhui Liu, Nikos Pagonas, Charlie Summers, Haonan Wang, Jiakai Xu, Tianle Zhou, Yusen Zhang, Zhou Yu, Zhuo Zhang, Tianyi Peng, Kostis Kaffes, Eugene Wu 7/9/2026

Agentic Data Environments

Agentic Data Environments framework for autonomous agents operating across files, APIs, applications, and system state with failure bounds.

Ax Vikas Reddy, Sumanth Reddy Challaram, Abhishek Basu 7/9/2026

Reason Less, Verify More: Deterministic Gates Recover a Silent Policy-Violation Failure Mode in Tool-Using LLM Agents

Deterministic gates detect silent policy violations in tool-using LLM agents where forbidden state transitions execute successfully.

Ax Xing Zhang, Yanwei Cui, Guanghui Wang, Ziyuan Li, Wei Qiu, Bing Zhu, Peiyang He 7/9/2026

The Blind Curator: How a Biased Judge Silently Disables Skill Retirement in Self-Evolving Agents

Shows biased reward judges silently disable skill retirement in self-evolving LLM agents, causing library drift below baseline performance.

Ax Songhan Wang, Haoang Chi, He Li, Zhiheng Zhang, Jiayan Yuan, Cheems Wang, Hao Peng, Xinwang Liu, Wenjing Yang 7/9/2026

SpaCellAgent: A Self-Evolving LLM-Based Multi-Agent Framework for Trajectory Analysis

SpaCellAgent: LLM-based multi-agent framework for autonomous trajectory inference analysis in spatial and single-cell transcriptomics.

Ax Dmitry Beresnev, Vladimir Makharev, Roman Khalikov, Ivan Oseledets, Petr Anokhin 7/9/2026

Search, Fail, Recover: A Training Framework for Correction-Aware Reasoning

Pyligent framework for training correction-aware reasoning in LLMs: validated search over partial solution chains with failure recovery.

Ax Wei-Jung Huang 7/9/2026

Do LLM-Generated Skills Make Better AI Data Scientists? A Component Ablation Across Data-Science Workflows

Compares LLM-generated vs expert-written reusable skills for AI data scientists across data cleaning, SQL, statistical testing, and formatting tasks.

Ax Azwar Abdulsalam, Nishil Patel, Andrew Saxe 7/9/2026

RL Post-Training Builds Compositional Reasoning Strategies

RL post-training enables Transformers to compose primitive skills into higher-level reasoning strategies beyond base model capabilities.

Ax Mingguang Chen, Licheng Wang, Bo Qu 7/9/2026

Recursive Self-Improvement in AI: From Bounded Self-Refinement to Autonomous Research Loops

Survey of 1,250 papers on recursive self-improvement in AI systems: self-refinement, self-reward, self-play, and autonomous research loops.

Ax Tianming Sha, Yue Zhao, Lichao Sun, Yushun Dong 7/9/2026

SkillCenter: A Large-Scale Source-Grounded Skill Library for Autonomous AI Agents

SkillCenter: open-source library with 216,938 structured skills for autonomous AI agents across 24 domain bundles with source grounding.