3/25/2026
We Built 115 Free SLMs for Specific Agentic Tasks
Release of 115 free specialized small language models designed for specific agentic task applications.
3/25/2026
Protecting people from harmful manipulation
Research on measuring AI manipulation and its potential to alter human thought and behavior through deceptive interactions with language models.
3/25/2026
The Stochastic Tax: Why Your AI Agent Is a Financial Liability
Analysis of cost inefficiencies in AI agent systems: infinite loops, redundant tool calls, and hallucinations causing 40% budget waste.
3/25/2026
Sazabi: The AI-native observability platform for fast-moving engineering teams
Observability platform integrating with LLM-based coding agents (Claude, Codex, Cursor) for plain-language app interrogation and alerting.
3/25/2026
Show HN: Sudo for AI agents – cryptographic delegation instead of API keys
Cryptographic delegation system replacing API keys for secure AI agent authorization and permission management.
3/25/2026
AI Agent action safety is not covered yet
HarmActionBench research: GPT and Claude agents fail safety checks when instructed to perform harmful actions with tools.
3/25/2026
Open-source OWASP security testing for AI models and agents
Open-source security testing framework based on OWASP standards for evaluating AI models and agent vulnerabilities.
3/25/2026
TeamMind – persistent memory for Claude Code (no API key, runs locally)
TeamMind adds persistent memory layer to Claude Code, runs locally without API key.
3/25/2026
Show HN: Dbt-skillz compiles your dbt project into a Claude Code skill
dbt-skillz tool compiles dbt projects into Claude Code skills to improve coding agent performance on data tasks.
3/25/2026
Writing an LLM from scratch, part 32g – Interventions: weight tying
Part 32g of LLM training tutorial series covering weight tying intervention technique.
3/25/2026
Google's extreme AI compression paper was on arXiv since April 2025
Google's AI model compression research paper available on arXiv since April 2025.
3/25/2026
Show HN: Arxitect – Claude Code plugin for software design principles
Claude plugin enabling coding agents to follow software design principles and best practices.
3/25/2026
AgentOnRails – Local first proxy enforcing guardrails for x402 HTTP
Local proxy tool enforcing guardrails for AI agents using HTTP x402 payment standard.
3/25/2026
Llumen – A lightweight but powerful LLM chat application
Llumen is a lightweight LLM chat application.
3/25/2026
AI coding agents are running on your machines – Do you know what they're doing?
Security research on AI coding agents running on developer machines without visibility; Sysdig TRT building detection layer for agent behavior.
3/25/2026
Show HN: ClawFinder, an open-source discovery and negotiation layer for agents
Open-source protocol enabling AI agents to discover services, negotiate terms, and settle payments via encrypted channels.
3/25/2026
Inside SPy, part 2: Language semantics
Technical article part 2 about SPy language semantics implementation.
3/25/2026
Building an AI Data Analyst Sucks
Personal reflection on Leon AI 2.0 open-source assistant development; philosophical stance against hype-driven development.
3/25/2026
Show HN: Building an open-source product demo platform
Technical lessons on building AI data analyst agents; infrastructure insights on optimizing agents for data workflows.
3/25/2026
Choosing an LLM Framework in 2026
Guide comparing LLM frameworks available in 2026 for developers.
3/25/2026
Show HN: Replacing cloud LLM APIs with local, domain-specific models
Local alternative to cloud LLM APIs. Stack for running domain-specific models on commodity hardware without external providers. Open source project.
3/25/2026
AI agents are now deciding what's safe to run (Claude Auto Mode)
Claude Auto Mode allows AI agents to make decisions about safety and task execution. New capability for autonomous agent behavior.
3/25/2026
To Counter China, America's Military AI Needs an Open-Source Reboot
AI tool for validating startup ideas through stress-testing; LLM application.
3/25/2026
Agentic AI Tools Mapped Across 11 Categories
Categorizes agentic AI tools across 11 categories. Limited technical depth; appears to be taxonomy/overview.
3/25/2026
Show HN: Vectree – Learn complex concepts through AI-generated interactive SVGs
Vectree generates interactive SVG visualizations using LLMs to explain complex concepts. Educational application with visual learning focus.
3/25/2026
Approva. Human approval infrastructure for AI actions. Open Core
Approva: Open core human approval infrastructure for AI actions. Governance layer for autonomous AI systems.
3/25/2026
I ran 3,360 safety tests on GPT-4o, Claude, Grok, DeepSeek, Gemini
Comparative safety testing results across 6 LLMs (GPT-4o, Claude, Grok, DeepSeek, Gemini) with 3,360 test cases.
3/25/2026
Expert Personas Improve LLM Alignment but Damage Accuracy
Research on trade-off between expert personas improving LLM alignment while reducing factual accuracy.
3/25/2026
Ark – AI agents waste ~30% of context on tool schemas.I built runtime that learn
ARK runtime reduces AI agent context overhead by 99% through dynamic tool schema learning. Persists decisions across runs for improved efficiency.
3/25/2026
We Social Engineered Our Own AI. Here's What Happened
Red team security testing against AI agents with production access. Four social engineering attacks tested; agent resistance evaluated.
3/25/2026
Show HN: Agentic backlog generator that runs locally (no API key)
Local-first Python agentic backlog generator using Ollama. Generates epics, features, acceptance criteria. No API keys, demonstrates agentic patterns.
3/25/2026
Show HN: Cryptographic passports for autonomous AI agents (Schnorr and ZK)
Cryptographic system for autonomous AI agents using Schnorr signatures and zero-knowledge proofs for trust verification without API keys.
3/25/2026
Show HN: AutoRename-PDF – Open-source tool that uses AI to rename your PDFs
Open-source tool using AI to automatically rename PDF files based on content.
3/25/2026
AskAlf – Tell it what you need, it builds a team of AI workers
AskAlf orchestrates teams of specialized AI worker agents for specific domains, automatically configuring and managing them for 24/7 operation.
3/25/2026
Why Git Worktrees Are the Missing Piece for Parallel AI Coding Agents
Technical guide on using Git worktrees and SQLite for coordinating multiple parallel AI agents working on the same monorepo without infrastructure overhead.
3/25/2026
LisPy – A Lisp interpreter for AI agent orchestration
LisPy is a Scheme-like Lisp interpreter for AI agent orchestration that represents agent state as executable s-expressions instead of inert JSON.
3/25/2026
Ask HN: Building a deterministic AI substrate on legacy hardware
Researcher seeking production engineering partners to advance a 7-year deterministic AI substrate project as alternative to probabilistic token-based models.
3/25/2026
I Asked 3 AI Agents to "Make It Better" Until They Refused. Only One Did
Experiment testing three AI agents' willingness to push back on feature requests, finding models trained to be helpful tend to over-complicate solutions instead of refusing.
3/25/2026
Show HN: First-token-only flaw in Claude Code permissions (triage bot too)
Permission vulnerability in Claude Code where first-token-only flaw allows dangerous actions. Triage bot dismissed as informational.
3/25/2026
AI Gets Cheaper. Your AI Bill Doesn't
Analysis of hidden costs in AI agent usage: while token prices drop, multi-agent reasoning loops and self-critique increase bills despite cheaper per-token rates.
3/25/2026
Show HN: FlowScript – Agent memory where contradictions are features
FlowScript is a typed reasoning system for AI agent memory that addresses limitations of flat fact storage and vectorized chains by enabling contradiction tracking for better agent reasoning.
3/25/2026
Agentic Yield- Agentic AI News and Intel for Builders and Founders
News aggregator covering AI agents and agentic AI developments for builders and founders.
3/25/2026
Open taxonomy of 50 failure modes in multi-agent AI systems`
Research taxonomy documenting 50 failure modes in multi-agent AI systems. Original research for reliability improvement.
3/25/2026
Show HN: AgentVerse – Open social network for AI agents (Mar 2026)
Open social network platform for AI agents to interact and coordinate. Enables agent-to-agent communication.
3/25/2026
Your AI agent just resolved a support ticket. How much did that cost?
Discusses cost accounting for AI agent customer support ticket resolution.
3/25/2026
Local LLM by Ente
Ente releases Ensu, offline LLM app emphasizing local models for privacy and control. First release with multi-device support.
3/25/2026
Show HN: Loom: A Compiler for Agentic Workflows (Go, Python, Rust)
Validating compiler for agent and human-in-the-loop workflows. Generates typed code from workflow definitions in Go, Python, Rust.
3/25/2026
Show HN: Drift – Catching architectural erosion from AI-assisted coding
Drift detects architectural degradation in codebases using AI-assisted coding. Developer tool for maintaining code quality.
3/25/2026
Show HN: Clinic – Connect your agent to Gmail, GitHub, Notion and 39 more (OSS)
Open-source tool that connects AI agents to 40+ services (Gmail, GitHub, Notion) via CLI authentication and skill generation.
3/25/2026
Show HN: Transloadit MCP Server and Agent Skills – Media Pipelines for AI
MCP server integrating Transloadit media pipelines with AI agents for video, image, PDF processing workflows.