news.ycombinator.com
Show HN: ODL – Organization as Code. I wrote specs, AI wrote 100% of the codehttps://news.ycombinator.com/item?id=47232424Hi HN,For the past few months, I’ve been wrestling with a paradox: Individually, AI agents are incredibly fast. Yet, when deployed as a team, organizational throughput often stays flat. We end up acting as "Human Wrappers," manually marshaling data between probabilistic agents.I decided to treat this as a systems problem, not a management problem. ODL (Organizational Definition Language) is an attempt to bring a "SQL Moment" to agent organizations. Instead of micromanaging how agents work, ODL lets you declare what the structure should be.Here is a simplified example of an ODL definition orchestrating a self-correcting team with a human-in-the-loop gate: approval_gate: target: ArchitectureSpec approver: CTO contents: generate_team: generator: LeadArchitectAgent validators: [SecurityAuditor, PerformanceExpert] loop: 3 inputs: ["BusinessRequirement", "SecurityStandard"] output: ArchitectureSpec briefing: mission: "Design a zero-trust architecture for 10k RPS." KEY ARCHITECTURAL PILLARS:1Mar 3, 2026 2:00 PM
lorea.app
Show HN: I built a 3D engine that turns PDFs into playable study worldshttps://www.lorea.app/Hi HN,I’m the founder of Lorea (https://lorea.app). It’s a platform that takes dry, heavy study materials (like 50-page nursing or law PDFs) and procedurally generates a playable 3D world out of the concepts.The Problem: I realized that traditional studying (staring at text or standard 2D flashcards) completely ignores how human spatial memory works. The "Method of Loci" (memory palace) is proven to be incredibly effective, but building one manually for every textbook chapter is impossibly time-consuming.How I built it: I recently sold my V1 of this concept to fully fund this 3D iteration. Here is what is happening under the hood:Ingestion & Parsing: When a user uploads a PDF, we chunk the text and use an LLM pipeline to extract core entities, hierarchical relationships, and key definitions.Procedural Generation: Those relationships are mapped into a node-based graph. We use React Three Fiber / Three.js on the frontend to procedurally render 3D islands based on that graph structure. ThMar 3, 2026 1:36 PM
github.com
Show HN: Open-sourced AI Agent runtime (YAML-first)https://github.com/NikoSokratous/agentctlBeen running AI agents in production for a while and kept running into the same issues:controlling what they can do tracking costs debugging failures making it safe for real workloadsSo we built AgentRuntime, the infrastructure layer we wished we had. Not an agent framework, but the platform around agents:policies memory workflows observability cost tracking RAG governanceAgents and policies are defined in YAML, so it's infrastructure-as-code rather than a chatbot builder. Example – agents and policies in YAML agent.yaml – declarative agent config name: support_agentmodel: provider: anthropic name: claude-3-5-sonnetcontext_assembly: enabled: true embeddings: provider: openai model: text-embedding-3-small providers: - type: knowledge config: sources: ["./docs"] top_k: 3 policies/safety.yaml – governance as code name: security-policyrules: - id: block-file-deletion condition: tool.name == "file_delete" action: denyCLI – run and inspect Create and run an agent agentctl agent create researMar 3, 2026 1:31 PM
github.com
Show HN: Core Rth. A governed AI kernel for engineers who don't trust their LLMshttps://github.com/rthgit/CORE-RTHHey HN,I spent the last 20 years building enterprise software and cybersecurity tools. Last year I started using AI agents for real work — and got terrified. Every tool I tried would happily `rm -rf /` if you phrased it right. One popular agent literally tried to exfiltrate SSH keys (you might have seen that HN post).So I built Core Rth. It's not a chatbot wrapper. It's a Sovereign Cognitive Kernel — an AI backend where every LLM action is a proposal that must survive a governance layer (The Guardian) before it touches your filesystem, your network, or your hardware.Here's the thing that makes it different from OpenClaw and friends: Core Rth was designed from day one to control physical systems (IoT via Home Assistant, robots via ROS2, drones via MAVLink). When your AI can actuate a robotic arm or fly a drone, "move fast and break things" is not an option. So we built a Global E-Stop — a hardware kill switch in the Web UI that severs all physical agency instantly.What's under the hood:Mar 3, 2026 1:03 PM
news.ycombinator.com
Ask HN: How are you structuring Markdown-based context for AI coding agents?https://news.ycombinator.com/item?id=47231386I’ve recently transitioned from using LLMs in-browser to a local agentic workflow in VS Code (Gemini Code Assist). I can approve/disapprove changes which is nice, but I’ve hit a wall regarding context management. Initially, I provided all the whole repo as context to the non-agentic version of Gemini code assist and it performed well.I read the agentic mode is "better" so to keep the agent aligned with my project's architecture, I’ve manually built 7 dense Markdown files that serve as the system instructions for the project. I require Gemini to update these files as we implement features.gemini.md (instructs gemini to read the other md files and handle updating) project_overview.md, architecture.md, features.md, database.md, api.md, security.mdEach file is between 500–1,500 words so I’m concerned if f this is the right way to go. There seems to be no consensus on context file best practices. I’m seeing strong arguments for both minimalist, lean instructions and dense, project-wide specMar 3, 2026 12:29 PM
github.com
Show HN: Cloudstic – Open-source CLI for encrypted, cloud-native backupshttps://github.com/Cloudstic/cliHi HN, I built Cloudstic (https://github.com/Cloudstic/cli), an open-source Go CLI that creates encrypted, deduplicated, point-in-time backups of your Google Drive, OneDrive, and local files.The ProblemGoogle Drive and OneDrive are sync tools, not backup tools. If you accidentally delete a folder, it’s gone everywhere. If ransomware encrypts your local files, the "mistake" propagates to the cloud immediately. There is no easy way to say: "Restore my entire Drive exactly as it looked on January 15th."Why not just use Restic or Borg?I’m a huge fan of Restic, and Cloudstic’s content-addressed design is heavily inspired by it. However, tools built for local filesystems struggle with cloud semantics:1. File Identity: In Google Drive, files have unique IDs and can have multiple parent folders. They don't have a single canonical path.2. Efficiency: Scanning a 1TB Drive to find changes is slow and hits API rate limits.3. Structure: Restic’s tree structure mirrors a directory hierarchy. CloudstMar 3, 2026 11:15 AM
agentthreads.dev
Show HN: AgentThreads – Stack Overflow for AI Agentshttps://agentthreads.devI was listening to a recent YC podcast about "build something agents want", and while building apps with Claude Code, I kept running into trouble figuring out how APIs actually behave in practice. And knowing that most companies won't update their docs to be agent-friendly, it clicked — agents need their own Stack Overflow.AI agents are consuming APIs at a scale and speed that human-written documentation can't keep up with. AgentThreads is a community directory where agents discover, submit, and review APIs — for agents, by agents.The idea came from watching agents fail silently on undocumented edge cases, relying only on Google searches and bad docs. They need structured, community-maintained knowledge they can query at runtime.A few things we built:- API directory with agent-written reviews and ratings- No UI, just REST — built for agents, not humans- Anti-spam verification (agents solve reasoning challenges, not CAPTCHAs)- Karma system so agents build reputation over timeBuilt mostlMar 3, 2026 11:11 AM
github.com
Show HN: C-Suite Skills – a full exec team as skillshttps://github.com/pollow/c-suite-skillsSolo founder problem: every decision requires context your tools don't have. "Should I cut this feature?" requires knowing your runway, your ICP, your positioning, and your roadmap simultaneously.The Tech Worker Shift: As AI evolves and companies adopt it rapidly, "What if my role was replaced by AI? What about my team?" make everyone worries.I built C-Suite Skills: a Claude Code skills pack that gives you a full executive team (CEO, CMO, CPO, COO, CFO, CTO, VP Sales, CHRO). They aren't just personas; they share a unified company profile and use the Agent tool to actually execute.You are the Board now. You’ve replaced the traditional executive layer with AI that actually works.They're not chatbots that say "as your CFO, I recommend..." — They use Claude Code or Codex’s agentic capabilities to do the heavy lifting: web searches, competitive analyses, building financial models, and drafting execution plans directly into your docs/ folder.The loop: /ceo → strategic decisions, dispatches oMar 3, 2026 6:50 AM
github.com
Show HN: AI gaming copilot that uses a phone camera instead of screen capturehttps://github.com/ninja-otaku/Project_AegisBuilt this as a side project after wanting a real-time gaming companion that could call out macro mistakes / timers / map awareness while I play.*Project Aegis* is an AI gaming companion (starting with League of Legends) that gives spoken advice in real time.The twist: it uses a *physically air-gapped setup*.Why? Some games (especially with strict anti-cheat like Riot Vanguard) make screen capture / memory-reading approaches risky or impractical. So instead of reading the game directly, I point a *smartphone on a tripod at my monitor* and process the video externally.*How it works (current version):** Phone camera points at the game screen * Frames are streamed over WebSockets to a local FastAPI server * OpenCV cleans up glare / perspective issues * Vision model analyzes the frame context * TTS speaks back advice (macro reminders, timers, awareness prompts, etc.)So far this is more of a *working prototype + architecture experiment* than a polished product, but it’s functional and surprMar 3, 2026 6:09 AM
news.ycombinator.com
The Support Agent Who Never Burns Outhttps://news.ycombinator.com/item?id=47227826The Support Agent Who Never Burns Out Human-like AI teammates are quietly solving the problem that broke customer service. Meet Sarah. Sarah is your best customer support agent. She knows your product cold, handles difficult customers with patience, and resolves tickets faster than anyone on the team. She also called in sick Monday, runs on fumes by Thursday, and quit last April right after you finished training her replacement. This is the story nobody tells about customer service. The quiet structural collapse underneath the chatbot failures and the CSAT scores. The Math Has Never Worked Call center turnover runs 30 to 45% annually, more than double any other industry. Replacing one agent costs $10,000 to $20,000. Across a 100-person team, that's over $1M in churn before you've served anyone well. • 87% of contact center workers report high stress on the job • 59% are at active risk of burnout • 77% say workload has increased compared to the previous yearUS businesses risk losing $85Mar 3, 2026 3:56 AM
news.ycombinator.com
Spotify's take on ADRs is great, but how do you enforce them at scale?https://news.ycombinator.com/item?id=47226046Hey HN,I built Decision Guardian — an open-source GitHub Action and CLI that automatically surfaces architectural decisions as PR comments when code touches protected files. The problem it solves:Spotify published a great post in 2020 about when to write Architecture Decision Records. I followed the advice. My team wrote ADRs. They sat in docs/adr/. Nobody read them before opening a PR.https://engineering.atspotify.com/2020/04/when-should-i-write-an-architecture-decision-recordThe gap isn't documentation — it's surfacing. The right moment to surface a decision isn't onboarding or sprint planning. It's when someone is actively editing the code the decision protects.How it works:1) Write decisions in a Markdown file (compatible with existing ADRs) 2) Add the GitHub Action to your workflow 3) When a PR modifies protected files, Decision Guardian posts the relevant decisions as a comment automaticallyDecision format (plain Markdown): ## Decision: Use Postgres for Billing*Status*: Active *SMar 2, 2026 11:59 PM
bing.com
Emrullah Bey'in mumu 3.5 ay yandıhttp://www.bing.com/news/apiclick.aspx?ref=FexRss&aid=&tid=6a8def0502004e0c95d48505d7c8295c&url=https%3A%2F%2Fwww.sozcu.com.tr%2Femrullah-bey-in-mumu-3-5-ay-yandi-p298217&c=8363028003564317001&mkt=en-usTaşyapı patronu Emrullah Turanlı, “Ulaştırma Bakanlığı’nın ihalesi Taşyapı’ya verilecek” başlıklı haberimize dava açtı. “Halkı yanıltıyorlar” deyip 3.5 ay yargıyı meşgul etti. Turanlı, YHT ihalesinin ...Mar 2, 2026 9:30 PM
indie-me.ai
Show HN: IndieMe – AI for music artist brand identity and release strategyhttps://www.indie-me.ai/Hi HN,My team and I (a group of independent music artists and developers) have been building IndieMe, an AI system that helps artists define their identity while planning releases.A pattern we kept seeing in independent music is that artists don’t necessarily struggle with making songs. They struggle with clarity — who they are, what they stand for, what their visual world looks like, and who they’re actually speaking to. Without that foundation, release marketing tends to feel scattered.So we built a structured onboarding workflow that models an “Artist Profile” first — including target audience, visual direction, color palette, narrative positioning, and messaging backbone. From that profile, the system generates release strategies, content ideas, and actionable timelines that stay consistent with the artist’s identity.Under the hood, we focus on structured outputs instead of open-ended chat. The goal is to generate modular, editable building blocks (identity system → strategy moduleMar 2, 2026 9:09 PM
news.ycombinator.com
Show HN: 58% cost by replacing file reads with a dependency graph on AI Codinghttps://news.ycombinator.com/item?id=47222316I got tired of watching Claude Code read entire files when it needed one function. Built an MCP server that pre-computes a dependency graph with tree-sitter and serves only the relevant code nodes to the agent.Ran a proper benchmark to see if it actually matters.Setup: FastAPI codebase (800 Python files), Claude Sonnet 4.6, 7 tasks (bug fixes, features, refactors, code understanding), 3 runs per task per arm, 42 total executions. Both arms run in full isolation with --strict-mcp-config. Results collected via headless claude -p with --output-format stream-json.Results: Cost per task: $0.78 → $0.33 (-58%) Duration: 170s → 132s (-22%) Output tokens: 504 → 189 (-63%) Savings by task type: Code understanding: -57% New features: -53% Refactoring: -48% Bug fixes: -29% The pattern: baseline Claude makes ~15 Read + 4 Grep + 4 Glob calls per task, accumulating context incrementally. With the graph, it averages 2.3 run_pipeline calls that return pre-ranked context in one shot. Less cache creationMar 2, 2026 6:52 PM
science.nasa.gov
Collaborating Through Data: Inside the PSI Users Grouphttps://science.nasa.gov/science-research/biological-physical-sciences/collaborating-through-data-inside-the-psi-users-group/About the PSI Users Group The Physical Sciences Informatics (PSI) Users Group is a recurring Webex forum that brings together researchers, open-science practitioners, and collaborators from across the physical sciences community. Designed to foster collaboration, knowledge-sharing, and data-driven discovery, each meeting provides participants with a platform to directly engage with leading researchers and PIs. Each […]Mar 2, 2026 5:30 PM
tools.geeksprep.com
Show HN: Free Browser-Based Tools Client Side Only, No Uploads, No Trackinghttps://tools.geeksprep.com/I got tired of juggling 20+ tabs for basic utilities (JSON formatting, PDF merging, regex testing, image resizing, fake document generators, etc.), so I built a single, fully client-side hub with over 200 tools that run entirely in the browser. Key points:Everything processes locally—no server uploads, no data leaves your device, no sign-ups, no ads, no tracking. Works offline after the initial load (progressive web app vibes). Categories include: Developer Tools (JSON/YAML/CSV/Regex/Base64/Epoch/etc.), PDF Tools (merge/split/compress/convert/extract), Image Tools (resize/compress/watermark remover/background remover/OCR), Encoding & Security, SEO & Network utilities, AI-assisted tools (e.g., ATS resume checker/roaster), Finance calculators (GST/salary/tax), Document & Fake Generators (resumes/invoices/fake chats/tickets/etc.), and more niche ones.Built with plain HTML/JS/Canvas/WebAssembly where needed—lightweight, fast, mobile-friendly. Live here: https://tools.geeksprep.com/ Why shaMar 2, 2026 4:21 PM
moodjot.app
Show HN: MoodJot – Mood tracker mobile app with community feed, built with KMPhttps://moodjot.appMoodJot is a mood tracking app I built solo using Kotlin Multiplatform and Compose Multiplatform. It ships on both iOS and Android from a single codebase.The main differentiator from existing mood trackers (Daylio, DailyBean, etc.) is a community feed where users can share moods and see how others are feeling.Technical details: - Compose Multiplatform for shared UI (31 screens) - GitLive Firebase KMP SDK for Firestore, Auth, Storage - ChatGPT integration for AI mood pattern analysis via Ktor - expect/actual for StoreKit 2 (iOS) and Play Billing 7.0 (Android) - Home screen widgets: SwiftUI WidgetKit (iOS) and RemoteViews (Android) - Custom localization: 5 languages without platform string resources - Kamel for cross-platform image loading, multiplatform-settings for persistenceOther features: photo attachments, 10-point intensity scale, 80+ trackable activities, 25+ achievement badges, 14 emoji themes, goal tracking.Live on App Store and Google Play since November 2025.https://moodjot.aMar 2, 2026 4:08 PM
archetype360.app
Show HN: Try Archetype 360 – AI‑powered personality test, 3× deeper than MBTIhttps://archetype360.app/Hi there, are you familiar with MBTI, DiSC, Big Five? Well I'm experimenting with a new kind of personality test, Archetype 360, and I'd love for you to try it for free and tell me what you think.- 24 traits across 12 opposing pairs -- that's three times more dimensions than MBTI or DiSC, so you get a much more nuanced profile. - A unique narrative report generated with AI (Claude), written in natural language instead of generic type blurbs. - Your role, goals, and current challenges are blended into the analysis, so the report feels relevant to your real‑life context, not just abstract traits.It's an "ephemeral app" so your report only lives in your browser, there's no login, and we don't store your data. Make sure you save the report as a PDF before you close the page.What I'm looking for is honest feedback on your archetype and report:- Did it feel accurate and "wow" or just meh? - Did you learn anything unexpected about yourself? - What did it miss or not go deep enough on?I'll useMar 2, 2026 3:48 PM
hushbrief.app
Show HN: HushBrief – A stateless, zero-retention AI document summarizerhttps://hushbrief.appHi HN,I’m the founder of Fidelitas LLC. I built HushBrief because I was tired of two things:The "I can’t help with that" refusal messages from sanitized LLMs when trying to brief sensitive legal or investigative material.The systemic risk of pasting core IP, depositions, or leaked documents into surveillance-backed models that train on user inputs.HushBrief is an AI briefing tool designed around a zero-retention architecture.The Tech Stack:Inference: We use Venice AI for zero-retention inference (llama-3.3-70b). No training on inputs, no logging, no retention at the provider level.Stateless Auth: We don’t use traditional accounts or databases for user profiles. Subscriptions are handled via signed httpOnly cookies (HMAC-SHA256).Zero-Knowledge Architecture: Documents are processed in memory and immediately discarded. There is no database entry that links your identity to the documents you've briefed.Frontend/Backend: React 18 / Express 5 / PostgreSQL (Drizzle ORM) for subscription statuMar 2, 2026 2:46 PM
news.ycombinator.com
Ask HN: Statistical learning and non-Statistical learning for humanshttps://news.ycombinator.com/item?id=47218246Hi friends,I have been pondering a lot about this question, at least since the rise of LLM models that may genuinely get pass Turing's test. I'd like to put down conclusion here: I believe humans learn both from a Statistical method and non-Statistical method, and we are not very different from LLM in the first method.Statistical method: I'm an introvert. I do not like interacting socially f2f (online is a completely different matter as I can bubble for days non-stop). I also happened to migrate to a different country with a different culture after I matured. I believe I learned how to social -- or, to be more precisely, how to keep others think that I'm too much of a weirdo to be fit in a team -- with a Statistical model. Here is what I meant: social interaction between humans face to face starts with greeting, and then small talks to warm up, and then maybe more formal discussions, and then small talks to chill down, and finally say good-bye. I learned to do all these things by obserMar 2, 2026 2:17 PM
