AI News Today

Long-horizon reasoning exposes a core weakness in AI agents: context windows fill up fast, and retrieval pipelines return noise instead of signal. To solve this, researchers at the National University of Singapore developed MRAgent , a framework that abandons the static "retrieve-then-reason" approach. Instead, it uses a mechanism that allows an agent to dynamically develop its memory based on accumulating evidence. This multi-step memory reconstruction is integrated into the reasoning process o

Article URL: https://old.reddit.com/r/MacStudio/comments/1ugkygr/due_to_the_ai_memory_crisis_apple_is_bringing/ Comments URL: https://news.ycombinator.com/item?id=48693024 Points: 1 # Comments: 0

Comments

https://archive.md/ArXuF https://www.nbcnews.com/tech/tech-news/us-government-gives-a... Comments URL: https://news.ycombinator.com/item?id=48692995 Points: 544 # Comments: 757

Exactly what the title says. I've recently realized I rarely need something as powerful as GPT5 (or as expensive!). GPT-4o-mini has been doing the job well. Curious to see if OSS has caught up to this model in anyone's experience. Comments URL: https://news.ycombinator.com/item?id=48692961 Points: 2 # Comments: 1

we built an agent plugin that lets an agent reflects after real work saves the reusable part as a "skill", or improves one it already has. No command, no trigger. inspired by hermes agent that improves itself automatically by creating skills i've used it for a few days and i've felt the magic already - 4 skills created for me, all look reasonable. more and better skills means faster and better job execution with fewer tokens. supports claude code and codex now - more coming soon. Feedback I want

LLMs are at this point a necessity to develop large scale systems - we couldn't do what we do without them. Our team is worried by the developments of Mythos/Fable getting restricted and the news about GPT 5.6 access being gated by the government customer-by-customer. The part that complicates the obvious answers: the Anthropic directive keyed on the nationality of the person accessing the model, not where the company is incorporated; even their own foreign-national employees lost access. So "ju

Article URL: https://parcle.ai/second-brain Comments URL: https://news.ycombinator.com/item?id=48692740 Points: 1 # Comments: 0

1min.AI gives you lifetime access to GPT, Gemini, Claude, and more

Article URL: https://www.bloomberg.com/news/articles/2026-06-26/anthropic-moves-toward-deal-with-us-to-lift-curbs-on-ai-models Comments URL: https://news.ycombinator.com/item?id=48692520 Points: 2 # Comments: 1

Enterprises can govern model context protocol (MCP) connections at scale by treating them as part of the agentic AI control plane. Every MCP server, exposed tool, permission, and agent relationship needs ownership, scope, monitoring, and auditability before it supports autonomous work. MCP governance is the discipline of controlling how AI agents discover, select, invoke, and... The post How can enterprises govern MCP connections at scale? appeared first on DataRobot .

Article URL: https://github.com/team-telnyx/telnyx-code-examples/tree/main/ai-content-translator-python Comments URL: https://news.ycombinator.com/item?id=48692436 Points: 2 # Comments: 0

Article URL: https://www.leapd.ai/resources/state-of-ai-run-businesses-2026 Comments URL: https://news.ycombinator.com/item?id=48692322 Points: 1 # Comments: 0

If you’re like me and have basically delegated all your development to Claude code, Codex, OpenCode and the other agents, you’d agree that one of the annoying things is the unmanageable terminals. Personally I love working on the terminal, but when building multiple projects, it quickly becomes messy. I built Even to solve this mess. I can control my terminal tabs per project, + comes with some nice features that make it so I don’t need to open any other app when locked in. In built Browser, Age

Obviously, no body likes the idea of having or living next to a data center. And AI companies want to solve UBI in a sensible manner. Why not combine these two things? AI companies, go to individual homes in a community, ask to install a GPU cluster (for a size that fits + owner agrees to), and pay the household some 100s or 1000s of dollars a month? Yes, this will cause wild incentives (for those with apartments, multiple homes, etc.) which would need to be addressed, but surely, this would lon

Article URL: https://stockonomy.net/proof Comments URL: https://news.ycombinator.com/item?id=48692192 Points: 1 # Comments: 0

We launched last week on 4chan and have been building fast ever since, integrating new features and new avatars. Our idea is to make desktop pets not only cute, but useful. They can serve as a language partner, help with homework or quickly help with whatever's happening on your screen. Plus, of course, they can just chat with you when you're a bit lonely or want to bounce ideas around as well. You can check out the website at: deskmatelive.com Or at the MS store: https://apps.microsoft.com/deta

Article URL: https://www.ft.com/content/04f98e21-4ce7-43d2-8651-44557e12c31c Comments URL: https://news.ycombinator.com/item?id=48692128 Points: 3 # Comments: 0

Article URL: https://www.agent-kits.com Comments URL: https://news.ycombinator.com/item?id=48692101 Points: 2 # Comments: 0

There’s a multitude of concepts that have been created during this advent of AI. Which do you predict will be resilient and long-lasting, and which do you think will churn away as harnesses evolve? - Model Context Protocol - Skills - Agentic workflows - Agents - Claude.md / Agents.md - etc. Bonus: Despite all of the above, what do you see as the unaddressed or “dark corners” of the AI ecosystem? Comments URL: https://news.ycombinator.com/item?id=48691972 Points: 3 # Comments: 3

As Anthropic tightens restrictions on access to Claude in China, users keep finding new workarounds, from proxy services to fake identities sourced on Telegram.

Article URL: https://www.macrumors.com/2026/06/25/2027-macs-m7-chips/ Comments URL: https://news.ycombinator.com/item?id=48691862 Points: 1 # Comments: 1

After years of juggling imprecise grilling recipes and temperatures, I've made a grilling calculator that derives perfect meat results from thermal models, Fick's law of diffusion, protein structures, and more. Along the way I calculated exactly how far in advance to salt meat (spoiler: even when it's frozen!), realized the best dry brine is usually "cling brine" (covered), derived approximate charcoal quantities with physics, and much more. As with the pancake calculator[1], I did an insane amo

Article URL: https://www.jackhamr.ai/startup-grant Comments URL: https://news.ycombinator.com/item?id=48691808 Points: 2 # Comments: 1

<p>Laravel and Symfony projects are great places to use AI agents. Not because PHP is "easy" (it isn't), but because PHP backends carry a lot of real business logic. Controllers, services, console commands, queue jobs, Doctrine repositories, Eloquent models, event listeners, validators, policies, migrations, and tests all coexist in one repo, and the relationships between them are exactly the kind of context that's expensive for a human to load and cheap for a careful agent to map.</p> <p>The be

Article URL: https://www.youtube.com/watch?v=F25i0sgrp9M Comments URL: https://news.ycombinator.com/item?id=48691745 Points: 1 # Comments: 0

Lately, I am finding myself using coding agents primarily for investigating and learning new topics, which include: * understanding how specific module works in codebase * overall architecture in the new codebase * onboarding faster in the new codebase * researching a topic * analyzing video content/meeting transcripts to quickly catchup with the topic and so on. My current workflow is just to point out to repository or folder with transcripts and ask agent to extract insights (assuming it will

<p><a href="https://dev.tourl"></a>Building an AI Chatbot Using AWS Bedrock and JavaScript</p> <p>I recently worked on building a full-stack AI chatbot using AWS Bedrock and JavaScript. The goal of this project was not just to create a chatbot interface, but to understand how modern AI applications are built end-to-end using foundation models and cloud services.<a href="https://youtu.be/6Es8E6t3KaQ?si=mG-vn4C-gwJKsjUB" rel="noopener noreferrer"></a></p> <p>The application connects a React-based

Article URL: https://arxiv.org/abs/2604.16106 Comments URL: https://news.ycombinator.com/item?id=48691502 Points: 1 # Comments: 1

Hey all - Scott here, I was a heavy Linear user until I noticed I hadn't opened the UI in days. I was just asking Claude to pull up the tickets I cared about and draw whatever view I needed in the moment. At some point it clicked that I didn't need Linear, I needed the database underneath it. Then I realized it was the same for Confluence, Jira, and Pipedrive. But I didn't want to stand up Postgres and babysit schemas and migrations to get structured, collaborative data into Claude. I just wante

Article URL: https://brennan.day/an-open-letter-to-compassionate-left-leaning-ai-hating-animal-loving-meat-eaters/ Comments URL: https://news.ycombinator.com/item?id=48691450 Points: 5 # Comments: 0

Article URL: https://epoch.ai/MirrorCode Comments URL: https://news.ycombinator.com/item?id=48691419 Points: 3 # Comments: 1

Article URL: https://gizmodo.com/duckduckgo-unable-to-resist-the-pull-of-ai-mistakenly-claims-trump-died-of-rabies-2000777863 Comments URL: https://news.ycombinator.com/item?id=48691368 Points: 6 # Comments: 0

NYT shifts OpenAI/Microsoft copyright claims after SCOTUS ruling against Sony.

<p>If you use Claude Code, Cursor, or any AI coding agent, you know the problem: you ask the AI to review its own work, and it says "looks good." Every time.</p> <p>The AI isn't being lazy. It's sharing the same mental model that produced the code. It literally can't see what's wrong — the blind spots are baked in.</p> <h2> Why "Review this as a senior engineer" Fails </h2> <p>Generic role-playing produces generic findings. The AI fills in what it <em>thinks</em> a senior engineer would say, whi

<p>I just published <strong>SuperCompress</strong> to PyPI! 🎉</p> <p><code>pip install supercompress</code> — that's all it takes.</p> <h2> What is it? </h2> <p>A tiny ~5K parameter CPU policy that scores every line of context for relevance before sending to the LLM. It keeps only what matters for the answer.</p> <h2> The Numbers </h2> <ul> <li> <strong>65% fewer tokens</strong> → same answers</li> <li> <strong>100% oracle recall</strong> → never drops the answer line</li> <li> <strong>~60ms CP

<p>I spent tonight building a methodology for turning personal tools into open source contributions. Before publishing it, I decided to let the methodology review itself.</p> <p>The results changed how I think about AI-assisted code review.</p> <h2> The Method: Named-Persona Adversarial Review </h2> <p>The core idea is simple: instead of asking an AI to "review this code as a security engineer" (which produces generic, shallow feedback), you <strong>web-search actual engineers' documented philos

Surely, we can have rules that allow us to continue doing what we're doing

<p>The paradigm of Artificial Intelligence is undergoing a fundamental shift. We are moving rapidly from the era of <strong>stateless chat completions</strong>—where an LLM simply acts as an advanced text-autocomplete engine—to <strong>stateful, autonomous AI agents</strong>. These agents don't just talk; they <em>do</em>. They plan multi-step workflows, execute tools, read and write files, run test suites, and react to background triggers.</p> <p>At the center of this revolution is a powerful s

<p>I've been working on a side project called <strong>SuperCompress</strong> — an intelligent prompt compression system for LLMs. The idea is simple: most tokens you send to an LLM never need to be processed. They're padding, boilerplate, irrelevant context. But they still burn GPU cycles.</p> <p>I wanted to fix that.</p> <h2> The Problem </h2> <p>Working with LLM agents, I noticed something: every agent loop was sending massive context through the GPU. 10K tokens. 50K tokens. Sometimes more. Mo