suggy

Anthropic engineer built a memory that dies unless the agent can retell it. If it can't say the fact in its own words, the fact is gone. He didn't store what the agent read. A copied fact sits there looking known - the agent has it, but doesn't actually understand it. INTAKE never saves the raw input. It hands the event back and says: explain this to me without looking. RETELL is the test. The agent has to reconstruct the fact from scratch, in its own phrasing, no peeking at the source. MATCH checks the retell against the original. If the meaning survives the rephrasing, it's understood. If it garbles, the agent never really had it. KEEP stores only the retold version - never the source. What lives in memory is what the agent could rebuild, not what it was handed. Here's the part nobody copies. Everyone saves the input word for word. He saves only what survived being re-explained, because a fact you can't restate in your own words you don't know - you're just holding a copy of it. Last month the agent failed to retell 35% of what it read, and dropped all of it. What stayed was smaller and actually usable, not a library it couldn't open. Your last forgotten lesson wasn't badly stored. It was never understood - highlighted, saved, filed, and gone, because you never made yourself say it back. You know this from studying. Re-reading the page feels like learning. Closing it and explaining it is the only thing that proves you learned anything. The agent that stores what it read isn't knowledgeable. It's a photocopier - full of text it can't reconstruct, mistaking having for knowing. Everyone is saving the input. He saves only what the agent could rebuild blind, and threw away everything it couldn't. Don't let this rot in your bookmarks. The article below is the whole memory layer - the intake that refuses raw input, the retell test, the meaning-match, and the rule that keeps only what the agent could re-explain from scratch.

One of the most useful ways to use Local AI is for long running tasks especially when managing the work of much smarter Cloud agents. ... but how to keep working for hours without losing focus (or coherence)? 🎯 This is where Occamy-1.0 comes in! Accio-Lab took Qwen3.6-35B-A3B and heavily post-trained it specifically for long-running agent/co-work tasks. 🧠 35B total parameters ⚡ only ~3B active/token 📚 262K context The training targets things agents actually need to do: 🔎 Search 💻 Terminal + coding 📁 Files 🔧 Tool calls 🌐 APIs 🤝 Delegated runs 🧠 Maintain state across long workflows 🔄 Recover from execution failures They even trained 2 specializations ... 🏃 Sprint → broad agent capability 🏃‍♂️ Marathon → sustained long-horizon execution ...then merged them into 1 checkpoint and refined it further. No second model required at inference. This is supported 💾 Q4_K_M = ~21.2GB ✅ llama.cpp ✅ Ollama ✅ Hermes Agent ✅ OpenClaw ✅ Apache 2.0 Creator-reported co-work results 👇 Claw-Eval 🦑 Occamy — 82.2 🐦 Ornith-1.5 — 64.4 🟣 Qwen3.6-35B-A3B — 69.5 AutomationBench 🦑 Occamy — 69.1 🐦 Ornith — 58.0 🟣 Qwen — 39.4 Ornith still wins Terminal-Bench 2.1: 🐦 Ornith — 67.8 🦑 Occamy — 59.0 So Occamy isn't claiming to win all benchmarks but rather be a ~3B-active Local AI model deliberately optimized to keep working, using tools and maintaining context across an entire job. Can it actually keep working? I have to test this!👀 🔗 HF: Accio-Lab/occamy-1.0

TypeSafeのJev使えるようになったので、日本語で試してみたが、、これすげえな…！👀✨ これ、よくある「チャットAI」じゃなくて、「判断AI」で、文章と判断基準を渡すと、 「どの担当？」「本番障害？」「重大度は？」 が、確率やスコア付きで返ってくる。 今回検証したら、サーバー側の評価時間は約76〜282msで早すぎw（通信時間は別）。 公式料金は、入力100万トークンあたり0.042ドルで、出力は無料。 仮に判断基準込みで1回1,000入力トークンなら、10万回呼んでも計算上4.2ドル。。 この速さと料金なら、小さな判断をアプリのあちこちに入れたくなるなこれは👀✨ しかも、不具合と請求が混ざった問い合わせには、 「技術担当78％、請求担当22％」 みたいに判断の割れ方まで返してくれる。 これ、エンジニア目線だと「その都度意味を理解するif文」をアプリに組み込める感覚。 問い合わせの振り分け、AIエージェントの処理先選び、回答の品質チェック…使いたい場所がめっちゃ浮かぶなこれは。。 もちろん型が正しくても判断が正しいとは限らないので、閾値と例外処理はこっちで設計する必要あるが、 その役割分担も含めて、かなり好きな方向性☺️ ウェイトリストこちらから登録できるのでぜひ試してみるのおすすめです💡 ・Home - TypeSafe AI 　　URL: typesafe.ai/

@Cloudflare 🤝 @typesafeai

DEEPSEEK HAS JUST KILLED THE ENTIRE CODING AGENT INDUSTRY It's called deepseek-harness It's the most complete framework for creating code agents Open source. Claude's most complete plan costs $200 a month. This is FREE And it comes with a brutal idea Everything is a plugin The model The tools The sandbox The UI Even the agent's own loop You can swap out any of those pieces just by tweaking the config Without touching the base code One single command and you have the local web interface: npx @deepseek-ai/dsh web Compatible with DeepSeek, Claude, GPT, Gemini and whatever else you want to plug in It reached over 160 thousand stars in just a few days A frontier lab just dropped for free the exact layer that other companies are selling as premium product I'll leave the repo in the comments

YC is partnering with over a dozen companies to offer students attending YC events free credits for AI developer tools! The YC AI Student Starter Pack includes over $20k in Azure & AWS credits, $5k+ in ChatGPT, Claude, and Grok credits, and more! ycombinator.com/blog/the-yc-…

Nobody wants to hear this, but distribution is the new moat. Almost every founder pitching me their new AI startup has no audience to sell it to.

Some news: 5 months ago I joined @coreauto, a small research lab focused on new ways for models to learn. I’m also hiring four interns to do something my younger self would've loved. It started with a pretty unusual project. We set up a handful of small online businesses, and put an agent in charge. We gave each business some initial funding and a seed idea, and off they went, trying to sell real things to real customers while we watched. It’s about as open-ended and unforgiving a task as you can probably give to an agent, with countless decisions to make and effects that take time to become clear. Each business got its own bank account, a debit card to pay suppliers, a web browser, an email inbox, a domain name, cloud infrastructure and a schedule. Every few hours, the agent woke up and decided what to do. It might update a landing page, adjust an ad campaign, reply to a customer, and then go back to sleep. The agents built actual storefronts and apps. They bought ads, attracted visitors, and even made some sales. The agents acted as operators that kept a business moving, but we noticed they never really stepped back to strategize about what might move the needle. If people weren’t buying, why? Was the landing page unappealing? Or maybe the offer itself was the problem? One amusing example was the very first order a customer placed. Our supplier had run out of stock. But it’s a dropshipping business, and thankfully tens of other suppliers carried the same product. Instead of trying another supplier, the agent almost immediately sent the customer a polite apology and attempted to issue a refund. Gah! So close! The agents could handle much of the work of getting a business up and running. What they struggled with was deciding what to change when it wasn’t working. That’s what we want to understand. We also want to see what changes when people take the lead. What does a business operator actually do all day when agents can do so much? What new work does this allow them to do? Which decisions can’t be delegated? — So this winter we're hiring 4 interns as Associate General Managers. Associate Product Manager internships are programs that give early career generalists a product to own. We want to try something similar, but with entire businesses. You’ll create and grow businesses, with agents handling the routine work so you can focus on figuring out what matters. Each AGM gets a budget and our agent infrastructure that’s designed for this. Which ideas to try is up to you. Put more money behind what finds customers and keep refining. Where you step in tells us what agents can't do yet. You'll be in our SF office with me and people I’m so fortunate to work with, like @_arohan_, @joannejang, @MillionInt, @marksaroufim and @juliacvillagra We don't care much where you went to school or what your resume looks like. We’re looking for enterprising people with a drive to create, who’ve made money on the internet before and already use AI extensively to enable their ideas. If that sounds like you: coreauto.com/AGM

Anthropic engineer (ex-Google): "I spent 14 years at Google and built 24 reusable skills for agents that I can use with any AI model So I stopped prompting my agents from scratch. The repo has already reached ~100K stars on GitHub" In a 40-min masterclass, Addy Osmani showed exactly how he uses his skills (~100K stars on GitHub) for his 24/7 multi-agent system His workflow and this workshop will replace 15 hours of other paid videos on agent engineering Watch it today, copy the GitHub repo - then read below how to use these skills for graphs ↓

We’re releasing new Codex features to make it a more effective coding collaborator: - A new IDE extension - Easily move tasks between the cloud and your local environment - Code reviews in GitHub - Revamped Codex CLI Powered by GPT-5 and available through your ChatGPT plan.

200,000+ new vibe coding projects get created every day yet almost NONE of them get customers 7 distribution strategies that actually work right now for your startup: 1. build an MCP server. when someone asks claude or chatgpt the question your product answers, your tool shows up. the AI becomes your sales team. 2. programmatic SEO. pick a keyword pattern (best X for Y). use firecrawl to pull real structured data so pages have actual value. one next.js template, AI generated content, human editing loop so it doesn't read like AI. 10,000 pages × 30 visits × 2% CVR × $10 = $60k/month from pages you built once. 3. vibe code a free tool (calculator, software etc). one problem, one tool, ship it today. it ranks, lives in people's workflows, markets your brand for years. ahrefs' free backlink checker has sent them more customers than most paid ads ever will. 4. answer engine optimization. people are getting answers from chatgpt and perplexity now, not just google. find the top questions your customer is asking AI. publish structured, definitive answers. one founder went from 4% to 20% AI referrals in a month just by doing this. 5. make the output of your product shareable. think spotify wrapped. think github graphs. think stripe atlas. what does your user want to screenshot and send? build that moment. add a pre-filled share button. every share is free impressions to your exact audience. 6. buy a niche newsletter. 10k subscribers for $5k to $20k. most owners are making $0 to $500 a month. DM them "ever thought about selling?" you inherit trust and a direct channel to your exact customer on day one. underrated. 7. 30 minute voice memo into claude: five tweet threads, three linkedin posts, one newsletter, short form clips. do this weekly. in 3 months you have more content than competitors who aren't doing this. obviously, your project needs to be optimized so it isnt ai slop, but you'll get there. code is commoditized. time to focus on distribution. pick 2 of these ideas and start this week to get customers. this episode was designed to get your creative juices flowing. maybe it'll give you more ideas on growth tactics you'll use this week. full breakdown on the @startupideaspod watch.

Some people have been contemplating an idea for years, maybe decades. Obsessing, attempting, discarding, agonizing, retrying. Some of these ideas are unpopular, niche, impractical. Not obviously capitalizable. They live on in the inventor's mind. In 2026, millions of these ideas will come to life thanks to superintelligent coding agents. AI doesn't get tired. It amplifies the individual, and for better and sometimes for worse, it always takes you seriously. "Great idea. Splendid. Wow. You're absolutely right." A world of digital wonders awaits us. This world will disproportionally favor the boldest ideas. Software that once seemed impossible will be one hyperlink away. I can't wait to see it.

Friday Developers Fun 🤣 Just don't touch it #developers #meme #fridayhumor

For as long as we've covered Robotics, the question has been "when?"… When do we move from novelty to utility, demos to deployment? In our latest Field Trip, we sent our analyst behind the scenes of the leading robotics lab to see with his own eyes what's happening on the bleeding edge of the curve. The verdict? Robotics has reached a tipping point. Useful productivity, autonomy and embodied intelligence are rapidly accelerating but most frameworks haven’t properly focused on the drivers. Instead, we lay our own framework for the coming wave of Physical AI.  citriniresearch.com/p/roboti…

【速報】 Claude Code、ついに「AIに仕事を任せっぱなしにできる新機能」を出してきた😳 その名も ーーーProjects（プロジェクト）！！ これ、一言で何ができるかというと👇 teapawt.net/claudeai/status/210063… 1つの会話に頼みごとを投げるだけで、Claudeが自分で作業を分けて同時に進めてくれます。 これまで複数の作業を同時に進めようとすると、 画面（セッション）をいくつも開く 毎回同じ前提を説明し直す どれが終わったか自分で見に行く みたいな運用が必要でした。 でもProjectsでは、 ・頼みごとは1つの会話に投げるだけ ・Claudeが作業ごとに「スレッド」を自動で立てる ・スレッドはクラウド上で同時に動く ・PCを閉じても作業は止まらない ・ルールは1回書けば全スレッドに共有 ・決めたことはプロジェクトの記憶に残る ・「あなた待ち」の作業だけ一覧で出る という進め方に変わります。 つまり、Claude Codeが 「1つずつ頼むAI」から、 作業を振り分ける“現場監督AI”に変わりつつある。 特にヤバいのはここ。 スマホのClaudeアプリからも進み具合を見て指示できる。 だから、 ・バグ修正 ・テストの追加 ・仕様書どおりの実装 ・資料の下書き ・調べもの ・問い合わせの分析 を朝まとめて投げて、 人間は「呼ばれた時だけ答える」働き方で回せるようになる。 これはClaude Codeの思想がかなり明確になったアップデート。 もうAIに1問1答でお願いする時代じゃない。 AIにまとめて任せて、人間は判断だけする時代に入ってきてる。 Pro・Maxの一部ユーザーから順次公開（ベータ版）。 左メニューに「Projects」が出ていれば使えます。 任せっぱなしにするなら、Anthropic社内で実際どう回しているかを知ってる人ほど強い。 Anthropic社内の運用ベストプラクティスをまとめた記事はこちら👇

强烈建议大家都亲自试试Jev，能让你的Codex操作速度提高10倍并省下大量token！ 因为Jev非常适合用于判断型工作，能在毫秒内做出正确判断，而且token价格极低。因此非常适合以下几个场景： 1、computer use 将Jev用作操作判断层，指导模型快速操作你的电脑 teapawt.net/gregpr07/status/210041… 2、上下文压缩 由Jev判断上下文内容的重要信息，实现近乎即时的上下文压缩 teapawt.net/tamarajtran/status/210… 3、模型路由 根据任务复杂程度自动分配合适的模型来处理 teapawt.net/mdlahfir/status/210031… 4、自动审核 比5.6 luna还要便宜且精准的自动审核模型，可能是Codex在安全审批方面最合适的工具 teapawt.net/fazxes/status/21003000… 使用的话要去官网 typesafe.ai 申请，趁现在还通过的很快大家可以来尝鲜体验一下

SpaceXAI engineer, Lauren Tan: "99% of people using GrokBot just for 1% of its real power. They run 1 agent without "loop" & "graph" I'm running a team of 20+ GrokBot agents, fully autonomous. I have a Chief of Staff agent, a PM agent and 20+ workers - that's the new stack of engineer" In a 1-hour session, a SpaceXAI engineer showed how to build a team of effective AI agents from scratch this is worth more than a $500 agentic engineering course watch this workshop today, then read how to build a fleet of GrokBot agents in the article below

We just open-sourced our entire marketing playbook. After huge viral success with our Marketing OS running on Grok Bot, we decided to bring it to Claude. Introducing Marketing AGI by Maxfusion. You no longer need your first 5 marketing hires. 📁 Marketing Department ┃ ┣ 📁 Head of Marketing ┃ ┣ 📁 Positioning & Offer Design ┃ ┣ 📁 Pricing Strategy ┃ ┗ 📁 Competitor Teardowns, Public Data Only ┃ ┣ 📁 Copywriter ┃ ┣ 📁 20 Variants, Panel-Scored ┃ ┣ 📁 Email Sequences, Written in Full ┃ ┗ 📁 Posts Built to Survive the Feed ┃ ┣ 📁 Creative Strategist ┃ ┣ 📁 The 18-Tactic Hook Engine ┃ ┣ 📁 Fatigue Caught at the Concept Level ┃ ┗ 📁 Next Tests, Ranked & Briefed ┃ ┣ 📁 Launch Lead ┃ ┣ 📁 The Runway Before Launch Day ┃ ┗ 📁 Product Hunt, Hour by Hour ┃ ┣ 📁 GEO Lead ┃ ┣ 📁 Ranked by ChatGPT & Perplexity ┃ ┗ 📁 App Store, Metadata to Screenshots ┃ ┗ 📁 Analyst 　 ┣ 📁 Audits, Weighted 0-100 　 ┣ 📁 Numbers That Mean Something 　 ┗ 📁 AI Slop, Caught on Sight And they coordinate all the work: A single Claude skill runs the whole loop: The Analyst finds the problems, the Copywriter rewrites it, the Creative Strategist tests the new version, and the Analyst grades the outcome. Comment "AGI" and I'll send you the setup.