Given the current state of affairs on Fable 5, it's great to remember Qwen3.6-27B (released April '26) running locally is slightly better than SOTA models released less than a year ago.

Don't trust leaderboard talk. Verify how they perform on actual hypothesis generation and experiment design.

you can access moonshot's new kimi k2.7 code right now, and it’s officially open-source and FREE😳 most people thought gpt-5.5-level performance was years away, but this model is already here and it's rewriting the rules of what a free assistant can build takes 1 minute to test what you get with kimi k2.7: →benchmark performance nearing gpt-5.5 levels across coding tasks →30% fewer reasoning tokens consumed compared to k2.6 →full open-source access through the kimi api and kimi code environments →high-speed 6x mode currently in development the performance breakdown (vs opus 4.8 & gpt-5.5): >mcp atlas: kimi 79.4 (beats opus 76.4) >mcp mark verified: kimi 81.1 (beats opus 76.4) >program bench: kimi 53.6 full setup guide (kimi k2.7 integration): step 1: access the platform →head over to the kimi api or kimi code dashboard to switch to the k2.7 environment step 2: optimize your tokens →leverage the 30% reduction in reasoning tokens to run more complex agentic loops for less cost step 3: integrate into your workflow >plug the new model into your current ide or client to start testing against your existing benchmarks step 4: prepare for the 6x speedup >keep your eyes on the update logs for the release of the high-speed mode to crush your latency issues already using opus 4.8? skip straight to step 3 and compare the output quality on your toughest codebase migrations this is the single biggest sign that frontier intelligence is becoming a commodity and the days of charging premium access for these models are numbered bookmark this before the free access tiers change or hit capacity

📅 Agenda for the #Copilot, #Microsoft365 & #PowerPlatform product updates call 16th of June • The latest updates ⚡ • Focus on #Copilot, #SharePoint & #SPFx • Presented by @vesajuvonen, Sahil Baid & Bert Jansen more! 🚀 👋 Join the call → msft.it/6019vcR2D

This Stanford lecture has 4,000 views. It should have 4 million. Fan-Yun Sun (CEO, Moonlake AI) made a claim that almost nobody heard: "World models will have arguably more economic impact than all language models today." More than ChatGPT. More than Claude. More than everything. Then he spent 48 minutes proving it, with the Wright Brothers, trillion-dollar simulation costs, a self-improvement loop that trains AI like a kid, and a 2-year prediction for when this hits the mainstream. If he's right, you're watching the next big AI wave before it has a name. Bookmark & watch ↓

Every FREE agentic engineering tool from the best AI builders that I've interviewed: @kieranklaassen Compound Engineering (⭐️ 21K): AI skills for planning, building, reviewing, and codifying lessons. My go-to for building with AI. github.com/EveryInc/compound… @kunchenguid gnhf (⭐️ 2K): Let agents work on features while you sleep. github.com/kunchenguid/gnhf No Mistakes (⭐️ 1.3K): Catch bugs with review, tests, lint, PR, and CI checks for AI-written code. github.com/kunchenguid/no-mi… Lavish (⭐️ 425): Review and give agents feedback on AI-generated HTML plans. github.com/kunchenguid/lavis… @mvanhorn Last 30 Days (⭐️ 41K): Research what people are saying about a topic across Reddit, X, YouTube, Hacker News, and the web. github.com/mvanhorn/last30da… Printing Press (⭐️ 3.3K): Make any website or app accessible to AI agents by sniffing out APIs and generating a CLI. github.com/mvanhorn/cli-prin… Agent Cookie (⭐️ 445): Keep your agent machine's sessions in sync with your main laptop. github.com/mvanhorn/agentcoo…

ANTHROPIC'S HEAD OF ENGINEERING SAYS THEY JUST SHIPPED THE MOST CAPABLE MODELS THEY HAVE EVER BUILT Mythos 5 and Fable 5. Fable 5 can run for days on a single goal and get it right on the first pass, on work that used to need an entire team. Most people use AI to summarize things. The real leverage is the long autonomous task you keep putting off.

I built the fastest coding agent on earth. 14,500 tokens per second. Seriously. It can create thousands of websites before @anthropic Fable can even make *1*. I'm using the @claude harness and powering it through a custom proxy with @taalas_inc silicon models.

📹 Writing 500 lines of agent code when 100 would do is a real problem. Session 6 covers how LangGraph fixes it. Here's what Zaid Ahmed taught in this session: 🔹 Why LangGraph's node-edge structure reduces the technical debt that piles up in fixed LangChain pipelines 🔹 How state, nodes, and edges give a single agent memory, decision points, and conditional routing 🔹 Why RAG retrieval quality depends on layering - metadata filtering, hybrid search, RRF, and cross-encoder reranking in the right sequence 🔹 How reflection and multi-agent patterns map directly to LangGraph implementations 🔹 Why document format affects LLM performance, and why Markdown is the most efficient input If you want to build agents that actually work in production, this is the foundation. The Agentic AI Bootcamp is now open for enrollment for the July cohort. Link in the comments. #AgenticAI #LangGraph #RAG #AIEngineering

Google quietly open-sourced a time-series AI that predicts anything. Sales trends. Market prices. User traffic. Energy demand. Crypto volatility. It's called TimesFM. Pre-trained on 100B real-world data points. Zero-shot forecasting with no fine-tuning.

Anthropic just showed a 24-minute workshop on how to actually prompt Claude. Taught by the people who built it. Free. No signup. No paywall. I've watched $300 courses that don't cover what they teach in the first 8 minutes.

Anthropic pays $750,000 a year for engineers who can build LLMs from scratch. Not how to prompt them. Not how to fine-tune them. Not how to build RAG pipelines. But how to build them from scratch. This 2-hour Stanford lecture teaches you everything. Scaling laws. Data collection. Architecture design. Post-training alignment. Free. From Stanford. Watch first. Then read this. The lecture is the theory. And this article shows you how to actually build it (with code) ↓

the skill that actually separates ai operators in 2026 isn't prompting. it's loop design. addy osmani just wrote a piece on what he calls loop engineering, designing the loops that coding agents run inside of. and it quietly explains why two people with the same model get wildly different results. here's the idea. a single prompt is one shot. you ask, it answers, you hope. a loop is a system. the agent acts, checks its own work against a goal, fixes what's wrong, and runs again until the output actually passes. the model is the same. the loop is the leverage. most people are still writing better prompts. the operators pulling ahead are writing better loops. test, observe, correct, repeat, with clear exit conditions so the agent knows when it's done and doesn't burn money spinning in circles. this is the part clients can't replicate. anyone can buy the same model you use. they can't buy your loops. the harness you've built around the agent, the checks, the guardrails, the way you've encoded "good" so the machine can grade itself, that's the actual product. that's the moat. stop thinking of yourself as someone who prompts ai. start thinking of yourself as someone who builds the loop the ai runs in. the prompt is the question. the loop is the business. read it: addyosmani.com/blog/loop-eng…

Google has published a paper that might end the transformer era. For the last 7 years, every major AI, ChatGPT, Claude, Gemini, has been built on the exact same architecture: The Transformer. But Transformers have a fatal flaw. To remember context, they have to process every single word against every other word. It’s called quadratic complexity. As your prompt gets longer, the compute cost explodes. The alternative is the old-school RNN (Recurrent Neural Network). RNNs are incredibly cheap and fast, but they have a fixed memory size. If you give them a long document, they get amnesia. Until today. Google researchers published Memory Caching: RNNs with Growing Memory. And it fixes the biggest bottleneck in AI. Instead of an RNN having a fixed, rigid memory that constantly overwrites itself, Google gave it a "save" button. The technique allows the RNN to cache checkpoints of its hidden states as it reads. The memory capacity of the RNN can now dynamically grow as the sequence gets longer. They built four different variants, including sparse selective mechanisms where the AI actively chooses exactly which checkpoints matter most. The results rewrite the rules of efficiency. On long-context understanding and recall-intensive tasks, these new Memory-Cached RNNs closed the gap with Transformers. They achieved competitive accuracy without the explosive, quadratic compute cost. It perfectly bridges the gap between the cheap efficiency of an RNN and the massive capability of a Transformer. We have spent billions scaling Transformers because we thought they were the only way an AI could remember a long conversation. But Google just proved we don't need to process the whole history every single time. We just needed a smarter cache.

The founder of Gamma: "we crossed $100M in revenue with 50 people" He started in 2020, peak pandemic, newborn at home, no paycheck, an investor literally hung up on him and said it was the worst idea he'd ever heard. He pitched 100 times in 2 weeks, 8pm to 2am every single night after putting the kids to bed. Today it's a $2.1 billion company, 100 million users, and word of mouth is still the number one acquisition channel.

Anthropic just dropped a serious policy package alongside CEO Dario Amodei’s new essay. They’re launching: - A $200M fund focused on AI labor research - A $150M national fellowship program - And two other major initiatives While most AI labs are racing purely for capability, Anthropic is putting real money into understanding and shaping how AI will impact the workforce and society. This feels like a mature move. They’re not just building powerful models, they’re actively trying to influence the broader consequences of the technology they’re creating. Whether you agree with their direction or not, credit where it’s due: they’re thinking several steps ahead. The era of “build fast and figure out the societal impact later” is slowly ending.

The AI era will be defined not only by technological breakthroughs, but by nations that can successfully integrate AI into governance, education, and economic development. In my latest Gulf News op-ed, I explore how the UAE is positioning itself to lead this transformation and shape a global model for AI-driven progress. Read the article: gulfnews.com/opinion/op-eds/… #UAE #ArtificialIntelligence #Innovation #FutureOfWork

UAE researchers develop injectable device that could treat chronic pain without surgery arnnewscentre.ae/news/uae/ua…

Playing with omnidreams (github.com/nv-tlabs/omni-dre…) inside flashdreams (nvidia.github.io/flashdreams) Fully interactive video generation at - 30FPS, - 720P resolution, - On a single RTX 6000 PRO GPU

If you want to get good at AI engineering (in 2026), learn these concepts: 1 LLM Evals Explained ↳ newsletter.systemdesign.one/… 2 Design Knowledge Q & A System ↳ newsletter.systemdesign.one/… 3 How OpenClaw Works ↳ newsletter.systemdesign.one/… 4 AI Agent Workflow ↳ newsletter.systemdesign.one/… 5 How MCP Works ↳ newsletter.systemdesign.one/… 6 Design AI Chat Assistant ↳ newsletter.systemdesign.one/… 7 How RAG Works ↳ newsletter.systemdesign.one/… 8 Agentic Patterns Explained ↳ newsletter.systemdesign.one/… 9 AI Coding Workflow 101 ↳ newsletter.systemdesign.one/… 10 Machine Learning System Design 101 ↳ newsletter.systemdesign.one/… 11 Multi-Agent Architecture Explained ↳ newsletter.systemdesign.one/… 12 How AI Agents Work ↳ newsletter.systemdesign.one/… 13 How Vector Databases Work ↳ newsletter.systemdesign.one/… 14 AI Agents: Memory, State & Consistency ↳ newsletter.systemdesign.one/… 15 AI Agents Design ↳ newsletter.systemdesign.one/… 16 Context Engineering 101 ↳ newsletter.systemdesign.one/… 17 What is Reinforcement Learning ↳ newsletter.systemdesign.one/… 18 LLM Concepts - A Deep Dive ↳ newsletter.systemdesign.one/… What else should make this list? === 👋 PS - Want my System Design Playbook (for free)? Join my newsletter with 201K software engineers now: → newsletter.systemdesign.one/… === 💾 Save & RT to help others get good at AI engineering. 👤 Follow @systemdesignone turn on notifications.