Image created with gemini-3.1-flash-image-preview with claude-sonnet-4-5. Image prompt: Animation cel style illustration of a friendly blue-skinned genie emerging from a brass oil lamp, carefully reading an unfurled parchment scroll labeled Constitutional AI with visible text about ethics and safety, magical teal wisps surrounding the scroll, warm golden lighting, clean background with Arabian Nights ambiance, Disney Pixar quality 2D animation aesthetic with bold outlines and volumetric magical effects, horizontal composition with space for title text.
more than claude code and moltbot, the main emerging trend in january was agent sandboxes. saw several posts from cloudfare, vercel, ramp, modal about agent sandbox and new features they were adding to them. i particularly enjoyed ramp’s article”” https://x.com/dejavucoder/status/2016979866651152898
BREAKING: @MiniMax_AI introduces MiniMax Agent Desktop! MiniMax Agent = Claude Cowork + Agent skills + Clawdbot. It’s really good! Watch how I use it to quickly create a visually stunning presentation from an AI paper. I was mindblown when I tried this for the first time.”” https://x.com/omarsar0/status/2016149402923200634
Anthropic integrates interactive MCP apps into Claude https://www.testingcatalog.com/anthropic-integrates-interactive-mcp-apps-into-claude/
A bit more context e.g. from Simon https://t.co/Yeq0lLOPBF just wow”” https://x.com/karpathy/status/2017297261160812716
Claude in Excel is really good. Its weird that using Microsoft’s own Excel agent using Claude 4.5 often yields weaker answers, It seems to be because the Excel agent relies on Excel alone (VLOOKUPs, etc) while Claude in Excel does its own analysis and uses Excel for output.”” https://x.com/emollick/status/2014891787051999566
Claude in Excel | Claude https://claude.com/claude-in-excel
16M impressions in 24 hours. if you’ve ever tried Claude in Sheets or Claude in Excel you will know how much more intelligent it is compared to Gemini in Sheets i have two current measures of Google-GDM product integration right now: – how long does it take Google to put a non”” https://x.com/swyx/status/2015207720237089146
We’ve launched the first official extension to MCP. MCP Apps lets tools return interactive interfaces instead of just plain text. Live in Claude today across a range of tools.”” https://x.com/alexalbert__/status/2015854375051428111
Your work tools are now interactive in Claude. Draft Slack messages, visualize ideas as Figma diagrams, or build and see Asana timelines.”” https://x.com/claudeai/status/2015851783655194640
MCP Apps – Bringing UI Capabilities To MCP Clients | Model Context Protocol Blog https://blog.modelcontextprotocol.io/posts/2026-01-26-mcp-apps/
Interactive tools in Claude | Claude https://claude.com/blog/interactive-tools-in-claude
On December 8, the Perseverance rover safely trundled across the surface of Mars. This was the first AI-planned drive on another planet. And it was planned by Claude.”” https://x.com/AnthropicAI/status/2017313346375004487
[AINews] Moonshot Kimi K2.5 – Beats Sonnet 4.5 at half the cost, SOTA Open Model, first Native Image+Video, 100 parallel Agent Swarm manager https://www.latent.space/p/ainews-moonshot-kimi-k25-beats-sonnet
We got Claude to teach open models how to write CUDA kernels. This blog post walks you through transferring hard capabilities (like kernel writing) between models with agents skills. Here’s the process: – get a powerful model (like Claude Opus 4.5 or OpenAI GPT-5.2) to solve a”” https://x.com/ben_burtenshaw/status/2016534389685940372
Given the attention to Claude Code/Codex, I think that people’s views about what AI can or can’t do are getting overly shaped by the affordances of CLI tools. A different agentic harness will radically change the ability profile of frontier models We just haven’t seen them yet.”” https://x.com/emollick/status/2014209839677812911
A few random notes from claude coding quite a bit last few weeks. Coding workflow. Given the latest lift in LLM coding capability, like many others I rapidly went from about 80% manual+autocomplete coding and 20% agents in November to 80% agent coding and 20% edits+touchups in”” https://x.com/karpathy/status/2015883857489522876
Continuing to build a weird game demo every day, this one is kind of fun: “”What if tic-tac-toe was given the Balatro/Slay the Spire roguelike treatment?”” As usual, I just gave feedback, all the major design decisions (& all code) was Claude Code. Play: https://x.com/emollick/status/2014793182576345471
MCP servers in @code can now return a UI in chat thanks to the MCP Apps spec. So I added a UI to the LIFX MCP server – get a control panel for the light if the AI can’t figure out what you want from your lazy prompt.”” https://x.com/burkeholland/status/2016208751200457088
Reading Claude Code playtesting a game it made is very cute, it is so proud of itself. (you can put whatever you think are the appropriate words to have in scare quotes in those quotes, depending on how you feel about anthropomorphising AIs)”” https://x.com/emollick/status/2015620461506293884
Someone in the comments asked for this to be made into a LucasArts style game instead. So I asked Claude to remake it that way (it added the jokey writing) Play: https://t.co/lM1FCw3cU3 (I was impressed that it figured out how to create sprites from images for the inventory)”” https://x.com/emollick/status/2015844994947465493
Introducing Claude Chic – Matthew Rocklin https://matthewrocklin.com/introducing-claude-chic/
Claude Coude、毎回キーボード触るの面倒すぎて、物理承認印鑑をつくりました。 これを使えばジャパニーズトラディショナルな気持ちでバイブコーディングできます。普通にちょっと便利”” https://x.com/takex5g/status/2017091276081156265
Goodnight, Claude. Hope those projects are done by the morning.”” https://x.com/emollick/status/2015686312028766514
This game was 100% designed, tested, and made by Claude Code with the instructions to “”make a complete Sierra-style adventure game with EGA-like graphics and text parser, with 10-15 minutes of gameplay.”” I then told it to playtest the game & deploy. Play: https://x.com/emollick/status/2015512532056764490
The degree to which Claude Code/Codex/etc are already capable of doing full development loops is super interesting. Here I tasked Claude Code with making a game more fun and balanced. It altered the code, but spontaneously also opened my browser and play-tested the game changes.”” https://x.com/emollick/status/2014758376354328655
The Claude vs Codex debate is missing the point. I use both on the same folder, same files. Claude for exploratory thinking, Codex for complex technical problems, and whichever has quota left when the other runs out. Stop pledging loyalty to AI companies. Use what’s”” https://x.com/fdaudens/status/2015188670408483058
BlenderMCP connects Blender to Claude AI through the Model Context Protocol (MCP). > Enables prompt assisted 3D modeling: Blender just became programmable by language. It connects Blender directly to Claude via the Model Context Protocol. Forget UI tricks and exports… Claude”” https://x.com/IlirAliu_/status/2014775922377752580
[AINews] Anthropic launches the MCP Apps open spec, in Claude.ai https://www.latent.space/p/ainews-anthropic-launches-the-mcp
Claude can make blue1brown animations in minutes. Education is about to explode.”” https://x.com/LiorOnAI/status/2016119374097084828
Hey Claude Code: “”change a diaper, plan an invasion, butcher a hog, conn a ship, design a building, write a sonnet, balance accounts, build a wall, set a bone, comfort the dying, take orders, give orders, cooperate, act alone, solve equations, analyze a new problem, pitch manure,”” https://x.com/emollick/status/2016532288675500539
Claude Code, heal thyself (It did, and then submitted a bug report)”” https://x.com/emollick/status/2014708077371556117
The author of Clean Code is using Claude to write software. I can hardly think of a clearer indicator that coding is now officially outsourced to LLM’s. Who else is mourning the death of coding by hand ?”” https://x.com/mischavdburg/status/2016389228356149460
(23) Can You Teach Claude to be ‘Good’? | Meet Anthropic Philosopher Amanda Askell – YouTube https://www.youtube.com/watch?v=HDfr8PvfoOw
Dario’s new essay on the risks of AI seems less in dialogue with his earlier essay on how AI can help us all, but rather with the recently released Claude Constitution In that context, the Constitution feels much more like a plea to future Claude from Anthropic than instructions”” https://x.com/emollick/status/2016036346192646214
Important new course: Agent Skills with Anthropic, built with @AnthropicAI and taught by @eschoppik! Skills are constructed as folders of instructions that equip agents with on-demand knowledge and workflows. This short course teaches you how to create them following best”” https://x.com/AndrewYNg/status/2016564878098780245
Huge thanks to @github for the amazing shout-out on MCP-UI and the new MCP Apps spec! We’re proud to join forces with @OpenAI and @AnthropicAI to create a unified spec for apps that run across chat platforms. Build once, run everywhere. 🚀 (cc @idosal1 )”” https://x.com/liadyosef/status/2002104900843679818
Today, the MCP community is announcing MCP Apps, the first official MCP extension. @code is the first major AI code editor with full MCP Apps support. With MCP Apps, tool calls can now return interactive UI components that render directly in the conversation. Learn more:”” https://x.com/code/status/2015853688594612715
This is basically vibe coding for 3d — clever scaffolding for any VLM to iteratively reconstruct 3d scenes. Think blender MCP on steroids. As major AI labs continue to hill climb spatial reasoning benchmarks the results we can get will only get more impressive. I love these”” https://x.com/bilawalsidhu/status/2015214794614227420
Anthropic just released the receipts on a fear everyone’s been hand-waving. 52 junior engineers learning a new Python library. AI group scored 50% on comprehension tests. Manual coding group scored 67%. That’s a 17% gap on foundational skills, and debugging showed the steepest”” https://x.com/aakashgupta/status/2017087521411477926
Clever scaffolding for any VLM to iteratively reconstruct (and even animate!) 3d scenes. No training required. Basically like blender MCP on steroids.”” https://x.com/bilawalsidhu/status/2015945325928649065
This research was led by Jackson Kaunismaa through the MATS program and supervised by researchers at Anthropic, with additional support from Surge AI and Scale AI. Read the full paper:”” https://x.com/AnthropicAI/status/2015870975238406600
AI can make work faster, but a fear is that relying on it may make it harder to learn new skills on the job. We ran an experiment with software engineers to learn more. Coding with AI led to a decrease in mastery–but this depended on how people used it.”” https://x.com/AnthropicAI/status/2016960382968136138
Anthropic faces new music publisher lawsuit over alleged piracy | Reuters https://www.reuters.com/legal/litigation/anthropic-faces-new-music-publisher-lawsuit-over-alleged-piracy-2026-01-28/
New research: When open-source models are fine-tuned on seemingly benign chemical synthesis information generated by frontier models, they become much better at chemical weapons tasks. We call this an elicitation attack.”” https://x.com/AnthropicAI/status/2015870963792142563
Anthropic co-founder Jared Kaplan had this to say about the future of physics research. A bold claim that I might not take as seriously if Kaplan wasn’t such a brilliant physicist before he left the field for AI:”” https://x.com/nattyover/status/2016239582220624177?s=20
I wish @DarioAmodei didn’t conflate “”having quasi-religious views about AI risk”” and “”thinking misaligned AI takeover risk is high””. If risk is actually high, we should want to believe risk is high! It’s possible to reasonably think takeover risk is very high! (IMO it’s ~40%)”” https://x.com/RyanPGreenblatt/status/2015869503385772037
The Adolescence of Technology: an essay on the risks posed by powerful AI to national security, economies and democracy–and how we can defend against them:”” https://x.com/DarioAmodei/status/2015833046327402527
Dario Amodei — The Adolescence of Technology
https://www.darioamodei.com/essay/the-adolescence-of-technology
I like that the current frontier models are polar opposites, it makes their use-cases and strengths pretty obvious GPT-5.2 = Exploration -> the reason why xhigh and Pro are so damn good Opus 4.5 = Exploitation -> the reason why Anthropic don’t need many tokens and reasoning”” https://x.com/scaling01/status/2016335491243676058
As an author, I feel I am lucky to have had a chance to establish my voice through trial & error before the invention of LLMs. Even if you don’t use them for writing, unless you are very careful, you start to pick up ambient Claudish or GPTish phrasing from all the other AI work”” https://x.com/emollick/status/2016235542791278756
Leave a Reply