# This Week in All Things AI - Week 3-2026

> Sunday 11th January 2026 to Saturday 17th January 2026

**Published by:** [This Week in All Things AI](https://paragraph.com/@twiata/)
**Published on:** 2026-01-18
**URL:** https://paragraph.com/@twiata/this-week-in-all-things-ai-week-3-2026

## Content

Greetings friends particularly those in India which celebrated multiple festivals this week such as Lohri, Pongal, Makar Sakranti As evident by the torrent of content posted on the Telegram group and thus reposted in this newsletter, this week has kept curators such as yours truly busy with Anthropic's release of Claude Cowork, Salesforce launch of their Slackbot AI agent, updated image models from Zhipu and Black Forest Labs as well as other news. This week's edition of the newsletter is in my opinion one of the longest if not the longest for the sheer number of messages its reposting. Encourage you to take your time to read it Readers who think others in their family, friends and acquaintances who are curious in knowing more about rapidly evolving AI tools/services/use cases and would benefit from being subscribers of this weekly newsletter are encouraged to share this publication link to them and invite them to subscribe The following messages were posted on the 'All Things AI' Telegram group from Sunday 11th Jan 2026 to Saturday 17th Jan 2026Sunday 11th January 20261In this 81 min conversation with a16z's Jen Kha, originally an a16z internal meeting, Marc Andreessen frames the current AI revolution as the most significant technological shift of his lifetime, surpassing even the internet in scope. He describes a rapidly evolving market where revenue is growing at unprecedented rates despite high operational costs and shifting business models. The discussion highlights a geopolitical race between the United States and China, emphasizing that the threat of competition is effectively reshaping domestic regulatory policy. Andreessen also explores the "trillion-dollar questions" regarding the future of open-source versus proprietary models and the transition from massive "god models" to specialized small-scale systems. He ultimately argues that while public anxiety about job loss persists, the revealed preferences of consumers show a global rush toward adoption. Finally, he outlines the portfolio strategy of his firm, which involves placing simultaneous bets on contradictory technological paths to navigate this messy but magical era. 2Some links about Luxreal which claims to addresses spatial-consistency issues in AI-generated video. It is built for professional creators in e-commerce, film, and related fields. For members in this group who have workflows which they think would benefit from Luxreal, I would appreciate if you can post your feedback in the future when you have sufficiently tested Luxreal https://www.luxreal.ai/ Manycore Tech @ManycoreTech Introducing LuxReal — the latest innovation from Manycore Tech! Create AI-powered videos with fully controllable effects and seamless spatiotemporal consistency. Elevate your visual storytelling like never before! #LuxReal #AIvideo #Innovation 2 4:54 PM • Dec 11, 2025 https://pandaily.com/manycore-tech-inc-unveils-strategic-roadmap-opens-spatial-intelligence-capabilities-and-launches-two-new-products Fakhr @iamfakhrealam The real problem with AI video isn’t creativity. It’s spatio-temporal inconsistency. Most AI videos collapse when camera angles change. Objects shift. Shapes break. Perspective fails. Why? Traditional AI video models are trained on: • Flat images • Short clips • Pattern 59 12:22 PM • Jan 9, 2026 https://www.threads.com/@luokai/post/DTS8YZZCEby3Meshy is claimed by many to be currently one of the leading AI-powered 3D model generators in 2026. It's widely used by many users and creators, allowing anyone — from beginners to professionals — to create high-quality, production-ready 3D assets in seconds without needing traditional 3D modeling skills. The platform specializes in turning simple inputs (text descriptions or 2D images) into fully textured, rigged, and even animatable 3D models. It's especially popular among game developers, film/VFX artists, product designers, 3D printing enthusiasts, educators, and indie creators. They unveiled the groundbreaking Meshy AI Creative Lab at CES 2026 (just days ago!) — the industry's first tool that turns AI-generated 3D models into physical, full-color 3D-printable products with one click (handling mesh repair, materials, slicing, and even fulfillment through partners). Meshy also recently launched version 6 Meshy Unveils AI Creative Lab at CES 2026California-based Meshy debuted the Meshy AI Creative Lab during CES 2026 in Las Vegashttps://www.advancedmanufacturing.org MeshyAI @MeshyAI Meshy 6 is officially here! We're raising the bar for 3D character generation with smoother surfaces, sharper edges and more accurate anatomy. The mesh speaks for itself. 878 3:09 PM • Jan 9, 2026 https://x.com/i/grok/share/rKzw4iwOQeyzogteZL2yzMyylMonday 12th January 2026Tuesday 13th January 20261via Coop Demo of GLM 4.7 running at 1200 tokens per second here spinning up apps in 7 seocnds: Coop | Nullshot @codingcoop WATCH THE VIDEO. LISTEN TO THE BASS. FEEL THE SPEED. 3,000 TPS is the new standard and we are the only ones doing it at @Nullshot_ai 23 1:15 AM • Jan 13, 2026 2Apple signs a multiyear Google deal to use Gemini models to power a more personalized Siri in 2026, saying Google's tech “provides the most capable foundation Elon Musk says Apple and Google's Gemini deal “seems like an unreasonable concentration of power for Google given that [they] also have Android and Chrome” https://www.cnbc.com/2026/01/12/apple-google-ai-siri-gemini.html News from Google @NewsFromGoogle Joint Statement: Apple and Google have entered into a multi-year collaboration under which the next generation of Apple Foundation Models will be based on Google's Gemini models and cloud technology. These models will help power future Apple Intelligence features, including a 52.9K 10:38 PM • Jan 12, 2026 Jo Bhakdi @JOBhakdi If you add up the user base and distribution power between Google and Apple, it’s probably above 50% of all humans and probably above 80% of all humans who are reasonably partaking in civilization. Definitely a huge concentration of power. It will require x.ai Welcome | xAI xAI is an AI company with the mission of advancing scientific discovery and gaining a deeper understanding of our universe. Elon Musk @elonmusk This seems like an unreasonable concentration of power for Google, given that the also have Android and Chrome 52 12:31 AM • Jan 13, 2026 3Anthropic launches Cowork for Claude, built on Claude Code to automate complex tasks with minimal prompting, as a research preview for Claude Max subscribersIntroducing Cowork | ClaudeClaude Code's agentic capabilities, now for everyone. Give Claude access to your files and let it organize, create, and edit documents while you focus on what matters.https://claude.comFirst impressions of Claude Cowork, Anthropic's general agentNew from Anthropic today is Claude Cowork, a "research preview" that they describe as "Claude Code for the rest of your work". It's currently available only to Max subscribers ($100 ...https://simonwillison.net claire vo 🖤 @clairevo x.com/i/article/2010… 2,304 3:36 AM • Jan 13, 2026 4via Steve Ip hey all, i just made a little tool for myself but thought I'd share it with all of you. It takes a children's tv show episode and turns it into a kids book. Have fun with it. https://github.com/deepseekcoder2/vid2bedtimestory5Response from Steve IP to the following question from Kevin DentCan it be used on the final season of Game of Thrones?Steve's responseIt would need massive modification to it since its made for 10-20 minute kids shows, but yes, the general principals would work. Would need to create more steps for the context cache to be able to handle an hour, and the franchise.json required for recognition would need to also be modified to handle it. Also, technically, this could then feed into a "recap" video machine and can make it into an auto recap video maker on youtube for the monetization technically. It has all the pieces.6via: Ben Jammin I made a Github widget that shows your Contributions as a heatmap. I made it open-source and you can add it to your site using the instructions in the repo or using npm "npm install github-heatmap-widget" Enjoy! https://github.com/xBenJamminx/github-heatmap-widget7via Uruba Niazi I'm head of marketing at www.fabi.ai Today we’re introducing Fabi.ai 2.0 - Your AI analyst for ALL your data. When we talk to teams we noticed they were running into the same problems we had: Getting any sort of insight to help drive the business requires wrangling data across applications, copy-pasting SQL queries and pivot tables and wrestling with expensive, clunky legacy BI. AND you need a PhD in data engineering. In the age of AI there’s a better way. A MUCH better way. Fabi connects to all your data sources (Postgres, Snowflake, Stripe, HubSpot…), learns the context about your business and empowers everyone in your organization to perform data analysis at all levels to turn your insights into dashboards or automated workflows in minutes. Now, with Fabi, insights that can actually help you make better product, go-to-market and strategic decisions are one prompt away. We’re entering the golden age of data analysis and we’re excited to be making Fabi 2.0 generally available. 8Salesforce Launches AI-Powered Slackbot as Personal Work Agent The upgraded Slackbot uses Anthropic's Claude AI to search data, summarize discussions, create task lists, and handle routines like a chief of staff, pulling from Slack chats, files, calendars, and tools such as Google Drive and Jira. It's now available for Business+ and Enterprise+ customers with no extra setup, and Salesforce co-founder Parker Harris called it the fastest-adopted feature in 27 years—even CEO Marc Benioff uses it daily. Internal tests showed two-thirds of employees trying it, 80% sticking with it, and many saving up to 20 hours a week Early users praise its smooth integration, positioning it against rivals like Microsoft Copilot while prioritizing data security. Slackbot @SlackHQ Introducing the all-new, intelligently redesigned Slackbot: The out-of-the-box personal agent for every employee. sforce.co/3NFHY7A 236 6:43 PM • Jan 13, 2026 https://www.salesforce.com/news/stories/slackbot-personal-agent-work/Wednesday 14th January 20261Ramp's AI Agent Authors 30% of Merged Code Pull Requests Fintech company Ramp launched Inspect, an AI coding agent that handled nearly a third of merged frontend and backend pull requests in its core repositories last week. It runs in secure cloud environments with VS Code, terminals, and databases, writing code, running tests, taking UI screenshots, and iterating until ready for Git commits. Even non-engineers like product managers and designers used it, with co-founder Eric Glyman noting submissions from every job function. Ramp open-sourced the full blueprint on Monday, built on tools like OpenCode and Modal, sparking excitement among tech leaders about empowering more people to code. Zach Bruggeman @zachbruggeman The craft of engineering is rapidly changing. At @tryramp, we built our own background coding agent to accelerate faster. We call it Inspect. It wrote 30% of merged frontend + backend PRs in the past week. It’s powered by @opencode, @modal and @CloudflareDev. It runs fully in 2,125 8:29 PM • Jan 12, 2026 rahul @rahulgs we built our own background coding agent at ramp: it's called inspect its powered by @opencode and @modal, and works with all the frontier models it has a cloud hosted version of vscode, chromium, and terminal. has all the tooling and skills a ramp engineer would have, and is 1,271 8:52 PM • Jan 12, 2026 Eric Glyman @eglyman One useful way to think about agents: they’re control systems. Generating output is easy. Feedback is everything. At Ramp we built a background coding agent, Inspect, that can actually translate requests in English into code, and then observe reality: tests, telemetry, and Zach Bruggeman @zachbruggeman The craft of engineering is rapidly changing. At @tryramp, we built our own background coding agent to accelerate faster. We call it Inspect. It wrote 30% of merged frontend + backend PRs in the past week. It’s powered by @opencode, @modal and @CloudflareDev. It runs fully in 384 11:39 PM • Jan 12, 2026 2Zhipu AI and Huawei Open-Source GLM-Image Model Zhipu AI, partnering with Huawei, released GLM-Image, a 16-billion-parameter hybrid model that excels in text rendering and complex prompts on open-source benchmarks. It combines a 9-billion autoregressive generator for semantic reasoning with a 7-billion diffusion decoder for detailed refinement, all trained on Huawei's domestic Ascend servers using MindSpore (Huawei's version of PyTorch) —no foreign chips needed. The model tops charts like CVTG-2K (0.9116 word accuracy) and handles tasks from text-to-image to style transfers, with demos showing sharp recipe cards and seamless scene edits. Fans praise its leap for dense text, though some see room for improvement in simpler aesthetics. Z.ai @Zai_org Introducing GLM-Image: A new milestone in open-source image generation. GLM-Image uses a hybrid auto-regressive plus diffusion architecture, combining strong global semantic understanding with high fidelity visual detail. It matches mainstream diffusion models in overall quality 2,036 6:52 AM • Jan 14, 2026 jietang @jietang Just released z.ai/blog/glm-image "GLM-Image: a very high quality image generation model with auto-regressive for dense-knowledge" GLM-Image is an open-source, industrial-grade discrete auto-regressive image generation model. It utilizes a hybrid architecture that combines an 124 10:37 AM • Jan 14, 2026 ModelScope @ModelScope2022 GLM-Image is now open-source, a breakthrough in "Cognitive Generation." By combining Autoregressive models with Diffusion Decoders, it finally solves the struggle of generating posters, PPTs, and knowledge-heavy infographics. Highlights: #1 on CVTG-2K & LongText-Bench 161 7:50 AM • Jan 14, 2026 fal @fal GLM-Image is now live on fal! Hybrid 9B autoregressive + 7B diffusion architecture Text rendering and editing capabilities Fast generation with high-quality output Perfect for text-to-image, style transfer & identity preservation 79 8:28 AM • Jan 14, 2026 Thursday 15th January 20261OpenAI strikes a multibillion-dollar agreement to buy 750 MW of computing capacity from Cerebras over three years; sources: the deal is worth more than $10B https://www.reuters.com/technology/openai-buy-compute-capacity-startup-cerebras-around-10-billion-wsj-reports-2026-01-14/ https://openai.com/index/cerebras-partnership/2Tool Search in Claude Code being rolled out to reduce how much context MCP servers take up. Thariq @trq212 x.com/i/article/2011… 4,599 1:07 AM • Jan 15, 2026 Boris Cherny @bcherny Super excited about this launch -- every Claude Code user just got way more context, better instruction following, and the ability to plug in even more tools Thariq @trq212 x.com/i/article/2011… 5,555 3:27 AM • Jan 15, 2026 Matthew Lam @mattlam_ Another great small improvement, or large you use lots of MCP servers with Claude Code. TLDR: MCP servers would previously be preloaded, so if you used multiple, they'd eat up your context window like crazy. Now they'll be searched for only when you need to use that MCP server Thariq @trq212 We're rolling out Tool Search in Claude Code to reduce how much context MCP servers take up. 1 2:47 AM • Jan 15, 2026 Aakash Gupta @aakashgupta Anthropic just solved MCP’s biggest lie: that 50+ tools means 50+ capabilities. The math tells a different story. Users with 7 MCP servers were burning 67k tokens before typing a single character. That’s 33% of Sonnet’s 200k context window gone. One Docker MCP alone consumed Thariq @trq212 x.com/i/article/2011… 588 10:28 AM • Jan 15, 2026 3 I had the opportunity a short while ago to speak briefly with my friend Ian Rogers, CXO at Ledger [prior to that he was Chief Digital Officer at LVMH and before that played key role in the launch of Apple Music]. Ian mentioned to me that he had been building with Claude Code for a while but during the holidays tried to find its limit. He told me that he wrote a LinkedIn article about it. Since Ian isn't on Telegram , I am sharing his notes to members here https://www.linkedin.com/pulse/reflections-claude-code-new-year-ian-rogers-igl4c/ Ian has graciously been both a guest in my conversation series as well as a co-host in my conversation with Roope Rainisto who late last year took on the role as the Creative AI Lead of Supercell. Roope is also the creator of Beat Bandit, an AI-powered screenwriting platform.Friday 16th January 20261Replit Launches AI Tool for No-Code iOS Apps Replit announced a new feature powered by its AI Agent that builds and publishes native iOS apps from simple natural language prompts, partnering with Expo for testing and submission. Users can preview on a simulated phone, scan a QR code to test on their device via Expo Go, and handle pro features like databases, authentication, and integrations with OpenAI, Stripe, and more CEO Amjad Masad, who co-created React Native, highlighted how this overcomes mobile development barriers, with early users like Manny Bernabe building games for family and Vic Kim shipping map-based apps praising the native feel from prompts alone https://x.com/Replit/status/2011891602365825489 https://x.com/MannyBernabe/status/2011896607076467179 https://x.com/victoriakimse/status/2011894338465280197 https://x.com/jordwalke/status/20119091703252500992via Dima Kuchin: Some alternatives to Replit's recent update of vibe-coding iOS apps already existRork | Idea to mobile app in minutesRork builds complete, production-ready mobile apps from your description using AI and Expo (React Native)https://rork.com3Black Forest Labs Launches Fast FLUX.2 Klein AI Image Models The klein models, meaning 'small' in German, come in 4 billion and 9 billion parameter sizes optimized for speed on hardware like RTX 3090 or 4070 GPUs. They handle text-to-image creation, single edits, and blending up to five images for style swaps or additions, with the 4B distilled version producing a 1-megapixel image in 0.3 seconds on high-end GPUs. Fully open 4B versions allow commercial use, and benchmarks show top quality scores at far less latency and memory than rivals. Developers can test instantly via playground.bfl.ai, Hugging Face, or hosted platforms like Replicate. https://x.com/bfl_ml/status/2011825819082244266 https://x.com/ComfyUI/status/2011830637062471830 https://x.com/replicate/status/2011826686900441247Saturday 17th January 20261Whilst I personally at present have a subscription to Wispr Flow and have also used Aqua Voice before, I recently came across VoiceInk which is a MacOS speech-to-text dictation tool which is open source, uses local models and is priced reasonably well [USD 25 for single-device] as a one-time purchase and seems to have good reviews. An iOS version is in beta [currently free to use] VoiceInk - Transform Speech to Text Instantly on MacInstantly Convert Speech Into Text with VoiceInk. Advanced AI-Powered Voice Recognition for Mac with Offline Capability, 100+ Languages, and Complete Privacy Focus. Try the Best Dictation Software for Mac.https://tryvoiceink.com2From Addy Osmani who was previously Engineering Manager for Google Chrome and currently bridges Google DeepMind, engineering, product, and developer relations teams comes a great document with actionable advice on "How to write a good spec for AI agents" How to write a good spec for AI agentsLearn how to write effective specifications for AI coding agents to improve clarity, focus, and productivity in your AI-driven development workflows.https://addyosmani.comBelow is my personal website which aggregates links to many of my socials as well as the various content and community that I curate. Feel free to share this link to others who you think may find this content/community useful to them https://linktr.ee/goolamabbas The cover image of this newsletter via generated via the Qwen model within the Freepik tool via the following promptIndian castle surrounded by water and nature, village, volumetric lighting, photorealistic, insanely detailed and intricate, Fantasy, epic cinematic shot, trending on ArtStation, mountains, 8k ultra hd, magical, mystical, matte painting, bright sunny day, flowers, massive cliffs, Sweeper3DBest Regards, Yusuf

## Publication Information

- [This Week in All Things AI](https://paragraph.com/@twiata/): Publication homepage
- [All Posts](https://paragraph.com/@twiata/): More posts from this publication
- [RSS Feed](https://api.paragraph.com/blogs/rss/@twiata): Subscribe to updates
- [Twitter](https://twitter.com/yusufg): Follow on Twitter