Last Week’s AI News #27

Most AI content sounds impressive, but rarely delivers real impact.

We’re changing that.

This week, we’re launching “Friday AI Clinic”, a live format where real business problems get solved with AI, step by step, in real time.

Meanwhile, the industry isn’t slowing down:
AI agents are gaining real control, infrastructure is scaling at record speed, and systems are starting not just to think, but to act on their own.

Here’s what you need to know from last week in AI:

New ChangeAI series: ‘’Friday AI Clinic’’
Elon Musk unveils Terafab AI chip megafactory
Anthropic gives Claude control of your computer
Luma AI launches Uni-1 image model
Zuckerberg builds personal CEO AI agent
ARC-AGI-3 resets AI reasoning benchmark
Google introduces TurboQuant memory breakthrough
Meta open-sources brain simulation model TRIBE v2
Agent-to-agent future reshaping businesses
Everything else that happened in AI last week

NEW CHANGEAI SERIES: “FRIDAY AI CLINIC”

Tired of AI webinars full of buzzwords but lacking real practical value? Then join our Friday AI Clinic, where we show how to apply AI in everyday work, from Excel to emails and reporting.

📅 Next session: Friday, April 3, 2026, 12:00–13:30
👉 Held every Friday, follow us on social media so you don’t miss the next one.

How it works:

1️⃣ You submit a problem
After registering, you’ll receive a short form to describe a process you want to improve (reports, analysis, data organization, writing, etc.).

2️⃣ We build solutions live
We select 3 real cases and create AI solutions from scratch, live on screen.

3️⃣ You gain real understanding
You’ll see every step, how to analyze workflows, evaluate AI tools, and handle AI mistakes.

🎁 Bonus: You’ll get the full session recording and all prompt templates, ready to use immediately.

Registration required: Secure your spot, get the templates, and submit your process.

👉 REGISTER NOW.

ELON MUSK UNVEILS TERAFAB AI CHIP MEGAFACTORY

Elon Musk revealed Terafab, a new Tesla-SpaceX-xAI facility designed to produce a terawatt of AI compute annually, roughly 50x current global output.

The Austin-based project would vertically integrate the entire chip process, from logic and memory to packaging and testing, something Musk claims doesn’t exist anywhere else. Two chip lines are planned: one for Tesla vehicles and Optimus robots, and another for space-based AI systems powered by solar energy and launched via Starship.

Musk also suggested that AI compute could move off Earth entirely, arguing that space-based infrastructure may become cheaper than land-based data centers within a few years.

Why does it matter for businesses?
The scale of AI infrastructure is becoming a competitive advantage. As compute becomes cheaper and more abundant, businesses will gain access to more powerful AI systems, accelerating automation, analysis, and product development.

ANTHROPIC GIVES CLAUDE CONTROL OF YOUR COMPUTER

Anthropic released a research preview that allows Claude to directly control a user’s desktop ,clicking, typing, and navigating across apps on macOS.

Combined with a new mobile tool called Dispatch, users can assign tasks remotely and let the system execute them on their computer. The system prioritizes direct integrations where possible, only using screen control as a fallback.

The feature is currently limited to Pro and Max users, with a Windows version in development. The release comes just weeks after Anthropic acquired computer-use startup Vercept.

Why does it matter for businesses?
AI is moving from assistant to operator. Instead of helping with tasks, systems can now execute entire workflows across tools, reducing the need for manual coordination and opening the door to fully automated back-office operations.

LUMA AI LAUNCHES UNI-1 IMAGE MODEL

Luma AI introduced Uni-1, a new image model that processes text and visuals through a single unified system.

Unlike traditional diffusion models, Uni-1 “thinks” through requests during generation, enabling more coherent outputs and better creative control. It supports use cases like infographics, manga, and stylistic edits, and ranked highly in human preference tests.

The model is also competitively priced, undercutting leading alternatives, though access is currently limited via waitlist.

Why does it matter for businesses?
Creative work is becoming fully AI-driven. Tools like Uni-1 signal a shift toward unified models that can handle design, content, and media generation, reducing reliance on multiple tools and speeding up production cycles.

ZUCKERBERG BUILDS PERSONAL CEO AI AGENT

Mark Zuckerberg is developing a personal AI agent designed to bypass traditional management layers and deliver instant answers.

The system can already retrieve information that would typically require multiple internal approvals. Inside Meta, employees are also building custom agents that read internal documents and interact with other bots directly.

Additional tools include AI systems acting as “chiefs of staff,” capable of pulling insights from across the organization on demand.

Why does it matter for businesses?
Decision-making is being compressed. AI agents can eliminate layers of communication, allowing leaders to access insights instantly, making organizations faster, leaner, and more responsive.

ARC-AGI-3 RESETS AI REASONING BENCHMARK

The ARC Prize Foundation released ARC-AGI-3, a new benchmark designed to test true reasoning capabilities in AI systems.

While humans can solve all tasks immediately, leading AI models scored below 1%, with the best-performing system reaching just 0.37%. The test requires agents to discover rules and strategies without instructions, simulating real-world problem solving.

A $1M prize is attached to the challenge, drawing increased attention from major AI labs.

Why does it matter for businesses?
Despite rapid progress, AI still struggles with general reasoning. Businesses should understand that while automation is improving fast, human oversight and judgment remain critical for complex, non-routine decisions.

GOOGLE INTRODUCES TURBOQUANT MEMORY BREAKTHROUGH

Google Research unveiled TurboQuant, a new method that compresses AI memory usage by over 6x without retraining models.

The approach also delivers up to 8x faster processing on Nvidia H100 chips, while maintaining near-perfect accuracy. This significantly reduces the cost and latency of handling long conversations and large datasets.

The research will be presented at ICLR 2026.

Why does it matter for businesses?
Efficiency is becoming as important as capability. Lower memory usage and faster processing mean cheaper AI operations, making advanced systems more accessible for SMBs at scale.

META OPEN-SOURCES BRAIN SIMULATION MODEL TRIBE V2

Meta released TRIBE v2, an AI model trained on brain scan data from over 700 individuals that can simulate neural activity across multiple functions.

The system expands dramatically on earlier versions, modeling up to 70,000 brain regions and outperforming real fMRI scans in predicting brain activity patterns.

Meta has open-sourced the model, allowing researchers to run experiments without physical scans.

Why does it matter for businesses?
AI is accelerating scientific research cycles. Breakthroughs like this can shorten development timelines in fields like healthcare and biotech, creating new opportunities for innovation-driven businesses.

AGENT-TO-AGENT FUTURE RESHAPING BUSINESSES

Industry leaders are increasingly pointing toward a future where AI agents communicate directly with each other to run business operations.

In this model, most workflows are handled autonomously between systems, while humans oversee decisions involving risk, finance, or legal responsibility. Even when agents execute tasks, final accountability remains with people.

This shifts the role of business owners from operators to supervisors of automated systems.

Why does it matter for businesses?
The skillset is changing. Running a business will increasingly mean managing AI systems, setting rules, and making high-level decisions, rather than executing day-to-day tasks.

EVERYTHING ELSE THAT HAPPENED IN AI LAST WEEK

• Microsoft – AI Talent Acquisition: Microsoft hired three top researchers from the Allen Institute for AI, including former CEO Ali Farhadi, to join its Superintelligence team. The move strengthens Microsoft’s position in the race for advanced AI systems. 
• Figma – AI Coding in Design: Figma opened its platform to coding agents, allowing tools like Claude Code to create and edit designs using existing brand components. This bridges design and development workflows in one system. 
• Roche – AI Factory for Pharma: Roche launched an AI-powered infrastructure with over 3,500 Nvidia Blackwell GPUs to accelerate drug discovery and production. The system is designed to scale research and manufacturing processes. 
• Cloudflare – Dynamic Workers: Cloudflare introduced Dynamic Workers, enabling AI agents to safely execute code they generate in real time at significantly higher speeds. The tool focuses on performance and security for autonomous systems. 
• OpenAI – $10B Funding Round: OpenAI is raising an additional $10B, pushing its total funding round beyond $120B with backing from major investors. The capital will support infrastructure, research, and product expansion.
• Google – Lyria 3 Pro Music AI: Google upgraded its music generation model to create full-length songs with structured sections like verses and choruses. The model is integrated into Gemini and Vertex AI platforms. 
• Sierra – Ghostwriter Agent Builder: Sierra introduced Ghostwriter, an AI system that builds customer service agents across voice and chat in over 30 languages. The platform automates the creation of support bots. 
• U.S. Department of Labor – AI Training via SMS: The U.S. government launched a free 7-day AI course delivered entirely through text messages. The initiative aims to improve AI literacy at scale. 
• Google – Gemini 3.1 Flash Live: Google released a faster and more realistic voice AI model designed for conversations across Search and APIs. The system improves responsiveness and task execution. 
• Cohere – Transcribe Speech Model: Cohere launched a free open-source speech recognition model that ranks #1 in accuracy across 14 languages. The model outperforms many existing solutions on benchmark tests. 

That's a wrap on last week in AI! From game-changing innovations to practical tools you can use today, the landscape is moving fast. Don’t forget to join our Friday AI Clinic to see real solutions in action, and stay ahead in the AI race. 

See you next week!