PLUS: GPT-5.5 class models on your laptop, Google's $75M bet on A24, and NVIDIA's liquid-cooled AI
Happy reading
Anthropic is taking AI out of the chat window and putting it right on your team. The company is launching a new agent that joins Slack channels as a persistent, collaborative member.
This represents a major shift from one-off commands to an AI that works alongside employees over hours or days. But as these digital teammates become more common, what will it mean for the future of team productivity and collaboration?
In today’s Next in AI:
Anthropic’s new AI agent for teams
GPT-5.5 class models on your laptop
NVIDIA’s new liquid-cooled AI hardware
Google and A24 partner on AI film tools
Meet Your New AI Teammate
Next in AI: Anthropic is launching Claude Tag, a new AI agent that integrates directly into your Slack workspace as a collaborative team member. You can delegate tasks, connect it to your tools, and let it learn from channel conversations to help your team get more done.
Explained:
Unlike a typical chatbot, Claude Tag acts as a single, shared identity within a Slack channel. This allows everyone to see what it's working on and build on previous interactions, creating a persistent, shared context for the entire team.
You can assign multi-step projects and it will work asynchronously over hours or days, even proactively flagging relevant information or following up on unresolved threads. Anthropic reports that its internal version already creates 65% of its product team's code.
The tool is built with enterprise needs in mind, allowing administrators to create separate Claude identities for different teams. This scopes its access and memory, ensuring an AI agent working with the sales team cannot access engineering data or tools.
Why It Matters: This marks a significant step in shifting AI from a simple query-and-response tool into a fully integrated digital teammate. AI agents that can operate autonomously within existing workflows could become the new foundation for team productivity.
GPT-5.5 On Your Laptop
Next in AI: Z.ai's new 744B-parameter model, GLM-5.2, now delivers performance rivaling top-tier models and can run on your local machine. This is made possible by Unsloth's new quantization techniques and its easy-to-use web interface.
Explained:
Z.ai's GLM-5.2 is a massive model with a 1M context window that performs on par with proprietary giants like GPT-5.5 and Claude 4.8 Opus on key benchmarks.
Unsloth's Dynamic 2-bit quantization shrinks the model by 84% from its original 1.5TB size while retaining ~82% of its accuracy.
To make it accessible, Unsloth Studio provides a free, open-source web UI for running the model locally on MacOS, Windows, or Linux, automatically handling complex setups.
Why It Matters: This development places elite AI capabilities directly into the hands of developers, removing dependencies on costly cloud APIs. It signals a major step forward for open-source AI, accelerating innovation by making state-of-the-art models more accessible than ever before.
AI's Hot Tub Moment
Next in AI: NVIDIA announced its next-gen AI infrastructure will be 100% liquid-cooled, using a system that operates with coolant as hot as 45°C (113°F). This breakthrough dramatically reduces the energy and water needed to power the world's most powerful computers.
Explained:
The higher 45°C operating temperature is key, as it allows data centers to use outdoor dry coolers instead of energy-intensive chillers for much of the year.
This new design can cut a facility's cooling water consumption to near zero in favorable climates, a massive shift from the millions of gallons typically used per megawatt annually.
Fully liquid-cooled servers eliminate the need for fans and traditional hot/cold aisles, enabling higher rack density and making AI factories significantly quieter.
Why It Matters:
As AI's demand for power and resources skyrockets, this kind of hardware efficiency is crucial for sustainable growth. Innovations like this change the fundamental economics and environmental footprint of building and operating AI at hyperscale.
Google's A24 Partnership
Next in AI: Google is investing $75 million in indie film studio A24 to develop new AI tools for filmmakers. The collaboration with Google's DeepMind unit aims to create tech that supports creative risk-taking rather than simply automating production.
Explained:
The partnership’s goal is to preserve creative control and support artistic risks, deliberately moving away from AI models focused on prompted content generation.
The official research partnership is a multi-year, non-exclusive deal that gives A24 access to DeepMind's infrastructure without granting Google access to its film library or data.
Initial applications will explore tools like AI-generated storyboards, a contrast to other Hollywood deals that use AI to generate new IP or make post-production tweaks without reshoots.
Why It Matters:
This deal positions AI as a creative collaborator rather than a replacement for human artists. If successful, it could provide a blueprint for how the tech and entertainment industries can work together, helping to win over skeptical creators and audiences.
AI Pulse
Alphabet slid 5% for its worst day in over a year following the departure of high-profile researchers, including AlphaFold co-creator John Jumper, to rivals Anthropic and OpenAI.
The Five Eyes warned in a rare joint statement that frontier AI models capable of disrupting governments are months, not years, away, urging a whole-of-society response to the threat.
Meta paused its internal AI training program that uses employee keystrokes as data after a leak exposed sensitive information, including private conversations and performance reviews, across the company.
Oak launched a new open-source version control system built specifically for AI agents, claiming to be 90% faster and use 50% fewer VCS-related tokens than Git.