Anthropic Supercharges Claude with Major Cowork Update
Anthropic is escalating the enterprise agent war with a major Claude Cowork update. The platform is now armed with department-specific AI agents, private plugin marketplaces, and crucial new software connectors, marking its transition into a full-fledged enterprise solution designed to boost employee productivity.
Organizations can now deploy pre-built agents covering ten departments, including HR, engineering, investment banking, and private equity. This is paired with new connectors for essential tools like Google Workspace (Gmail, Drive), DocuSign, FactSet, and WordPress. These integrations allow Claude to pull context and data directly from enterprise software without users needing to switch interfaces.
A key feature of this update is the ability for companies to build and manage their own private plugin stores. Administrators can push custom AI agents to specific teams and use granular controls to assign and limit access, ensuring secure and relevant AI assistance across the organization. In a compelling new research preview, Anthropic also demonstrated Claude's ability to operate across Microsoft Excel and PowerPoint, pulling data from a spreadsheet to build a complete slide deck.
New Agentic Platforms and Foundational Models
The AI tooling landscape is rapidly advancing beyond single-purpose assistants. Several new platforms and models are focused on creating autonomous, persistent agents and providing the high-speed reasoning necessary to power them.
Inception Labs Launches Mercury 2
Inception Labs has introduced Mercury 2, a novel "diffusion LLM" that edits text rather than generating it sequentially. This architectural difference results in incredible speed. Independent tests show Mercury 2 reaching 1,196 tokens per second, a speed over three times faster than its nearest competitor in the same price class. It's built for production speed in agent loops where latency is critical.
|
Model |
Speed (Tokens/Sec) |
Key Feature |
|---|---|---|
|
Inception Mercury 2 |
1,196 |
Diffusion-based reasoning for extreme speed |
|
Anthropic Claude 4.5 Haiku |
~89 |
Balanced speed and intelligence |
|
OpenAI GPT-5 Mini |
~73 |
General-purpose use |
Standard Intelligence FDM-1 Learns from Video
A new model from Standard Intelligence called FDM-1 learns to operate computer software by watching videos. Trained on 11 million hours of screen footage, it can replicate complex tasks like CAD modeling in Blender and has even been used to drive a car in San Francisco using live data feeds. This approach could dramatically expand the capabilities of computer-use agents.
Other Notable Model and Platform Releases:
-
Notion Custom Agents: Notion has shipped fully autonomous AI agents that can be configured in plain language to run 24/7 across tools like Slack, Mail, and Figma to automate repetitive tasks.
-
KiloClaw: This service from Kilo allows anyone to deploy OpenClaw agents in under 60 seconds, providing an "always-on" capability with built-in monitoring and access to over 500 models.
-
Qwen3.5-35B-A3B: This powerful open-source model family from Hugging Face integrates multimodal learning and supports a massive context length of up to 262,144 tokens. It can run locally on a Mac with just 24GB of RAM.
Fresh AI Tools for Productivity and Creativity
A wave of new tools aims to integrate AI seamlessly into daily workflows, from managing emails to generating high-quality media content. These applications focus on saving time and enhancing creative output without requiring deep technical knowledge.
Productivity and Workflow Enhancers:
-
Oura AI Model: Oura launched its first proprietary AI model dedicated to women's health, powering its Oura Advisor chatbot for personalized guidance.
-
Wispr Flow: This voice-to-text tool, now available on Android, cleans up spoken words into ready-to-send text, understanding complex code syntax and technical terms.
-
Slackbot: A context-aware AI agent inside Slack that can search your entire workspace, automate tasks, and synthesize information instantly.
-
Lemon & Granola: Lemon is a Mac app for replying to emails with voice commands, while Granola is an AI notepad that synthesizes meeting transcripts with your notes for actionable summaries.
Creative and Content Generation Tools:
-
ProducerAI: Now part of Google Labs, this generative music platform is powered by DeepMind's Lyria 3 model for creating and refining tracks.
-
HeyGen: A powerful and easy-to-use tool for translating videos into other languages, cloning the speaker's voice, and adjusting lip movements.
-
Lunair: An AI tool designed to generate studio-quality explainer videos from simple prompts.
-
Reve v1.5 & Seedream 5.0 Lite: Two new text-to-image models, with Reve offering 4k resolution outputs and ByteDance's Seedream providing an upgraded image generation experience.
-
Gemini Video Templates: Google's Gemini has rolled out new video templates for Veo 3.1, simplifying the creation of AI videos from text.
New Tools for Developers and Businesses
The developer community and business sector also saw a number of significant releases, focusing on coding automation, UI generation, and enterprise readiness.
-
Devin 2.2: An update to the AI software engineer from Cognition AI that now writes, tests, and auto-fixes its own code with a 3x faster startup time and a redesigned UI.
-
Google Stitch: This tool from Google generates complete app and web UI designs from text prompts, which can then be exported to Figma or as clean frontend code.
-
Claude Code Remote Control: A new feature allowing developers to start a coding session in their terminal and seamlessly hand it off to their phone to monitor or continue the task.
-
WorkOS: A platform designed to make applications enterprise-ready in days, helping startups close deals without rebuilding their entire stack.
-
Basis: An AI agent platform specifically for accountants, which just raised a $100M Series B to expand its capabilities across tax, audit, and advisory services.