OpenAI and Google Escalate the AI Agent Race
The competition to own the enterprise workflow intensified this week with significant announcements from two of the industry's biggest players. Both OpenAI and Google are pushing further into creating sophisticated, collaborative AI systems designed for team environments.
Introducing OpenAI Workspace Agents
OpenAI has officially launched OpenAI Workspace Agents, a new feature available in a research preview for select ChatGPT plans. These are shared AI agents that teams can create and deploy to handle complex, multi-step tasks and workflows. Powered by Codex, they can generate reports, write code, and manage communications across integrated tools like Slack.
These new agents are being positioned as an evolution of the solo-user Custom GPTs introduced in 2023. OpenAI confirmed that existing GPTs will continue to function, with a conversion tool planned for the future. The company is even using them internally for tasks like sales account research and accounting reconciliations, showcasing their practical application in a business setting.
Google's Counter with Gemini Enterprise Agent Platform
Not to be outdone, Google debuted its Gemini Enterprise Agent Platform. This is a comprehensive solution for building, scaling, and governing enterprise-grade agents. It integrates model selection and building capabilities with new features for DevOps, orchestration, and security.
Alongside the platform, Google launched Workspace Intelligence, a semantic layer that integrates emails, chats, files, and projects. This allows for powerful new features, such as building spreadsheets in Google Sheets using natural language. This move aims to position Google Workspace as a central control layer for business operations, powered by the Gemini model series.
Breakthroughs in Open Source and Image Generation
Beyond the agent-focused updates, the industry also saw significant advances in open-source models and AI-powered imaging, offering developers and creators powerful new capabilities.
Alibaba's Qwen3.6-27B Model Raises the Bar
Alibaba's Qwen team has open-sourced Qwen3.6-27B, a 27-billion-parameter dense model that is making waves in the coding community. According to the development team, this smaller model surpasses its own massive 397B predecessor across all major coding benchmarks. Its manageable size, with quantized versions available on Hugging Face, makes flagship-level agentic coding performance accessible on consumer-grade hardware.
The Quiet Debut of ChatGPT Images 2.0
OpenAI is also quietly testing GPT Image 2, its next-generation image model, on the LM Arena platform. Early tests reveal a significant leap in quality, particularly in generating coherent and typo-free text within images. Users are already creating everything from realistic UI screenshots and multi-page magazines to creative QR codes, signaling a new challenger to other models in the AI image market.
A Flood of New and Specialized AI Tools
The ecosystem continues to expand with a diverse range of new tools targeting specific use cases from video creation to development and productivity.
| Tool Name | Primary Function | Key Feature |
|---|---|---|
| Ella | Video Creation | Generate, edit, and export video using multiple AI models in one workflow. |
| Ideogram Custom Models | Image Generation | Fine-tune image models on your own assets for consistent branding. |
| Chronicle | Presentations | Turns ideas into presentation slides, described as a 'Cursor for slides'. |
| Brex CrabTrap | AI Security | Open-source proxy that inspects and blocks risky outbound agent actions. |
| Somio | Music Generation | Turns text or lyrics into melodies using AI. |
| Claras | Productivity | Transcribes YouTube videos and allows you to chat with the content. |
Enhancements for Developers and Researchers
Several new tools are focused on improving the developer and researcher experience. Euphony is a new open-source viewer from OpenAI for inspecting chat data and Codex session logs. For those working on agentic workloads, Applied Compute has released an open-source benchmarking tool. Hugging Face also launched ml-intern, an open-source agent designed to devise and run research experiments.
Niche Productivity and Automation
Other notable launches include Norm by Bland AI, which builds production-ready voice agents from a single prompt, and TinyFish, an API that provides agents with scalable web access tools like a stealth browser and web fetch capabilities. On the consumer side, noscroll offers to doomscroll X (formerly Twitter) for you and text you a summary, while Toki helps organize and plan your time with AI.
