AI agents go enterprise
- OpenAI and Google rolled out “agent” tools that let models act across business apps and automate multi-step tasks. - OpenAI introduced Workspace Agents and Agent Mode while Google announced new agent tooling at Cloud Next. - Those products position models as workflow orchestrators that can connect Slack, Salesforce and browsers, changing where enterprise work gets done ( ).
OpenAI and Google both spent April 22 pitching the same idea to corporate buyers: artificial intelligence should not just answer questions, but do work across the software companies already use. (openai.com) OpenAI said its new Workspace Agents in ChatGPT are “an evolution of GPTs” for Business and Enterprise customers, built to handle multi-step tasks, run in the cloud, and keep working after a user closes the chat. The company said teams can use them inside ChatGPT or in Slack, with rollout starting over the next few weeks. (openai.com, help.openai.com) OpenAI’s help pages say Workspace Agents can connect to tools including Google Drive, Google Calendar, Slack and Microsoft SharePoint, and can be created from templates, previewed before publishing, shared inside a workspace, and scheduled to run automatically. A separate ChatGPT “agent” product can browse websites, fill out forms, edit spreadsheets and connect to outside data sources while keeping the user in control of approvals. (help.openai.com, help.openai.com) Google made a parallel case at Cloud Next 2026 in Las Vegas, where it framed the “agentic enterprise” as the next phase of cloud software and introduced a Gemini Enterprise Agent Platform for building, governing and optimizing agents. Google said the platform combines Vertex AI model tools with new features for agent integration, security and operations. (cloud.google.com, blog.google) Google also launched an official Agent Skills repository for products including BigQuery, Google Kubernetes Engine and the Gemini application programming interface, aiming to give agents smaller, task-specific bundles of product knowledge instead of stuffing full manuals into prompts. (cloud.google.com) The shift here is from chatbots as front ends to agents as coordinators. Instead of a worker copying information from Slack into Salesforce, checking a browser tab, then drafting an update, the model is being sold as the software layer that moves among those systems itself. (venturebeat.com, openai.com) That pitch lines up with where enterprise spending is moving. Reuters reported from Cloud Next that Google is putting AI agents at the center of its enterprise software push, while Bloomberg said the new Google tools are aimed directly at OpenAI and Anthropic in the race to automate office work. (msn.com, bloomberg.com) Both companies are also selling control, not just autonomy. OpenAI said Workspace Agents operate within an organization’s permissions and controls, and Google said its new platform is built around governance, security and optimization for technical teams that have to manage fleets of agents, not one-off demos. (openai.com, blog.google) The immediate question is whether companies trust these systems with real workflows, not just experiments. After April 22, the competition is less about which model writes the best paragraph and more about which one can move through the most business software with the fewest human handoffs. (openai.com, blog.google)