Enterprise AI moves to agents

A new fight in enterprise artificial intelligence is taking shape around agents: software that can keep working after a prompt ends, inside the apps companies already use. OpenAI rolled out workspace agents on April 22, and Google answered the same week at Cloud Next with a broader agent platform push. (openai.com) (cloud.google.com) OpenAI said its workspace agents are powered by Codex, run in the cloud, and are built to handle “complex tasks and long-running workflows” for teams. The company said they can be shared across an organization and used in ChatGPT or Slack under existing workplace permissions and controls. (openai.com) (help.openai.com) OpenAI’s help documentation says admins can turn workspace agents on for eligible ChatGPT Enterprise workspaces, connect them to apps and tools, test them before publishing, and run them on a schedule. OpenAI’s apps page lists business integrations including Slack, SharePoint, Airtable and Google Drive, while outside reporting said Salesforce is part of the new enterprise pitch. (help.openai.com) (chatgpt.com) (venturebeat.com) Google used Cloud Next in Las Vegas to reframe its own stack around agents. The company introduced Gemini Enterprise Agent Platform on April 22 as the successor to Vertex AI for building, scaling, governing and optimizing agents. (cloud.google.com 1) (cloud.google.com 2) Google also announced a $750 million partner fund for agentic development and said partner-built agents would be available inside Gemini Enterprise. Google’s partner blog said those agents can come from companies including Accenture, Adobe, Atlassian, Oracle, Salesforce, ServiceNow and Workday. (cloud.google.com 1) (cloud.google.com 2) An agent is basically a model with tools, memory and permission to act, not just answer. That shifts the product contest from who has the best chatbot demo to who can connect safely to email, chat, files, customer records and internal approval systems. (docs.cloud.google.com) (openai.com) (cloud.google.com) That is why both companies are emphasizing governance language alongside model language. OpenAI says workspace agents operate within organizational controls, while Google says Gemini Enterprise is built to centralize visibility and control over Google-made, third-party and custom agents. (openai.com) (cloud.google.com) Google is also pushing standards as part of the platform argument. Its Agent2Agent, or A2A, protocol is meant to let one agent communicate with another across tools and vendors, and Google says the protocol now has support from more than 150 organizations. (developers.googleblog.com) (cloud.google.com) The timing is not accidental. OpenAI’s launch turns custom workplace bots into shared, persistent workers, while Google is bundling models, orchestration, security and partner distribution into one enterprise sales pitch. (openai.com) (cloud.google.com) The next test is whether companies trust these systems with real work, not just drafts and demos. In this phase of enterprise artificial intelligence, the winning feature may be less about sounding smart than proving who did what, in which system, with whose permission. (help.openai.com) (cloud.google.com)