Google Cloud puts Gemini agents in the company directory as named coworkers

Google has built the agent into the systems companies already use to manage people ^[1]. A persistent coworker gets an address on a separate @agents.company.com domain and its own Drive storage ^[4]. Coworkers can call on it with an @mention in Chat or a tag in a Doc comment, and version history records its edits under its own name ^[5]. Google says the agent sees only information shared with it ^[5]. Under that rule, each colleague who shares a file with the agent decides what it can read.

Google Cloud CEO Thomas Kurian said, "You give it objectives, not just instructions." ^[3] An instruction leaves the judgement with the person who gave it. An objective hands part of that judgement to the agent. The controls Google has described apply to the software itself: administrators can restrict which models agents use, and Smart Routing picks among the permitted models based on task requirements and cost ^[7]. In my view the harder choice is about people. Someone in the business has to own each objective an agent is given and sign off on what it produces.

Mahmoud Ramin, a research director at Info-Tech Research Group, argued that the launch does not open a new category, since multi-agent capabilities are already on offer from OpenAI, Anthropic, Microsoft and Meta ^[11]. On capability, the release dates support him: Google launched 13 days after Microsoft's Copilot update ^[20]. What a manager has to act on is the identity model. An agent with a directory entry and an attested identity is an account that someone has to provision and later retire ^[6]. That work falls to IT and HR as well as to whoever bought the software.

Little of this needs settling this week, though an account policy is easier to write before agents arrive in large numbers. The agent is in private preview and open only to selected customers ^[14]. Google said nearly 90% of Fortune 100 companies use Gemini Enterprise, and that close to 500 of its Google Cloud customers had each handled over a trillion tokens in the year before ^[10]. Those figures cover existing products and do not measure adoption of the new agent, implicator.ai noted ^[10]. Google has supplied no independent benchmarks of how reliably the agent completes long assignments across applications ^[16].

The order of decisions matters most in regulated work. Google has preview versions for financial services and legal work. The finance version carries more than 50 skills plus FactSet and LSEG data, and the legal version connects to NetDocuments and iManage ^[9]. In both sectors, firms are expected to show who did what on a client file. A company that sets its agent-account rules this quarter will set them before the UK data regulator has formed a view. Britain's Information Commissioner's Office opened a general call for evidence on agent risks on the day Google launched ^[17].