Microsoft remains convinced that AI-driven agents will reshape how work gets done, and the company has introduced a new tool meant to let businesses manage those programs much like they manage human staff. The offering arrives at a moment when companies are experimenting with generative AI inside core processes, creating a need for an administrative layer that keeps automated helpers visible and controllable.
Called Agent 365, the product is designed to help organizations corral an expanding fleet of automated assistants. Microsoft positions the tool as a way to bring order to agents that are being deployed across tasks such as email triage, scheduling, vendor interactions, and procurement. Companies that start running many different bots inside their software environments can use Agent 365 to keep an inventory, gauge performance, and change operational settings.
The company stresses that Agent 365 is not intended to be a development platform for enterprise AI. Rather, it functions as a control plane where IT and ops teams can group agents, apply policies, and modify what each agent is allowed to do. Microsoft said the tool is rolling out today in its early access program for customers that want to test management at scale.
Microsoft described the product as a trackable workspace for agents. “Tools that you use to manage people, devices, and applications today, you'd want to extend them to run agents as well in the future,” says Charles Lamanna, a president of business and industry for Microsoft’s Copilot, its AI chatbot.
Lamanna sketched out how quickly agent counts could multiply inside an organization. He suggested that if a company employs 100,000 people, it might end up running “half a million to a million agents,” assigned to duties from basic email organization to managing the “whole procurement process” for a business. He added that Microsoft itself uses millions of agents internally, a signal of the scale the company expects.
Those automated programs can be given permission to take actions within corporate systems and automate parts of employees’ routines. That capability brings efficiency gains, yet it can be difficult to track who created an agent, what data it touches, and whether it is still operating correctly. Security teams warn that a lack of centralized oversight raises the prospect of misconfigurations or misuse that could expose sensitive information.
Agent 365’s central feature is a registry that lists an organization’s active agents in a single location. Each agent receives a specific identification number and associated metadata describing how people at the company are using it. From that console administrators can alter configurations, change an agent’s permissions, and limit the portions of an enterprise application the bot can access.
The product includes tools intended to monitor agent behavior in real time and to produce records for later review, so that operations and security staff can see what an agent did and when. “As data flows between people, agents, and applications,” says Lamanna, “it stays protected.” Those monitoring capabilities are meant to give IT teams a way to respond if an agent acts unexpectedly.
As pilot programs expand, questions about safety have become more common among companies that are thinking about putting agents into workflows that contain confidential business records. A "prompt injection attack" — where a website or app contains hidden instructions that try to take control of an agent or alter its outputs — is one example of the sorts of vulnerabilities researchers have been flagging in current agent designs.
Lamanna argued that leaders who resist widespread agent deployment over concerns about security or occasional errors are pushing against a broader shift in workplace technology. “Resisting having agents enabled is kind of like resisting giving internet or PC access to your employees,” he says. Microsoft has been steering its enterprise offerings so customers can adopt generative AI alongside their existing software subscriptions.
Across the tech industry, many leading AI vendors have concentrated on agent technology this year. The implementations vary, and the systems can be brittle: an agent that looks useful in a lab can still introduce surprising mistakes into automated processes if it follows an unexpected instruction or misinterprets a prompt.
Hands-on evaluations done over recent months have tended to emphasize consumer-facing scenarios rather than the enterprise applications that Agent 365 targets, and testers often found limits to what current agents can reliably accomplish. In some early trials, agents failed to complete routine errands such as shopping for a birthday gift, underscoring reliability gaps that will matter if organizations scale agent usage for business tasks.
White-collar workers who face pressure from managers to bring agents into daily work can expect that push to continue into the coming year. “2025 is the year of agents,” says Lamanna. “2026 will be even more agents.”