Microsoft says AI agents that work for hours or days need a different computing foundation, and offers three questions to test whether your business is ready.
Microsoft says the next stage of AI is software that works on a task for hours or days, not just one that answers a single question. It argues this needs a different kind of computer foundation. The claim comes from a post on the Microsoft Azure blog, which carries a note saying it was written and paid for by Microsoft.
An AI agent is software that carries out a job on its own, step by step. Microsoft says one agent may move between reasoning, finding data, using tools and running code, while keeping its progress along the way. Its needs can change mid-task. It may need a burst of computing power, then fast access to large amounts of data, then a pause while another tool or agent finishes.
Earlier cloud software often followed predictable paths, such as a request followed by a response. Microsoft says agents are different because their work is long and unpredictable. Its answer is that chips, computing, storage and networking should be designed and tuned as one system. More computing power alone will not help, it says, if data cannot move quickly or progress cannot be saved between steps.
The post points to Levi Strauss & Co. The company moved its data centers to Microsoft Azure, consolidated its ERP platforms, modernized applications and unified its data in a governed, real-time layer. It is now building a "super agent" to coordinate more than 1,000 agents, each focused on a specific area of the business.
Microsoft suggests technical leaders start with these checks:
The post gives no test results, prices or timelines. Levi Strauss is described as building its super agent, with no results shared. The post also concludes that Azure is the right foundation, so it works as a sales pitch. The three questions are still a reasonable way to review your own systems, whichever vendor you use.
Before committing to AI agents, check whether your systems can handle long, unpredictable work, and treat vendor claims with care.
World models, AI that learns physics from video, are being built into home robots and driverless cars, and your own videos may help train them.
Read more: AI that learns how the physical world works is coming to robots and carsA new Alibaba-led test of real coding requests found the best AI agent fully finished just 26.4 percent of tasks, so human review still matters.
Read more: AI coding helpers fully finish only about 1 in 4 real tasksA University of Chicago law professor asked Claude to fix page numbers. It deleted his syllabus and a year of notes instead. Dropbox saved him, not the AI.
Read more: An AI assistant deleted a law professor's year of lecture notes
