Microsoft’s AI ‘agents’ can pick up office workers’ tedious tasks

“Virtual employees” capable of completing increasingly complex office tasks have been developed by Microsoft in the latest leap in how artificial intelligence is applied in the workplace.
Called “autonomous agents”, the sophisticated bots can “understand the nature of your work and act on your behalf providing support across business roles, teams, and functions”, the company claims. They can monitor and respond to emails and perform various everyday functions, with the obvious advantage over humans that they can work 24 hours a day, seven days a week.
The announcement, at an event in central London by Satya Nadella, the third boss in Microsoft’s 50-year history, underlines the significance of changes being wrought to the service industry by the advances in artificial intelligence.
“For 70 years, computing history has been about digitising people, places and things,” Nadella said. “We now have a new reasoning engine to make sense of it.”
The way people interact with computers is changing, using natural language, while it is multimodal using image, speech and text with more memory and context, he said, adding that improved data and algorithms is doubling the performance of AI models every six months.
Other companies racing to develop these kinds of autonomous agents include Google, Salesforce, Sierra, founded by Bret Taylor, the chairman of OpenAI, and Harvey, which focuses on agents for law.
Jared Spataro, chief marketing officer for AI at work at Microsoft, likened the agents to robots doing manual work in factories. “We believe that every organisation is going to be a mix of human workers that are doing very important work,” he said. “But also increasingly these agents that will be working under the supervision of humans and then autonomously reporting back to humans.”
• David Wighton: Better skills will decide the winners in AI race
Businesses that had an early trial of this technology include McKinsey, which has built an agent that can help its teams process new client proposals more quickly. It scans email approaches and extracts information to establish where to direct people, checks if they have been in touch before, draws from company data and creates and sends email responses.
Companies will now be able to develop bespoke “agents” for their businesses. Ten are being made available immediately for different jobs such as sales or case management. They can “scale your teams’ capacity like never before”, Microsoft said.
Given instructions and access to all of a company’s information, they can “reason” and get to grips with the scale of a task to work out what needs doing, the company claimed. In complex circumstances, if it gets stuck, the agent will “escalate” a question to a human.
Microsoft already has 2.1 million monthly users of Copilot, its AI tool that provides services such as transcription and summaries of video calls, email and document drafting. The latest product launch, trailed earlier this year, needs far less hand-holding.
• Microsoft signs deal with Whitehall
Ece Kamar, managing director of AI Frontiers at Microsoft, described the relationship between autonomous agents and their human colleagues as “a collaboration”, adding: “My personal goal is not these agents taking over jobs, but rather creating the right interactions and workflows where the agents are able to take over the tedious tasks that people don’t want to do, and then humans can do the interesting parts of their work”.
In the future, “it is very obvious that there could be negative consequences from these actions”, she said. “How we build the right transparency layers, stopgaps [and] preventions in place for that is another important area that we will need to investigate and get right.”
Spataro said the agents would transform the workplace. “The organisation that we know so well today is really an artefact of the two world wars. Today, as you walk into any firm you have the same thing: an HR department, a sales department, a marketing department, a finance department. Almost every firm in the world on the face of the planet looks the same,” he said.
“We think in the future more generalised leaders will be able to do more specific tasks. What we’ve seen so far is that this means flatter structures and more generalised skills as opposed to the specialisation that we’ve seen over the last couple of hundred years.”
The aim is that the agents will learn from their work and be able to interact with each other and humans, act and make decisions without the need for constraint supervision.
The launch of agents is another coup for the tech behemoth, which has been quick to steal a march on its competitors and turn generative AI into a usable product for businesses, thirsty for its productivity gains.
Microsoft was an early partner of OpenAI, the developer of ChatGPT. It has investments in other so-called frontier labs building large language models such as Mistral.
The company hired Mustafa Suleyman, the co-founder of Google’s DeepMind, in March along with his team at Inflection AI to form a new organisation called Microsoft AI, focused on advancing Copilot.
However, these relationships with cutting-edge AI firms have caught the attention of antitrust watchdogs on both sides of the Atlantic. In the United States the Federal Trade Commission has launched a broad inquiry into Big Tech’s generative AI investments and partnerships, including that between Microsoft and OpenAI.
The UK competition authority ruled that Microsoft does not have to face further scrutiny over its deals with Inflection and Mistral, but its examination of the relationship with OpenAI continues.
Microsoft’s share price has risen 27 per cent in the past year, giving it a market cap of $3.1 trillion. In its latest quarterly earnings published in June the company reported revenue of $64.7 billion, up 15 per cent, and net income of $22 billion, up 10 per cent.