AI Agents: Redefining the Future of Work
It’s a Monday morning, and the to-do list is already daunting. Perhaps you’re buried in invoices, need to update field technicians, or struggle to provide efficient IT support. Now, imagine getting help with all of this – all while you focus on your team’s long-term vision. This is the promise of AI agents.
Agents are designed to work with you or on your behalf. They can manage projects, reconcile financial statements, and handle various other complex tasks. Microsoft 365 Copilot, for instance, is a personal assistant designed to tackle daily tasks and stimulate creative projects. Interacting with diverse agents through this platform opens a world of possibilities for businesses, empowering employees and driving efficiency. Agents can operate around the clock, reviewing and approving customer returns or analyzing shipping invoices to prevent supply-chain issues. They might also scan extensive product information to provide field technicians with step-by-step instructions, or manage IT help desk tickets.
“Think of agents as the new apps for an AI-powered world,” says Jared Spataro, Microsoft’s chief marketing officer for AI at Work. “We’re adding new capabilities to make the biggest pain points at work go away and drive real business results.”
Understanding AI Agents
AI agents represent the next evolutionary step in generative AI. Unlike assistants that simply help, agents actively work alongside you or independently on your behalf. These agents can respond to questions and manage complex, multi-step assignments. What sets them apart is their ability to be customized for specialized tasks, providing them with unique expertise.
For instance, imagine an agent with comprehensive knowledge of your company’s product catalog. It could draft detailed responses to customer inquiries or prepare product details for presentations. Other agents can handle sales order fulfillment, allowing you to focus on building relationships. By automating these routine activities, AI agents can dramatically boost productivity across a wide range of industries.
Microsoft 365 and Dynamics 365 already offer pre-built agents, and Copilot Studio allows for the creation of custom agents, too. Consider a salesperson with ambitious quarterly goals. With Copilot, they can draft emails, summarize missed meetings, and design sales presentations. Concurrently, a sales lead generation agent identifies new prospects for the salesperson to follow up with. Through a partnership between Copilot and specialized agents, end-of-quarter targets become attainable.
The Evolution of AI Agents
AI agents are not a new concept. Microsoft has been researching this area extensively and even created a multi-agent library for developers worldwide, a project that has greatly influenced the current capabilities of agents. Their increased prominence is a result of recent advancements in LLMs, which allow anyone to communicate with AI. This combination makes AI tools more practically useful.
“People expect AI to do things for them,” not just produce text, says Ece Kamar, the managing director of Microsoft’s AI Frontiers Lab. “If you want a system that can really solve real world problems and help people, that system has to have a good understanding of the world we live in, and when something happens, that system has to perceive that change and take action accordingly.”
Essentially, agents operate as layers on top of language models, observing and gathering information. They then provide input to the model to formulate an action plan and communicate that to the user. Both agents and models are vital components of generative AI tools.
The continued evolution of agents relies on advances in three key areas: memory, entitlements, and tools. Memory ensures continuity, preventing each request from starting from scratch. As Sam Schillace, Microsoft’s deputy chief technology officer, explains, “To be autonomous you have to carry context through a bunch of actions, but the models are very disconnected and don’t have continuity the way we do, so every prompt is in a vacuum and it might pull the wrong memory out. It’s like you’re watching a stop-motion animation, one isolated frame after another, and your mind puts it into motion. The clay model doesn’t move on its own.”
To address this, Schillace’s team is developing a process of chunking and chaining. This entails breaking down interactions into interconnected components. Those components can then be stored and cross-linked by relevance, similar to memory. Entitlements ensures that agents have secure access to necessary, authorized information – like who your boss is, for example – while tools provide access to the computer programs needed to operate. This includes applications like Teams and PowerPoint, allowing agents to take action.
Building and Utilizing AI Agents for Work
Users are now able to easily create agents in Microsoft 365 Copilot to assist with daily work, similar to creating a spreadsheet or presentation, without coding skills. Copilot Studio empowers people without coding knowledge to build agents. They can be linked to business data—emails, reports, and customer management systems—to perform tasks and share valuable insights. Several new agents will soon be available in Microsoft 365, designed to streamline common workflows.
Interpreter in Teams can provide real-time speech-to-speech translation during meetings and even simulate your voice. The Employee Self-Service Agent simplifies HR and IT help desk related tasks, such as helping workers resolve a laptop issue or find out if they’ve maxed out certain benefits. It can connect to different company systems with additional customization within Copilot Studio.
Microsoft Dynamics 365 also supports agents for common business workflows in sales, supply chain, finance, and customer service functions. Plus, every SharePoint site will soon have an agent tailored to the organization’s content. It will allow employees to quickly discover project details, internal memos, and more.
Developers have even more options. The new Azure AI Agent Service allows them to select small- or large language models to design, develop, and scale automated agent-powered apps. This streamlines complex tasks, such as order processing and customer data synchronization. It offers a software development kit (SDK) with tools for agent development, permitting efficient integration of agent capabilities using Visual Studio Code and GitHub.
One type of model, OpenAI’s o1 series, provides advanced reasoning capabilities, which empowers agents to manage more complex assignments by breaking them down into steps. These agents could gather the information that an IT help desk employee would need to resolve an issue, consider previous solutions, and establish a new plan.
Agents are also being developed for inclusion in LinkedIn, with the social media platform’s initial agent designed to aid recruiters in their hiring process.
Prioritizing Responsible AI and Risk Assessment
As AI agents take on more autonomy, it’s crucial, according to Sarah Bird, chief product officer of Responsible AI at Microsoft, to address safety concerns. “Agents certainly up the stakes from a responsible AI point of view,” Bird says. “So we have to have much, much lower error rates. And there’s many more nuanced ways in which something could be an error. This is the big challenge with agents.”
The Copilot Control System provides IT departments with tools for managing Copilot and agents. This includes data access, governance, management, security controls, measurement reports, and tools to track adoption and business value. Agents built for Microsoft 365 and Dynamics 365 often have “human in the loop” approvals, which means a person must review and send emails or other communications.
Ultimately, accuracy requires careful testing and moderation. As Bird notes, organizations should choose a suitable starting point to meet specific needs.
Looking Ahead: The Future of AI Agents
Kamar, who has been dedicated to AI agents since 2005, envisions a new ecosystem or marketplace of agents, akin to how smartphone apps enable users to increase their productivity. Agents already have “the basic building blocks of what it takes to complete a task,” she says. “Like observing, ‘I can see your meeting is taking longer; I should delay the next meeting.’”
As agents continue to develop autonomy through advances in memory and entitlements, they grow in usefulness. They’re alleviating pain points for employees by offering support with expense reporting, project management, and facilitating meetings. Their value allows businesses to thrive by managing duties like inventory alerts. Ultimately, agents are making tasks simpler for employees and expanding business capabilities.
“AI agents are not only a way to get more value for people but are going to be a paradigm shift in terms of how work gets done,” Kamar says.
As Microsoft is adding new capabilities like Copilot Actions, designed to help tackle time-consuming tasks, such as composing emails or generating routine reports, Copilot is set to grow and improve every employee’s ability to work efficiently.
“Copilot will empower every employee to do their best work in less time, and focus on more meaningful tasks,” Spataro says. “And agents created in Copilot Studio will transform every business process, helping companies streamline operations, enhance collaboration and drive innovation at scale.”
Illustration of various AI agent features.
