“When we talk about transformation, we all love to talk about the tech. The reality of transformation is to all our business processes,” explained Ray Smith, VP of AI Agents, Microsoft, as he kicked off Day 2 of AI Agent & Copilot Summit with the opening keynote, “How Agents are Defining the Future.”
During his discussion, Smith explained how transformation starts with an outcome focus – we need to be looking at our processes and reimagining those processes to be more efficient and drive growth. Process improvement is one of the three types of transformation made possible with AI; the other two include general employee productivity and creating new products and experiences.
Overview of Agents and Agentic Transformation
To lay the foundation, Smith gave a direct description of an agent: Agents are apps that use AI to reason, plan, connect to systems and execute business processes, working alongside or on behalf of a person, team, or organization. Further, agents vary in complexity and capabilities depending on your need, ranging from “simple” (retrieval) to “advanced” (autonomous).
Agents fundamentally change how we run our business, service customers, and redefine our value, noted Smith. There are four levels of agentic transformation:
- Level 1: Human first
- Level 2: Human + agents
- Level 3: Agents-first
- Level 4: Agents only
Key components of agents include user experiences, orchestrators, knowledge, tools, autonomy, foundation models, as illustrated in the image shared during the keynote:
“The biggest thing in the agent world is the Orchestrator – this is where the IP is.”
Impact of Agents
Over the last three months, over 400,000 agents have been built. Further, over 160,000 companies have adopted Copilot Studio. Why are companies using agents and Copilots? Smith highlighted several reasons, including:
- Time to Market
- ROI
- Enterprise Knowledge
- Connectors
- Security and Governance
To support this point, he shared a use case from a UK-based company, Pets at Home, which created a profit protection agent for its profit protection team to more efficiently compile cases for skilled human review. Since it’s rollout, the company has saved approximately $1 million.
In the example, Pets at Home’s agent is designed to handle customer support emails efficiently. When a customer sends an email about a product issue, the agent, using natural language instructions, automatically:
- Triggers into action upon receiving the email
- Reviews the customer’s product history and details
- Searches the knowledge base for relevant information
- Composes an empathetic, personalized email response
Then, the agent follows a step-by-step process and:
- Retrieves previous customer information
- Identifies common issues
- Finds helpful knowledge sources
- Sends a follow-up email
However, in a more complex scenario, such as a damaged product, the agent:
- Asks for clarifying information
- Assess damage using image analysis
- Escalates to a human agent if needed
- Follows up with a specific and tailored email referencing the specific customer and their pet
As a result, the agent creates an experience that is “hyper-vertical and personalized, but very easy, just based on those simple things.”
Autonomous Agents
Smith discussed the premise of autonomous agents which can handle variability and complexity at infinite scale. Autonomous agents are triggered by business events or actions (emails, schedules, database value changes), and eliminate the need for a “human triggers,” instead replaced by autonomous triggers.
Effectively, autonomous agents automate long-running processes, continuously improves, orchestrates other agents, and follows human guardrails.
“So really, the big difference here is the dynamic reasoning, tool selection, the ongoing processes, and we started making memories as its orchestrated across the process.”
Key Takeaways
As the discussion wrapped up, Smith shared a variety of key takeaways – the biggest is that everyone can be a builder and creator of agents. Other takeaways on Smith’s list include:
- Agents are just the new apps
- Agents are not RPA v2 – rather, agents are a reasoning tool that unlock business transformation
- Get hands-on with agents
- Garbage in = garbage out
- Technology is no longer the limiting factor
- Breakdown a use case into multiple agents
- Take advantage of Microsoft Partner Programs
- It’s easy to get started with Copilot Studio
Related Posts
