Experts of Experience: The Art of Conversation Design for AI Agents

Podcast: Experts of Experience
Host: Lauren Wood
Guest: Irina Gutman, Global Leader of AI Professional Services at Salesforce
Presented by: Salesforce Customer Success
Release Date: April 9, 2025
Duration: 53 minutes and 50 seconds

Introduction

In this enlightening episode of Experts of Experience, host Lauren Wood engages with Irina Gutman, the Global Leader of AI Professional Services at Salesforce, to explore the transformative realm of Agentic AI. Building upon their previous discussion six months prior, the duo delves deep into how businesses are implementing Agentic AI, the challenges and opportunities it presents, and the future trajectory of AI in customer experience (CX).

Understanding the Different Types of AI

Predictive AI, Generative AI, and Agentic AI

At the outset, Irina Gutman provides clear definitions to distinguish between Predictive AI, Generative AI, and the focus of today's conversation, Agentic AI.

• Predictive AI (00:16 - 03:22):
  "Predictive AI focuses on making predictions based on data, based on rules and based on the structure that we provide it with."
  Example: Salesforce's lead qualification system categorizes and prioritizes sales leads based on predefined rules and data.

• Generative AI (03:22 - 05:18):
  "Generative AI... generates dynamic content based on the request or what we call as prompt or basically what you put in ChatGPT."
  Example: ChatGPT, which can generate text responses based on user prompts without being confined to strict predefined rules.

• Agentic AI (05:18 - 08:27):
  "Agentic technology... elevates it to the next level. It makes a decision how to act based on those rules, uses data to reason on what action to take, and converses with us in natural language."

  Example: Autonomous driverless cars like Waymo, which not only follow traffic rules but also make intuitive decisions based on real-time road conditions.

Agentic AI vs. Traditional Chatbots

Lauren Wood draws a clear comparison between Agentic AI and traditional chatbots to highlight the advanced capabilities of the former.

• Chatbots (08:47 - 11:17):
  "Chatbot follows a very prescribed process flow... it cannot deviate, it cannot change a course of action or interaction."
  Limitations: Struggle with understanding natural language variations, slang, or handling complex queries without predefined scripts.

• AI Agents (08:47 - 11:17):
  "Agent... should be able to interpret the slang... infer based on the information provided... pick up on the tone and change the way [it] responds."
  Advantages: Enhanced flexibility, ability to understand and respond to nuanced language, and adaptability to different customer tones and contexts.

Notable Quote:
Lauren Wood (08:01):
"The thing about agentic AI that is just so incredible is that it can think for itself. It's kind of creepy, but here we are and there's so much opportunity."

Implementing Agentic AI: Key Considerations

1. Starting Small with Low-Hanging Fruit

Irina emphasizes the importance of beginning with simple, repeatable tasks when implementing Agentic AI.

Quote:
Irina Gutman (18:59):
"Actually, you want the most boring, the most repeatable, the most low hanging fruit agent to start with."

Example: Implementing agents to handle frequently asked questions or standard service inquiries before tackling more complex interactions.

2. Organizational Readiness and Operating Models

Successful implementation requires a strong partnership between business and IT.

• Readiness Assessment (23:29):
  Evaluating data availability, identifying use cases, and ensuring organizational support.

• New Operating Models (25:04):
  Establishing roles such as Agent Owners and Agent Monitors, and potentially creating Centers of Excellence (COE) to oversee AI initiatives.

Quote:
Irina Gutman (27:00):
"The best way to handle emerging technology is to have some kind of a center of excellence around it."

3. Foundational Technology

Robust data management and integration capabilities are crucial.

• Salesforce Components (23:35 - 24:25):
  • Data Cloud: Facilitates data syncing across various sources.
  • MuleSoft: Acts as the integration layer connecting structured and unstructured data from within and outside Salesforce.

Insight:
"Combination of Data Cloud and MuleSoft unlocks any data needs, any communication needs."

4. Upskilling and New Roles

Introducing Agentic AI necessitates new skill sets and roles within the organization.

• Key Roles (37:14 - 41:02):
  • Prompt Engineers: Crafting precise prompts for AI agents.
  • Conversation Designers: Designing the interaction flow between agents and humans.
  • Content Creators: Structuring knowledge base articles for AI consumption.

Quote:
Irina Gutman (37:43):
"Prompt engineering is absolutely foundational skill set to have."

5. Incremental Implementation and Roadmapping

Adopt an iterative approach, allowing organizations to see tangible outcomes without overwhelming transformations.

Quote:
Irina Gutman (26:25):
"You don't want it to be a year-long transformation program where you don't see any outcome until we implement all complex agents in the world."

Responsible AI Practices

Addressing the ethical and operational risks associated with Agentic AI is paramount.

• Defining Guardrails (48:09):
  Establishing clear boundaries on what the AI agent can and cannot do, ensuring compliance and preventing misuse.

• Stress Testing for Bias and Toxicity (50:42):
  Rigorous testing to ensure agents do not produce biased or toxic responses, maintaining brand integrity and avoiding legal repercussions.

Quote:
Irina Gutman (48:09):
"Nobody wants to be a headline for the wrong reason or have a lawsuit. So that type of testing is absolutely critical."

The Future of Agentic AI: Multi-Agent Collaboration

As organizations mature in their AI journey, the next evolution involves multi-agent systems.

• Internal Assistive Agents:
  Agents that operate within organizations to assist employees.

• Customer-Facing Agents:
  Agents interacting directly with customers to enhance CX.

• Agent-to-Agent Collaboration (42:05):
  Enables different agents, possibly from different companies, to communicate and collaborate, expanding the scope and efficiency of AI interactions.

Quote:
Irina Gutman (42:09):
"When we talk about agentic technology, the natural progression for a lot of companies is internal assistive agent... agent to agent collaboration."

Real-World Examples

Saks Fifth Avenue's Agent "Sophie"

Saks Fifth Avenue has implemented an AI agent named Sophie to handle customer service queries such as order placement, status updates, and issue resolutions. This demonstrates the practical application and benefits of Agentic AI in retail.

Quote:
Irina Gutman (44:14):
"Saks actually has an agent like that. Her name is Sophie and she handles this type of questions for Saks."

Starbucks’ AI-Enhanced Customer Experience

Irina shares her personal experience with Starbucks' AI-driven app, which remembers favorite stores, orders, and even allows for order prioritization in busy scenarios. This seamless integration of AI improves customer convenience and operational efficiency.

Quote:
Irina Gutman (51:30):
"It has just enough key elements to make my life easy and also from a customer service perspective... human plus technology working together provides amazing experience."

Conclusion

The episode concludes with a reaffirmation of the immense opportunities Agentic AI offers, balanced by the necessary caution and responsibility required in its deployment. Irina Gutman expresses excitement about partnering with organizations to harness Agentic AI's potential responsibly, ensuring enhanced customer experiences and operational efficiencies.

Final Thought:
Lauren Wood emphasizes the importance of embracing change and proactive adoption to stay competitive in the rapidly evolving AI landscape.

Final Quote:
Lauren Wood (53:30):
"We have to have those conversations. What are the risks at play? Just put them all out on the table... let's jump on the bandwagon."

Key Takeaways:

1. Agentic AI represents a significant advancement over traditional and generative AI by enabling autonomous decision-making and natural language interactions.
2. Successful implementation requires starting with simple, repeatable tasks, ensuring organizational readiness, and building robust technological foundations.
3. Responsible AI practices, including defining guardrails and stress testing, are essential to mitigate risks.
4. The future of AI lies in multi-agent collaborations, enhancing both internal operations and customer-facing interactions.
5. Real-world applications, such as those by Saks and Starbucks, demonstrate the tangible benefits and transformative potential of Agentic AI in enhancing customer experiences.

For those looking to explore Agentic AI further or implement it within their organizations, partnering with technology leaders like Salesforce and investing in upskilling teams are critical steps toward harnessing this groundbreaking technology responsibly and effectively.