L
How We Built It: Decagon - Fireside Chat with Harrison Chase & Bihan Jiang
@ LangChain
02/20/25
Takeaways
BJ
Bihan Jiang
02/20/25
@ LangChain
I'm Bihan, and I lead product at Decagon, a Series B startup building AI agents for customer service and support use cases.
BJ
Bihan Jiang
02/20/25
@ LangChain
Decagon builds AI agents that not only respond to customer inquiries but also create a comprehensive support ecosystem around them.
BJ
Bihan Jiang
02/20/25
@ LangChain
We're creating a better, more human-like customer support experience with AI agents to help users get the support they need efficiently.
BJ
Bihan Jiang
02/20/25
@ LangChain
Customer support is an interesting space for AI agents because it involves a lot of unstructured information that can be effectively utilized by AI models.
BJ
Bihan Jiang
02/20/25
@ LangChain
AI models have become proficient at processing unstructured data, allowing them to generate helpful responses and take actions in customer support scenarios.
BJ
Bihan Jiang
02/20/25
@ LangChain
The core AI agent acts as the brain of the system, handling logic, knowledge, and workflows, similar to how human agents are trained.
BJ
Bihan Jiang
02/20/25
@ LangChain
Different modalities like chat, email, and voice require different user expectations and response times, but the core AI agent logic can be applied across all channels.
BJ
Bihan Jiang
02/20/25
@ LangChain
Standard operating procedures for human agents can often be adapted for AI agents, but the instructions need to be tailored for how AI executes tasks.
BJ
Bihan Jiang
02/20/25
@ LangChain
We implement guardrails around sensitive actions like issuing refunds, allowing customers to define specific criteria for when these actions can be executed.
BJ
Bihan Jiang
02/20/25
@ LangChain
We've built in security measures to prevent prompt injections and ensure that AI agents behave as expected, especially in regulated environments.
L
LangChain Cast
02/20/25
@ LangChain
Routing is essential in customer support, especially in industries like healthcare and finance, where certain cases must always be directed to a human agent due to incomplete knowledge bases or workflows.
L
LangChain Cast
02/20/25
@ LangChain
Our routing product allows customers to dynamically define criteria for directing conversations to human agents, while our agent assist product acts as a co-pilot for human support agents, enhancing their efficiency by providing AI-driven insights.
L
LangChain Cast
02/20/25
@ LangChain
Some companies prefer to start with human agents handling customer interactions before gradually integrating AI, allowing them to build confidence in the AI's capabilities over time.
L
LangChain Cast
02/20/25
@ LangChain
We can customize AI agents to align with a company's brand guidelines, tone, and structure, ensuring that the AI represents the brand accurately and maintains its reputation during customer interactions.
L
LangChain Cast
02/20/25
@ LangChain
During the deployment process, we work closely with customers to define how their AI agent should respond, including tone and response structure, to meet their specific needs and preferences.
L
LangChain Cast
02/20/25
@ LangChain
The admin dashboard is crucial for customers to build and configure their AI agents, allowing them to test and roll out the agent safely while monitoring performance and user interactions.
L
LangChain Cast
02/20/25
@ LangChain
Key metrics for evaluating AI agents include deflection rate, which measures how many conversations the AI can handle without human escalation, and customer satisfaction scores, which reflect user experience.
L
LangChain Cast
02/20/25
@ LangChain
Testing AI agents involves pre-deployment testing with curated conversation sets to ensure expected responses, followed by ongoing evaluation to adapt to changing user interactions and questions.
L
LangChain Cast
02/20/25
@ LangChain
Quality management is vital for AI agents, involving metrics analysis and deep feedback on user interactions to continuously improve performance and address any issues that arise during deployment.
L
LangChain Cast
02/20/25
@ LangChain
We are hiring across various roles, including product design, engineering, and sales, as we continue to build and scale our AI agent solutions, inviting talented individuals to join our team.