Guardrails, not guard dogs: Shipping reliable AI actions

Описание: Intercom’s Chief AI Officer Fergal Reid sits down with Principal Machine Learning Engineer Pratik Bothra to explore what it really takes to make AI actions reliable and safe. They dive into the pitfalls of moving from demo to deployment, why deterministic checks trump LLM watchdogs, and why most resolutions still come from strong informational answers. It’s a candid look at the challenges and trade-offs in building agentic workflows that work in production.

Chapters
00:00 Introduction to AI and Engineering Background
02:58 Transitioning to AI Products and Co-Pilot
06:12 Understanding Tasks and API Integration
09:05 Security and Reliability in AI Systems
11:55 Complexity in Task Management and Business Logic
15:00 Setting Up Tasks and User Authentication
17:53 The Role of LLMs in Customer Service
20:46 Challenges in Task Product Development
24:06 The Importance of Prompt Engineering
27:12 Reliability and Performance Metrics in AI Tasks
34:33 Navigating the Challenges of UX and AI
39:05 Building Effective AI Products
44:41 Balancing Control and Complexity in AI
52:05 The Future of AI Product Development
01:01:14 Understanding Informational vs. Personalized Responses

Follow the people
  / fergalreid  
  / pratik-bothra-bba03074  

https://fin.ai/research/
Insights and blogs from the AI Group building Fin at Intercom

Say hi
  / fin  
https://x.com/Fin_ai

https://www.fin.ai

#AI, #CustomerSupport, #Automation, #MachineLearning, #Intercom, #FinAI, #AIAgents, #LLM, #CustomerExperience, #APIs, #TechPodcast