Trustworthy AI Agents

Tanmai Gopal (@tanmaigo, CEO/Founder @HasuraQL) discusses the importance of reliability and trustworthiness for both generative and agentic AI. We discuss the pitfalls in existing data pipelines and how to enhance the results. SHOW: 935 SHOW TRANSCRIPT: The Cloudcast #935 Transcript SHOW VIDEO: https://youtube.com/@TheCloudcastNET  CLOUD NEWS OF THE WEEK:  http://bit.ly/cloudcast-cnotw NEW TO CLOUD? CHECK OUT OUR OTHER PODCAST:  "CLOUDCAST BASICS"  SPONSORS:[VASION] Vasion Print eliminates the need for print servers by enabling secure, cloud-based printing from any device, anywhere. Get a custom demo to see the difference for yourself.[US CLOUD] Cut Enterprise IT Support Costs by 30-50% with US Cloud SHOW NOTES:Hasura websiteHasura GitHub Topic 1 - Welcome to the show, Tanmai. Give everyone a quick introduction. Topic 2 - Our topic today is Reliable and Trustworthy AI Agents. First off, what’s the problem we’re solving for here (define reliability and trustworthiness)? Are we solving for hallucinations? Reliability? Connecting private and Enterprise data to models with fine-tuning or RAG? Topic 3 - How is reliability or trustworthiness measured? I would imagine this isn’t black and white, but maybe a bit more subjective? Topic 4 - How do Agentic and GenAI differ, if at all, with this model? I would think that since Gen AI lends itself more to the creative side and Agent AI is very deterministic, the approaches to solving the problem might be different. Thoughts? Topic 5 - Let’s talk about data pipelines. Today, many organizations take an off-the-shelf frontier or foundational model and then apply a RAG pipeline to it for customization. Sometimes fine-tuning is involved, but in my experience, this is the exception rather than the rule. What is wrong with that architecture today? How is this less reliable? Topic 6 - Let’s talk about Hasura and PromptQL. As I understand it, you are decoupling query planning from execution, thereby creating a more deterministic AI workflow. Now… that’s a mouthful. Can you break down what this means and explain how the architecture differs? FEEDBACK?Email: show at the cloudcast dot netBluesky: @cloudcastpod.bsky.socialTwitter/X: @cloudcastpodInstagram: @cloudcastpodTikTok: @cloudcastpod