Super Data Science: ML & AI Podcast with Jon Krohn · Jon Krohn

958: Without Trusted Context, Agents are Stupid (featuring Salesforce’s Rahul Auradkar)

·24 min·1 clip

If AI models are so intelligent, why do they keep doing such stupid things? For my guest today, the answer is simple

AI can sound sharp and still make a mess. Rahul frames the problem as the space between model intelligence and the company context needed to use it responsibly. Jon starts with the plain question: if models are this capable, why do they still fail? The answer is often weak data foundations, not reasoning alone. Salesforce's unified data engine sits in the middle of the conversation. Data 360 is treated as one piece of a wider system for getting enterprise data into usable shape. Informatica gets a lot of time because its metadata, catalog, data quality, governance lineage, and ETL tools give agents cleaner ground to stand on. MuleSoft adds the connectivity layer, handling iPaaS patterns and helping data move between systems. Rahul keeps returning to trusted context. Ordinary context is not enough inside a business. The context has to be governed, consent aware, traceable, and ready for real workflows. Jon makes the point easier to hear: recent AI apps show that context can be the difference between an agent that works and one that breaks. Rahul pushes on the silo problem too. Companies may be data rich and still context poor when information is split across systems. Disconnected prompts lead to disconnected answers. The practical thread is Salesforce's bet that cataloging, cleansing, integration, governance, and real-time signals make agents less brittle. This is a data platform conversation, but the real question is reliability: can an agent understand the business setting well enough before it takes action?

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Super Data Science: ML & AI Podcast with Jon Krohn