Reddappa Gowd Bandi

Interaction in the land of LLMs - Separating Hype from Reality TL-DR: Move thinking from creating scenarios where LLMs respond and act based on journey specific knowledge to where LLMs are provided agency to guided by access to knowledge and able to trigger deterministic actions that bring in a level of control and recovery. Chatbots of old were designed around requirements described as intent-based scenarios (or customer journeys). For example, ‘customer wants to block their card - the bot must be able to identify the intent, gather the information required, trigger appropriate back-end systems to carry out the task, and provide confirmation back to the user’. You would start out by drawing up a list of such journeys and implement them using deterministic flows. This is good as it gives good control but reduces the flexibility of the conversation especially around more complex multi-intent tasks. Instead of just replacing ML/diagloue engines with LLMs to understand input and generate responses we need to reimagine the role of the LLM. LLMs need to be given access to knowledge around process, controls, products, and brand. This can be through LLM-driven retrieval or prompt instructions. They also need to get access to actions. The actions should still remain deterministic. LLM is controlling the conversation based on knowledge it has. The scenarios/journeys of old are now the test cases instead of the requirements.

22

7 Comments

By modernizing workflow and workforce in tandem, AI can eliminate inefficiencies and strengthen trust. Learn how teams are unlocking productivity. https://bit.ly/40oP57E

13

Agentic AI Security: New Risks and Controls in the Databricks AI Security Framework (DASF v3.0) As organizations accelerate adoption of autonomous AI agents, security must evolve just as fast. The latest update to the Databricks AI Security Framework (DASF v3.0) introduces a significant milestone—Agentic AI as the 13th system component. What’s new? • 35 new technical security risks tailored for Agentic AI • 6 new mitigation controls to strengthen secure deployment • Focus on emerging risks in agent memory, planning, and tool usage • Coverage of Model Context Protocol (MCP)—a key standard enabling agents to interact with enterprise tools Why this matters Agentic AI introduces a fundamentally different risk model compared to traditional AI systems. With autonomy comes increased exposure—especially in how agents reason, access tools, and retain context over time. The DASF Agentic AI Extension whitepaper and updated compendium are now available—essential resources for security leaders, architects, and practitioners navigating this new frontier. #AgenticAI #AISecurity #Databricks #CyberSecurity #AIGovernance #CloudSecurity #LLMSecurity #GenAI #SecurityFrameworks #ZeroTrus::

11

1 Comment

Below the line, AI compresses your cost. Above it, AI creates the MOAT. Powerful framing from Zinniv President Nilesh Thakkar at #ZINNOV CONFLUENCE Day 2. Everyone is using AI below the line. For efficiency. For cost. But only 13% of CEOs have moved ABOVE the line. That's where real advantage lies. Above the line is messy. It's autonomy. It's interoperability. It's agents making decisions across systems. And you can't live above the line without resilience. Are you building for cost or for moat? #ZinnovConfluence #AI #AgenticAI #Leadership Quinte Financial Technologies

21

1 Comment