On Repeatability and the Application of AI Agents

Many processes require repeatability rather than the inherent probabilistic nature of generative AI — generated through ChatGPT

Unless you have been living under a rock for the past couple of years, you will have seen the buzz around generative AI and how it has come to dominate the tech news cycle. The most recent update and drive is towards so-called “Agentic AI”, where AI agents take in data from various sources, utilising large language models to reason and understand the inputs and tasks, and run other actions based upon this understanding. For a fantastic detailed overview, I recommend this Nvidia blog.

Salesforce have released Agentforce as their offering in this space, with Agentforce for Service being live on the Salesforce help site, and Agentforce for Developers available to help developers (and non-developers) generate and tweak code solutions.

I have spoken on a few podcasts and at some events on agents and Agentforce in particular at a high-level, and I wanted to write a deeper dive into where my thinking and reasoning comes from. What I am excited about and what I am nervous about. And most importantly, why. I have worked with AI and built a few small AI systems over the years, so I have some understanding of the technology and mathematics underlying these systems, which also feeds into this. So here we go.

The Probabilistic Nature of AI —Stochastic Parrots