Agentforce

Chapter 18How Do You Make Sure AI Is Governed?

It can seem at times that data is out of control – and in a sense, it is. We cannot stop the flow of bits into our world, nor should we want to. Not only do they have the potential to do nice-to-have things like make shopping more fun, but they can also save lives through cyber-medicine and cyber-pharma, analyze and (one hopes) address climate change, and improve learning.

According to the analyst firm IDC, almost 300 zettabytes of data will be generated by 2027 – and it's up to us to store and manage this deluge.¹ It's needed for the modern enterprise, not least because data fires up AI: it's the lithium-ion in the renewable engine of the agentic AI future.

AI actually requires more data than other activities for a number of reasons. AI models learn from data, so the more they have, the better they perform. Their learning is becoming highly nuanced, which gives it extra power but requires vast amounts of data. GenAI models in particular often have billions of parameters, which require significant data for performance and accuracy. They improve over time, requiring more data to do so. And their strength in handling variability (scenarios, questions, dialects) requires there be a lot of variability in the data, which means – as you've guessed – more data.²

Data and AI come together in the enterprise in one of two divergent forms: chaos or governance. In the case of chaos, an enterprise is not quite sure how their data is collected, ...