Oracle AI World - Tuesday - Day 2

it’s the first official day for AI World. Launch day, and already I have seen a really exciting announcement.

Multi-Cloud Universal Credits

Every cloud provider has a means of training a commitment for spend with lower costs. For example AWS has Reserved Instances and Savings plans, both of which you make a commitment for hourly usage for a discount.

AWS also has private pricing agreements that are negotiated under NDA with specific service teams.

The problem with these constructs is that you are pegged to an hourly spend. This means 2 things:

• You are limited to covering the minimum steady state

• If you go below the covered amount, you end up paying for underutilization

With OCI’s universal credits, you buy credits. They apply to any services you consume. And best of all, there is no hourly Burn down, so if your business is 5x during the day vs overnight, you can cover everything with universal credits. You consume more when you need them and less when you don’t.

Can this flexible model be used with Oracle services being offered by other cloud providers? Read about it here:

Multi-Cloud universal Credits

First Keynote:

AI is built into the database, into Fusion apps, Support, and other areas. And Oracle eats our own dog food, or drinks our own champagne. I can say from personal experience that the Epense submission process is much easier than others I have used.

Exelon: Leveraging Oracle and AI to help build and maintain the power grid. Removing costs but delivering on higher expectations. We will see more changes in power generation in the next 10 years than we have in the last 100. We can’t keep doing the same thing. We need to minimize risk. Every piece of equipment we add to the grid adds risk. We need to ensure that security is a top priority.

Communication is critical, setting expectations. This is data-driven and leverages AI. Customers want to have accurate estimations of when they will be back online.

If you think of AI from the mindset of eliminating jobs, you will miss the real opportunity.

Avis: # Pillars: Customers, Vehicles, People. Actively migrating data from legacy Oracle database to Oracle DB 23ai on OCI. Leveraging the ability for employees to explore data using natural language actively. (As a former business consultant for IT I remember government organizations having teams of data specifically to pull data together… and the business teams not trusting the data they were given.)

What matters isn’t the technology. Can AI accept the time to deliver business value? Are we gaining market share? Are our customer satisfaction scores increasing? Can we get to market faster? AI isn’t Artificial Intelligence, it’s Augmenting Individuals.

Marriott: How do you utilize AI to support employees and improve the customer experience? If we get this right, AI isn’t replacing the human touch; it’s bringing the human forward. The process to check someone into a hotel often interacts with multiple systems.

For so long, the message has been How do you do more with less. But instead, it should be How do you make what you do more impactful. “If your associate is having a better day, then your guest will have a better day.”

Recurring messaging

One of the recurring messages from all three speakers at the first keynote is that companies should not be looking to replace humans with AI, but rather augment human activities with AI. Can you leverage AI to make your customer interactions better and give your people more time to spend with customers rather than interacting with systems?

Instead of trying to leverage AI to eliminate “entry-level” positions, can you empower entry-level personnel to do different jobs? Providing people with more interesting and fulfilling work helps you to retain them as long-term employees.

Zero Trust with Agentic AI

For those who are at all confused about what Agentic AI means, it’s the use of Agents to embed AI functionality. Agents are kind of like the microservice of the AI landscape. Agents are created to be specialists at a single task, or a small collection of related tasks, rather than a wide-open chat system.

The use of Agents makes it easier to focus on specific tasks, facilitates testing, and, in theory, enhances security because agents should only require access to specific information and perform only the necessary actions. If an agent doesn’t create records, it shouldn’t have write access to anything except maybe the operational logs, where it can append.

The same security principles that apply to microservices also apply here. Think lest privilege. The idea that the agent is going to expose data is really no different to the old practice of putting data access web services out on the internet. How many times have people seen a web service that accepts an SQL String as a parameter, and then executes it blindly against the database, and returns a serialized dataset? Bonus points when the programmer used admin credentials for the web service.

The big differentiation here is that the web service was dumb. It did what you told it, but that was it. You had to possess a certain level of skill to locate the web service and then work to identify the available exploits. Often, an attacker would have to spend hours executing commands to try to determine the structure of the database. When you add an AI agent, you have a bit of code whose purpose in life is to try to be helpful. Does your agent assist the attacker in determining the database's structure?

• We need to secure access to the data

• We need to be very explicit about what the agent can and can not do

That’s a wrap for day 2… time to start day 3!