Today at AI World London, Oracle announced twenty-two new Fusion Agentic Applications and significantly expanded AI Agent Studio. Steve Miranda, Executive Vice President of Applications Development, said that the goal is to move enterprise software beyond "passive systems of record" toward applications that can reason, decide, and act in pursuit of defined business objectives. In its own words, Oracle is shifting Fusion from a system of record to a system of outcomes.
For buyers, this is well beyond the capabilities of copilots and generative AI assistants. Those tools may have provided a marginal uplift in productivity (and even that is debatable), but this suite of agentic applications essentially changes how work gets done.
The speed at which this has moved is worth acknowledging. Eighteen months ago, Oracle was embedding large language models to generate text across Fusion - item descriptions, report summaries, etc. Useful, but limited and incremental. Then came individual agents, handling discrete steps in a process. Describing those agents, Miranda said:
Then we moved to agents. We now have hundreds of agents across the applications. The way to think about it is that each agent is a small piece of AI that handles one step in a process. Take a recruiting process - you shortlist candidates, schedule interviews, do background checks, do onboarding. Each of those steps has an agent.
Today's announcement is the step beyond that. Not individual agents handling individual tasks, but coordinated teams of agents working toward a business objective. And it's available now across the full Fusion portfolio - ERP, Supply Chain and Manufacturing, HCM, and Customer Experience. On the scale of the change this could bring, Oracle CEO Mike Sicilia compared it to aviation's shift to jet engines. On stage during his keynote, he said:
I like to think about aviation's shift from propellers to jet engines as a comparison. When that happened, airlines didn't just change their routes - they were still moving people and goods - but what changed was capability. Jet engines expanded what was possible: faster routes, longer distances, entirely new markets, because the underlying technology could finally keep up with rising expectations. And that's exactly what AI is doing, and will do, for the enterprise.
What's actually being announced
The twenty-two new Fusion Agentic Applications cover use cases that include:: a Collectors Workspace aimed at reducing days sales outstanding and accelerating cash collection; a Workforce Operations app targeting payroll errors and scheduling delays; a Design-to-Source Workspace for supply chain sourcing; a Cross-Sell Program Workspace for sales teams chasing expansion revenue.
This means that rather than having users operating individual screens, entering data into predefined fields, telling the system what to record - the approach to ERP that’s barely changed in thirty years - these applications accept a business objective and work toward it. Miranda used a good analogy:
Think back a couple of years ago…if you were going on vacation, you'd go into Google, search for hotels, search for available flights, search for things to do, restaurants, and so on. You'd get some answers, and then manually book airline tickets, the hotel, restaurants, tours. With an agentic framework, instead of asking what's a good hotel or what are the tourist attractions, you just ask: what should I do in London? And it gives you a fully mapped-out itinerary.
A simple example, but that's the shift Miranda and Oracle are laying out for enterprise. From asking questions and doing the typing yourself, to giving objectives and having the system do the work. In a pre-brief ahead of the event in London, Natalia Rachelson, who leads Fusion Applications product management, said:
The enterprise systems of the last 30 to 40 years recorded what happened, reported on what happened, and we all made decisions and moved business forward outside of the system.
With agentic applications, it's very much a system that can make decisions autonomously, and act and execute."
The governance piece
I pushed hard on governance with the Oracle team this week, given that this is the primary concern amongst CIOs and digital leaders in the diginomica network. In January, we ran a micro-pulse survey with 124 CIOs asking about their AI implementation experiences in 2025.
Governance and risk management came out as the highest urgency and highest impact theme of the entire survey. One CIO described their experience as "overall quite challenging and at times frustrating - much of this stems from the organisation still defining AI governance and operating models, which has resulted in complex, unclear processes that are difficult to navigate." Another flagged the agent-specific tension, stating that they “worry about over-dependence on one LLM and have raised the company risk level as we become more dependent on agents”. These aren’t challenges specific to Oracle, but they provide context for the selling environment Oracle is stepping into.
Oracle's response is that Fusion Agentic Applications inherit the entire governance model of Fusion itself. Rachelson outlined why this matters architecturally:
The governance comes from the broader Fusion system of record. These applications inherit the user's role-based access controls, the security associated with that user, the approval hierarchy, all the rules and policies stored in the system. So if I can only see a subset of HR data, my agentic application can also only see that subset.
The common critique of bolted-on AI in enterprise contexts is that it creates a new governance layer sitting outside the core system - new risks, new audit requirements, new conversations with regulators and workers' councils. Oracle's argument is that because its agents run inside Fusion, with native access to data, policies, and approval structures, that problem is largely avoided.
I also asked specifically about audit trails - what does traceability actually look like if something goes wrong? Rachelson's answer was more substantive than I expected:
Every single step is recorded - not just cases where an agent gets stuck and calls for human intervention, but every decision, why it was made, and how. Customers get a complete and full audit trail. They can test against it and review it, and that's often how they get comfortable with the system - seeing traceability and auditability in action.
I also asked about automation and the level of autonomy these agents have. Oracle is taking a measured position - customers set their preferred level on a spectrum, most will start somewhere in the middle and increase over time (towards full autonomy) as confidence grows.
The question nobody can answer yet
Another point that I put to Miranda was around how organizations are successfully - if at all - changing operating models towards agentic AI architectures. The genuinely disruptive thing about these agents is that they don't care about organizational boundaries. They cross departments, functions, and reporting lines to accomplish a goal. Which is precisely why they're powerful - and why they create a real problem for the humans who have to reorganize around them.
This is what the CIOs in our network raise consistently as their primary concern. The technology is becoming easier to understand, but changing operating models to really reap the rewards feels like a complex and challenging job. I asked Miranda: are you seeing examples yet of how to effectively bring that organisational change about? He said:
I can't point to concrete examples of success just yet, but I can say there's enormous pressure across the ecosystem. Think about what I described in terms of business process flows: there are going to be certain tasks that AI does extraordinarily well.
If today three out of five of your tasks fall into that category, you probably won't be doing those much longer. But you'll still have the other two. What does that mean? Your peer sitting next to you might have ten tasks and AI only handles three of theirs.
He went further on the cross-functional dimension:
We see that changing fundamentally too. The agentic applications we're building are all role-based, at a higher role level, and what happens underneath will change fundamentally. People are trying to figure out what's being automated and what they need to do as a result. It is reforming the structure of teams and roles.
That's the work that sits ahead of every enterprise considering this. And it's the work that no vendor can do for you.
A few other things worth noting
A couple of other points worth highlighting, even if they might be obvious to some, are around how Oracle tests its agents across multiple models and picks the best for each task, with no commitment to any single provider. The infrastructure position is that because the major LLMs are available on OCI, customers will always get the benefit of whatever the best option turns out to be.
On pricing, agentic AI is included within existing Fusion subscriptions, with a consumption model - measured in action units - for usage beyond the base allotment. Miranda said that user-based pricing will eventually give way to something transaction or company-size based:
There will come a day when our pricing model for our base subscription shifts from user-based to some sort of transaction-based or company-size-based metric. Nobody is in business to run ERP - they have to run ERP. The more we can save them on the ERP side, the more they can invest in what they're actually in business to do.
My take
The system of record to system of outcomes framing is the right way to think about what Oracle is attempting here. The architectural argument - agents living inside Fusion, inheriting its governance model, meaning enterprises can move faster without creating new risk - is coherent. If it holds up in production, it will be impressive.
However, I keep coming back to the enterprise reality of it all. This is so far ahead of where so many organizations are, as it really does require buyers to think through how they operate (which could be quite drastically different). Teams, departmental lines, job roles - these agentic architectures impact them all. That’s not totally Oracle’s job to have the answer. Everyone is thinking through the use cases and implications, but I really do urge CIOs (or anyone in charge of this) to recognize this isn’t a straightforward tech rollout. This is not like for like. We are all still figuring out the playbook.