Exclusive: Oracle's marketplace passes 400 Fusion AI agents

Oracle has significantly expanded its Fusion AI agent platform, now boasting more than 400 agents across its enterprise applications. The company has also opened its Agent Studio marketplace to partners such as Deloitte, Accenture, Infosys and TCS, enabling them to develop and distribute their own AI-driven agents.

"We've gone up to probably past 400," said Kaushal Kurapati, Group Vice President of Product Management for Fusion AI at Oracle. "We've been fortunate to have about 20-plus launch partners, and they've created more than 100 agent templates available in the Agent Studio."

The platform serves as a single AI agent layer across Oracle's suite of enterprise software - spanning ERP, HCM, supply chain and customer experience applications - and allows users to automate and streamline workflows.

Building and customising

Kurapati said that Oracle's design goal is to make agent creation accessible to a wide range of enterprise users, though for now, those building them are typically IT professionals, administrators, solution engineers and product managers.

"Our aim is ultimately so that anybody in the organisation should be able to create a knowledge worker of any kind," said Kurapati. "For now, it's mainly IT folks, fusion admins, and solution engineers."

Users can either build agents from scratch or customise existing templates, making deployment easier. "You could create something from scratch, but even easier is taking a template and customising it for your own purpose," said Kurapati.

Secure by design

Security and compliance are built into the platform, according to Kurapati. Agents inherit Oracle Fusion's existing security architecture and policies, ensuring that user permissions and data constraints remain intact.

"Agents are built natively into the Fusion platform, which means they inherit all the goodness of the Fusion security architecture," he said. "We enable auditing capabilities for agents, so audit trails are available."

Each agent must be assigned to a specific user role, determining what data it can access. "Associating it with a user role is a mandatory field," said Kurapati. "You cannot proceed without that."

Range of capabilities

Kurapati described a spectrum of agent capabilities, from simple task automation to complex orchestration. Some agents handle tasks such as summarising meeting minutes or drafting emails, while others manage more advanced workflows involving multiple subordinate agents.

"It's across the spectrum," said Kurapati. "Some of the simpler ones are just executing a simple completion prompt… but the more sophisticated ones then execute workflows or coordinate between worker agents."

Oracle is also developing "workflow agents" that perform sequential, deterministic tasks, contrasting with goal-based agents that respond dynamically to user intent.

Practical applications

Oracle has embedded generative AI agents directly into its enterprise applications. For example, employees can use AI assistance when writing annual performance goals.

"If you enter a fragment, it can generate a goal for you based on company policies, last year's goals and your role," said Kurapati. "You're not starting from a blank sheet of paper anymore."

The agents also integrate with collaboration platforms such as Microsoft Teams and Slack, enabling conversational workflows. Kurapati said internal Oracle staff already use these integrations for IT support. "You can send a message saying your laptop is rebooting," he said. "The agent responds, troubleshoots, and if needed, creates a helpdesk ticket or brings in a technician."

He added that product managers are also using internal tools such as a "PRD agent" to generate initial drafts of product requirement documents within seconds.

Competitive differentiation

Kurapati argued that Oracle's integrated technology stack provides a distinct advantage over rivals. "We are not only a business applications provider, but also a hyperscaler," he said. "The scale of models we are operating gives us a unique advantage of being an end-to-end provider."

He pointed to Oracle's longstanding reputation in database management and its ability to run large AI models on its own Oracle Cloud Infrastructure (OCI). "We can run the best models in the world on our compute infrastructure, which gives us a cost advantage," he said. "If you're using a model on OCI, we are not charging additional tokens - all our competitors are."

Kurapati said Oracle's customers also benefit from data proximity and performance advantages. "The data is already there. We don't need to go to some other data lake," he said. "We know who's eligible to see that data, and it's sitting right there in a highly performant database."

Pricing and adoption

Oracle is including many of its standard agents within existing Fusion subscriptions to encourage adoption. "Most of it is included in your Fusion subscription," said Kurapati. "We are focused on adoption of AI and agents - we don't want to put that up as a first barrier."

Custom-built agents, however, may incur additional charges. "If you create a custom agent from scratch, then we will charge you," said Kurapati. "But a lot of customisations we won't even charge for."

Partner ecosystem

Oracle's marketplace currently consists of around 70% Oracle-created agents and 30% partner-built templates. Kurapati expects this balance to shift rapidly as more partners join.

"It's probably like 70–30 right now," said Kurapati. "We actually rejected some agents because the quality wasn't meeting our bar… I fully expect there'll be an explosion of that whole category."

System integrators such as Deloitte, Infosys, TCS and Accenture are among the key contributors, with some also building their own agent platforms. Kurapati said this was expected. "They are looking at connecting custom implementations with our agents," he said. "The field is still very Greenfield - it's early days to predict how this will shake out."

Integration and interoperability

While Oracle's agents primarily operate within its ecosystem, interoperability remains a design priority. "We work in a very interoperable manner," said Kurapati. "You can connect to data sitting in Databricks, Snowflake, Salesforce or ServiceNow."

He added that Oracle supports modern integration standards such as Model Context Protocol (MCP), allowing agents to access and act on data stored in platforms like Box, OneDrive and Google Drive.

Kurapati said he expects multiple enterprise marketplaces to coexist. "In the enterprise context, there will be multiple," he said. "Ultimately, the ease of creation of agents and the utility of the marketplace will prove itself out."

The company plans to host the latest open-source models within Oracle Cloud. "We'll host that in November in OCI," said Kurapati. "We'll have the best models under the sun in OCI - and that's what is exciting for us."