Oracle Announces 22 Fusion Agentic Applications and New Agent Tools for Its Database and Cloud

Title: Oracle Announces 22 Fusion Agentic Applications and New Agent Tools for Its Database and Cloud

Date: 26 March 2026

Type: Blog

Author: SAASiQ (contact@saasiq.ai)

Word count: 1486 words

Reading time: 6 min

Published: 26-03-2026

Tags: #Oracle #AI #OCI #Database #EnterpriseAI

Oracle used its AI World Tour event in London on 24 March to announce 22 Fusion Agentic Applications, new features in AI Agent Studio, a set of agent tools for Oracle AI Database and a developer service called OCI Enterprise AI. It also renamed its database services inside the other large clouds, so Oracle Database@AWS is now Oracle AI Database@AWS. The announcements came two weeks after third-quarter results in which Oracle's cloud infrastructure revenue rose 84 per cent and its contracted order book reached $553 billion.

The Fusion agentic applications

Oracle describes a Fusion Agentic Application as a workspace in which a coordinated team of specialised AI agents works to deliver one business outcome, using the data, workflows, policies, approval hierarchies and permissions already held in Fusion Cloud Applications. It said 22 were available now, across finance, HR, supply chain and customer experience, and it gave four examples.

The Collectors Workspace is aimed at collecting cash faster and lowering days sales outstanding. The Workforce Operations application is meant to cut manual data gathering, speed up scheduling approvals and reduce payroll issues. In supply chain, the Design-to-Source Workspace targets product cost, cycle time and compliance risk, and in sales the Cross-Sell Program Workspace looks for expansion revenue and lower acquisition costs. The announcement named no customers and gave no measured results.

Oracle said the applications run inside Fusion's existing security framework, with role-based access, approval frameworks and end-to-end traceability. An agent acting for a user reaches only the data and functions that the user's roles already allow, so the role design a customer set up during implementation now governs its agents as well. Steve Miranda, who leads applications development, called the launch a huge step forward for the industry.

What changed in AI Agent Studio

AI Agent Studio, launched in March 2025, is the tool Fusion customers and partners use to build agents of their own. The 24 March update added an Agentic Applications Builder, which assembles applications from agents in natural language, along with workflow orchestration with points for human oversight, content intelligence that combines documents with transaction data, contextual memory between interactions, multimodal models, and monitoring with a prompt playground. An Agent ROI dashboard reports the time and cost saved by each agent.

Oracle said the studio is still available at no additional cost, and that more than 63,000 partner staff are now certified in it. On the third-quarter earnings call on 10 March, co-chief executive Mike Sicilia said Oracle had delivered well over 1,000 agents inside its business and industry applications, and that the embedded agents also come at no additional cost.

Agent tools in the database

Oracle AI Database 26ai was announced at Oracle AI World in Las Vegas on 14 October 2025 as the successor to Database 23ai, delivered through the October 2025 release update rather than as an upgrade, and Enterprise Edition for on-premises Linux followed in the January 2026 release update. It carries the AI Vector Search that Oracle introduced in 23ai, so similarity search and retrieval for AI models run in the same database as the transactional data.

The 24 March database announcement added nine items. The Autonomous AI Database MCP Server lets outside agents and MCP clients query the database without custom integration code, alongside the MCP server Oracle already offers in its SQLcl command-line tool. Deep Data Security applies each end user's access rules inside the database, by persona and by function, so an agent acting for a user sees only what that user is allowed to see, and Oracle presents it as a protection against prompt injection. The Private Agent Factory is a no-code agent builder for business analysts, shipped as a container that runs in a public cloud or on a customer's own hardware, with pre-built Database Knowledge, Structured Data Analysis and Deep Data Research agents.

The Unified Memory Core stores agent context in one engine across vector, JSON, graph, relational, text, spatial and columnar data. The Private AI Services Container runs private model instances, including in air-gapped sites, so that tasks such as generating embeddings do not send data to a third party. Trusted Answer Search uses vector search to match a question to a report that already exists, returning a tested answer instead of a generated one. Vectors on Ice adds vector data to Apache Iceberg tables, Exadata customers get accelerated AI search, and a new Autonomous AI Vector Database is in limited availability on Oracle's free and developer tiers.

Most of these run wherever Oracle's database runs: in OCI, in the other clouds and on-premises. Juan Loaiza, executive vice president for database technologies, said the next wave of enterprise AI would be defined by whether customers can use it safely in business-critical production systems.

OCI Enterprise AI

OCI Enterprise AI is for developers building their own agents on Oracle's cloud. Oracle announced it on 24 March as one service with three layers: models and inference, agents and tools, and governance and security. It is compatible with OpenAI's Responses API, so code written to that pattern needs less rework, it supports MCP for tools and the A2A protocol for agents talking to each other, and it offers hosted deployment so that a team does not have to run the agent infrastructure itself. Oracle said general availability was coming and listed nine regions, including London and Frankfurt.

It sits beside Oracle AI Data Platform, which became generally available at AI World in October 2025. That service combines data ingestion, semantic enrichment and vector indexing with generative AI tools, so that enterprise data can be prepared for models in one place, with open formats such as Apache Iceberg and Delta Lake for lakehouse storage.

New names for the multicloud database

Oracle renamed its database services inside the other clouds on 24 March, to Oracle AI Database@AWS, Oracle AI Database@Azure and Oracle AI Database@Google Cloud, and said the new names reflect the AI features built into the database engine. In all three, Oracle runs Exadata hardware inside the other provider's data centres and the customer buys through that provider's marketplace. When Oracle Database@AWS became generally available in July 2025, the two companies said the spending counts towards a customer's existing AWS commitments.

On the 10 March call, co-chief executive Clay Magouyrk said Oracle began the quarter with two AWS regions live, ended it with eight and would end the fourth quarter, which closes on 31 May, with 22. He put the counts at 33 live regions with Microsoft and 14 with Google. Multicloud database revenue grew 531 per cent in the quarter, although Oracle gave no dollar figure for it.

The infrastructure and the order book

Oracle's third-quarter revenue, reported on 10 March, was $17.2 billion, up 22 per cent. Cloud revenue was $8.9 billion, up 44 per cent, of which cloud infrastructure was $4.9 billion, up 84 per cent, and cloud applications $4.0 billion, up 13 per cent. Fusion Cloud ERP and NetSuite brought in $1.1 billion each. Remaining performance obligations, the contracted revenue not yet recognised, reached $553 billion, up 325 per cent on a year earlier.

Oracle said most of the quarter's increase in remaining performance obligations came from large AI contracts in which the customer either prepays so that Oracle can buy the GPUs or supplies the GPUs itself, so Oracle did not expect to raise extra funds to support them. In December, when the figure reached $523 billion, Oracle had named Meta and NVIDIA among the new commitments. Oracle expects capital spending of $50 billion in the fiscal year ending in May, on revenue of $67 billion, and its free cash flow over the previous 12 months was negative $24.7 billion.

The largest cluster design Oracle has announced is OCI Zettascale10, unveiled in October 2025. It connects hundreds of thousands of NVIDIA GPUs across several data centres within a two-kilometre radius, on Oracle's Acceleron RoCE networking, for up to 16 zettaFLOPS of peak performance, with initial deployments of up to 800,000 GPUs. Oracle said it underpins the supercluster it is building with OpenAI in Abilene, Texas, as part of Stargate, and that Zettascale10 would be available in the second half of 2026.

How the pieces connect

Access control runs through all three sets of announcements. A Fusion agent works within its user's Fusion roles, Deep Data Security carries a user's identity into the database for agents built outside Fusion, and OCI Enterprise AI has a governance layer of its own. The agentic applications and AI Agent Studio are Fusion features, while the database tools are available to on-premises and multicloud estates as well.

SAASiQ's view is that for an existing Fusion customer the practical place to start is the security role design, because every one of these agent features inherits it.

Oracle's fiscal year ends on 31 May, by which date it expects 22 AWS regions to be running its database, and Zettascale10 is due in the second half of 2026.

SAASiQ - Intelligent Solutions for SaaS ©