Oracle unveils AI Agent Marketplace

LAS VEGAS—Oracle has introduced an online store that lets its Fusion Cloud Applications customers buy and install AI agents from a wide range of software developers, and it has enhanced its supply chain, sales, and marketing software with new LLM-powered tools.

Oracle Fusion Applications AI Agent Marketplace, accessible on the Oracle AI Agent Studio development platform, lets customers of the company’s finance, logistics, HR, and customer management applications find more than 100 agents from third-party software vendors and IT consultancies—then deploy them directly inside those applications, Oracle said this week at its AI World customer conference. The agents available in the marketplace from more than two dozen partners, including Accenture, Deloitte, IBM, Infosys, KPMG, PwC, Wipro, Box, and Stripe, supplement the roughly 400 Fusion Cloud assistants and agents that Oracle supplies.

Oracle also announced new AI agents for its Fusion Cloud supply chain management, sales, and marketing applications, and it has redesigned the applications’ starting screens to include conversational prompt windows for AI queries and status updates on agents’ work.

In addition, the company updated AI Agent Studio with support for Model Context Protocol (MCP), which lets agents communicate with other enterprise software outside of Oracle Fusion Applications, and Agent2Agent (A2A), a communication protocol for interoperability among agents made by different vendors. AI Agent Studio, a graphical setup tool Oracle released earlier this year for customizing prebuilt agents or building new ones from scratch, now lets organizations measure the amount of generative AI “tokens,” or word fragments, that applications consume. That can indicate the running costs of interacting with LLMs.

“Because these agents are embedded inside Fusion applications, it builds context,” Oracle Executive Vice President Steve Miranda said during his keynote at AI World. For instance, users of accounting software can monitor variance in a company’s liabilities, get alerts for notable information, and drill into the invoices that caused an unexpected result, without switching among screens and with their security rights applied to the results. “You don't need to code the context with Fusion,” Miranda said.

AI agents work inside applications to gather data, interact with computer users, and take step-by-step actions to complete processes. Companies are starting to use them to analyze sales pipelines and offer reps tips to close deals, help supply chain managers react to changing market conditions, resolve customer service requests, and set up interviews with job candidates.

Oracle and other developers of back-office and customer-facing applications are supplementing their own AI agents with those from outside vendors and consultants, positioning them as productivity boosters. Boston Consulting Group forecasts the AI agents market will grow ninefold through 2030, to $52.1 billion.

While three-quarters of the 1,803 C-level executives that BCG surveyed globally in January (PDF) rated AI as one of their top three strategic priorities, only a quarter of respondents said their organizations were generating “significant value” from their AI initiatives.

Near-term value

Agents could help generate some near-term value from enterprise AI projects by streamlining processes and paring customer support costs.

Agents available in the new marketplace include

• An HR assistant from Infosys for updating staff profiles while adhering to Fusion’s controls for accessing and changing employee data
• KPMG’s procurement helper, which helps speed the process of calling up supplier, purchase, and price data
• A sales order entry assistant from IBM that prompts users for additional information in order to reduce errors
• An invoice collection agent from Stripe and Infosys that can offer incentives for some billing arrangements to reduce days sales outstanding, a metric related to cash flow

New Fusion agents available directly from Oracle can help accountants create journal entries, procurement planners automate quote requests, and fulfillment managers speed shipping requests. Others can help marketers prioritize customers by their likelihood to buy, as well as arm sales staff with pricing information and customer references.

Unlike previous generations of software assistants that relied on coded rules, AI agents use pretrained large language models to converse with users and in some cases, understand their intent, adding context from organization-specific information, such as job categories, policies, and customer accounts. Agents can also draw on their memory of previous interactions and learn from past scenarios to provide better answers or complete tasks more accurately in the future.

View more Oracle Connect articles