Future of enterprise software is agentic apps, says Oracle

Oracle has unveiled its new AI-native builder experience, which will move businesses beyond standalone agents, to AI-powered applications capable of reasoning, planning and executing tasks within existing workflows.

The announcement builds on the company's broader push into agentic AI and introduces new tools designed to help both business users and developers create AI-powered applications with minimal coding, while addressing one of the largest barriers to enterprise adoption: governance.

Agentic applications are the next evolution of enterprise software, replacing traditional applications with systems centred on AI agents, according to Kaushal Kurapati, GVP of Applications Development at Oracle.

"Enterprise apps today are based on static APIs," Kurapati said.

"We're suggesting that agentic applications are based on agents as the primary artefact. They can decide, reason, execute and plan, while helping users solve a business outcome."

Going beyond standalone agents

The builder experience builds on Oracle's AI Agent Studio for Fusion Applications platform, as well as existing agents designed to execute tasks in a predictable and deterministic manner.

Most business processes already follow well-defined steps, making them suitable for AI systems that can operate within clearly governed and structured parameters.

The new agentic application framework layers multiple specialised agents together under a single orchestration layer. The orchestration layer shares context between agents, delegates tasks and presents recommended actions to users.

Beneath that, subject-matter expert agents are focused on specific functions, supported by tools, connectors and enterprise systems. 

The resulting application is packaged as a self-contained deployment that includes the user interface, agents, metadata and integrations, allowing it to be deployed into enterprise environments with a single click.

What is builder experience capable of?

One example of the AI builder experience's considerable capabilities is a sales command centre, designed to help account managers prioritise customer opportunities and risks.

It can quickly optimise sales workflows, using specialised agents focused on renewals, account expansion opportunities and customer risk assessments.

The agents analyse information across systems and provide recommended actions to users.

For example, the application could identify a high priority account at risk of churn in the near term, highlight unresolved objections from previous meetings, and recommend immediate follow-up actions.

The goal is to give employees the information they need at their fingertips, allowing them to focus on important decision-making instead of scavenging for data.

Another application is a manager coaching tool that monitors team health, reviews one-on-one meeting notes, tracks employee goals and identifies engagement issues.

Using natural language prompts, a manager can generate recognition emails, create meeting agendas, and build PowerPoint presentations summarising team performance without leaving the application.

Again, this is designed to cut down on time-consuming and often cumbersome admin that takes attention away from the customer's core business operations and goals.

Won't require advanced coding skills

For those perhaps a bit rusty on their coding, Oracle has focused on simplifying application creation.

Business users can create prototype applications using natural language through Agent Studio's 'agent brain' interface. Users describe the outcome they want to achieve, and the platform identifies suitable agents, assembles them into an application and generates a working prototype within minutes.

For developers or those looking to improve their coding, while retaining more granular control of their application, AI Studio Skills integrates with popular tools such as VS Code or Claude Code.

The integration allows developers to create, test and refine agentic applications, while maintaining access to debugging tools, connectors and deployment workflows.

Developers can also compare different large language models and optimise applications for factors such as accuracy, latency and cost.

Preventing hallucinations

Many organisations rely on lengthy policy documents containing eligibility rules, approval requirements and compliance obligations. Traditional retrieval-based AI systems may interpret these documents differently across interactions, creating risks for regulated industries.

Oracle's approach converts policy documents into executable code using an LLM. The generated code is then validated through automated testing and human review before deployment.

At runtime, the AI agent executes deterministic code rather than interpreting policy documents through retrieval-augmented generation.

This approach delivers consistent outputs and eliminates hallucinations when enforcing business policies.

"When you do retrieval-augmented generation, it can be 90 per cent or 95 per cent accurate, but it can also be non-deterministic between runs," Kurapati said.

"With the policy node, you're now running deterministic code that encapsulates all the rules."

Governance remains critical

As enterprises continue to weigh the benefits of AI automation against concerns about reliability, security and compliance, governance features have been integrated into builder experience from the outset.

These include human approval checkpoints for high-value transactions and critical business decisions, audit trails that capture every action taken by an agent, and debugging tools that allow organisations to inspect how decisions were made.

The platform also incorporates AI guardrails developed within Oracle Cloud Infrastructure to detect prompt injection attacks, identify harmful content and manage personally identifiable information.

Organisations can also restrict agents to specific subject areas using guardrails, preventing them from responding to unrelated prompts.

Customers can choose to run models hosted entirely within Oracle Cloud Infrastructure, or bring their own LLMs, if data residency and privacy requirements demand it.

Kurapati believes these controls will be critical as businesses begin deploying agents into core enterprise processes.

"We've paid a lot of attention to governance mechanisms," he said. 

"Enterprise workflows need to scale and require safeguards so customers can execute these workflows in a deterministic manner with safety, trust and reliability."