Terminology 01
Ontology: Giving AI an Understanding of Your Business
Ontology sounds intimidating, or medical. But the basic idea is fairly simple.
Think of it as a structured description of what your business data means and how different things relate to one another.
A database might contain a field called SUPPLIER_ID. Metadata can explain that this field identifies a supplier, along with alternate names, a business definition, and technical information about the field.
An ontology goes further by defining concepts and their relationships. A supplier has sites. A supplier receives purchase orders. Purchase orders have schedules and receipts. Receipts represent physical or financial business events.
Why does this matter for AI?
Imagine asking an AI system: "Which suppliers consistently deliver late, and what is the financial impact?"
AI needs to understand the relationship between suppliers, purchase orders, promised delivery dates, receipts, prices, and business organizations.
Simply knowing the individual database columns is not enough. Business terminology, relationships, and calculations provide important context for a useful answer that isn't AI generated nonsense.
Understanding "Late Supplier Delivery"
A user might ask about late shipments, but the database stores purchase orders, schedules, promised dates, receipt transactions, and suppliers in separate objects. Don't get me started on version control among those!
An ontology describes how these objects relate and why you need to consider the various objects. A semantic layer defines exactly how a metric such as on-time delivery percentage is calculated. Not so complicated any more is it?
My takeaway: Ontology is not complicated, neither are Semantic Layers. It is part of the structure that helps AI understand what the business means instead of guessing based on column names.
Terminology 02
FSCR: AI-Assisted Oracle Fusion Configurations
FSCR, pronounced "fi-skur," stands for Fusion Setup and Configuration Skills Registry. You could describe it as "cutting edge" technology, if puns are your thing.
Oracle publishes this open-source reference toolkit on GitHub. This may be new for you, but technical nerds are rejoicing. It includes AI-compatible skills, reference information, scripts, and workflows to assist with planning and carrying out supported Oracle Fusion setup activities, all bundled up in a way that agents can quickly understand.
You "download the repo" (don't worry if this sounds like a made up phrase) and use it alongside an AI coding tool, like Codex or Claude Code, maybe even Antigravity.
How does it work?
Think about a typical Oracle Fusion implementation. Consultants hold workshops, collect requirements, document configurations, identify dependencies, prepare setup files, enter configurations, test results, and maintain implementation workbooks.
FSCR provides a software for accelerating and documenting portions of this process.
Provide your requirements
Give the coding assistant documents, spreadsheets, meeting notes, or other input formats supported by your chosen tools. PDF, Word, Excel, PowerPoint, and images are all supported today. No word on smoke signals or mind reading.
Interpret the business requirements
AI references the FSCR skills and relevant setup guidance to identify potential configuration decisions, dependencies, and unresolved requirements. While it ships with what Oracle thinks, your system implementer will augment this with their own industry specific leading practice design, and their own example configuration workbooks, so you continue to leverage the huge advantage of an experienced consultant.
Propose a configuration plan
Review the proposed configurations, implementation sequence, assumptions, and questions before execution. Still makes great bedtime reading materials, my kids fall asleep before I'm done with purchasing category hierarchy review.
Approve the proposed work
Your or your team review suggestions and approve what may be executed against the Fusion environment.
Execute supported setup activities
Depending on the task, FSCR will use supported REST services, setup import/export capabilities, data loading mechanisms, and other supported methods. At this point, it should be able to run while you drink coffee, sleep, or write articles for your website.
Document and validate
Maintain setup records, runbooks, validation results, exception logs, and other implementation artifacts. FSCR logs everything!
What makes this special?
A system implementer can incorporate its own leading practice standards, methodology, templates, and client-specific guidance.
For example, a healthcare Fusion implementation might add design standards for PAR replenishment, inventory organizations, procurement business units, ship-to locations, and receiving. I know I have my own playbooks ready based on what I know about how healtchare works, and my knowledge of the quirks within Fusion.
Is it free?
FSCR is published as open source, so you don't pay Oracle to download and use it. But, you run it on your computer with your AI coding tool, where you pay them directly for tokens consumed.
What FSCR does not mean
FSCR is also not a replacement for functional design, experienced consultants, approvals, testing, or security. It is primarily an implementation accelerator rather than a general-purpose automatic optimization service.
Oracle's repository requirements and supported releases should be checked before any deployment. Future migration or optimization offerings should not be assumed available until documented.
Today, this is built for new implementations only. Optimizations and EBS to Fusion should be supported in a later release. My best guess would be mid 2027.
My takeaway: FSCR will reduce repetitive setup activity. But the value will depend on good requirements, approved configurations, and effective oversight by experienced implementation professionals. We are talking about weeks cut off of a project, not months.
Terminology 03
AI Agents: You Ask, They Help Do the Work
An AI agent is typically a chat interface that interprets an objective and take steps toward completing it, using the tools and permissions available to it. You bring the issue to it.
The easiest way to understand an agent is through a chat interface. You explain what you need, and the agent attempts to help.
"Why hasn't this purchase order been received?"
An agent might retrieve the purchase order, inspect its shipment and receipt information, identify a likely exception, and explain what is blocking the process.
If it has access to appropriate tools and permissions, it might also be able to initiate a transaction or request corrective action.
Not every agent has the same level of capability. Some are designed primarily to retrieve and explain information. Others may use multiple tools or carry out authorized work.
Agents also do not necessarily require a chat window. They can be initiated by workflow events, application activities, or other software, but it's still something bringing the task to the agent.
Terminology 04
Agentic Applications: The Application Knows What Needs Attention
Agentic Applications take the concept further. Instead of requiring users to initiate every interaction, an Agentic Application will proactively identify issues, coordinate specialized agents, and present work in the context of a business process.
Consider a procurement manager who is responsible for ensuring that critical supplies arrive on time.
Traditional Report
A dashboard displays twenty overdue purchase orders. The manager must investigate them one by one in the UI and determine which ones require action.
AI Agent
The manager asks which purchase orders deserve attention. The agent investigates and provides an explanation, which may include a report showing 20, or focused on a subset, based on the prompt. The agent can bring the required information, so you can complete the full investigation and take action from the chat window. A moderate time savings over traditional reports.
Agentic Application
The application identifies delivery risks, considers their business impact, prioritizes work, and presents recommended actions or tasks for resolution, even without you asking. The actions are primed and ready with one click, instead of interacting via chat for each item. A significant time savings over traditional reports.
How is this different from ordinary alerts?
A traditional alert generally relies on predetermined rules. For example, if a purchase order is more than five days overdue, send a notification.
An Agentic Application can consider additional factors such as material criticality, current inventory, expected demand, alternate suppliers, and delivery patterns.
Instead of only reporting an exception, it will determine what should happen next.
My takeaway: An agent is often a specialist assigned a task. An Agentic Application coordinates capabilities around business outcomes.
Terminology 05
Fusion Claw: Research, Compare, Plan, and Execute
Fusion Claw is one of the more interesting technologies discussed in Oracle's 2026 AI announcements.
Fusion Claw is a governed execution runtime for Fusion Agentic Applications.
That is a complicated way of describing software that enables sophisticated analysis, planning, computation, simulation, and execution.
The AI helps determine what needs to be investigated and which options deserve consideration. Deterministic software performs calculations and other operations that require precise, repeatable results.
A supply chain example
Imagine a planner responsible for delivering material to several facilities.
There are different carriers, transportation costs, capacity constraints, delivery windows, and opportunities to consolidate shipments.
Identify the problem
The application identifies that the current transportation plan is at risk.
Research possible solutions
Review available transportation alternatives, constraints, expected arrival dates, and costs.
Evaluate alternative plans
Compare scenarios using calculations and optimization or simulation capabilities available to the application.
Recommend a plan
Present the preferred alternative together with tradeoffs, constraints, and expected outcomes.
Execute within approved boundaries
Initiate authorized actions, with human approvals where required by the operating model.
Reevaluate as conditions change
Replan when new information makes the earlier decision less appropriate.
Does a human always have to approve?
Not necessarily. Depending on the solution and its governance model, organizations may require approval, permit guided execution, or allow actions automatically within defined authority and risk limits. For example, you could allow the agent to take actions itself when the impact is minor, while still requiring human in the loop for high impact decisions.
Can you use Fusion Claw without Agentic Applications?
Fusion Claw is an execution runtime on top of Fusion Agentic Applications, not simply as a standalone chat interface. So you need Agentic Applications in order to use Fusion Claw.
Does Fusion Claw have an extra cost?
Oracle's Agentic Apps Platform has its own subscription requirements, but Fusion Claw is included in that subscription. Claw may consume more tokens than a basic Agentic Application, but the search methods have been optimized so you may find yourself pleasantly surprised.
Don't confuse automation with unlimited autonomy
Complex actions still require appropriate permissions, business policies, approval rules, testing, auditability, and controls. The degree of autonomy should match the risk of the decisions being made.
My takeaway: Fusion Claw provides a foundation for tackling more sophisticated business problems involving multiple possible plans and changing conditions, rather than simply answering questions or reporting exceptions.
