/*Making AI useful for DataFlex development*/

Key takeaways

Context matters more than clever prompting

A general-purpose language model knows surprisingly little about your application, your coding conventions or the meaning of your business data. Without that context, it can produce answers that sound convincing but are technically wrong.

Retrieval Augmented Generation, or RAG, addresses that problem by adding relevant source material to the prompt before the model generates an answer. The DataFlex documentation assistant combines traditional keyword search with semantic search, then re-ranks the results to select the most useful documentation fragments. This matters for developers because exact DataFlex class and property names often require keyword precision, while broader questions benefit from meaning-based search.

AI quality depends on source quality

One of the most useful examples in the session is also one where the AI gets the answer wrong. The system retrieves the correct documentation page but interprets an ambiguous passage incorrectly and suggests the wrong data type.

The lesson is important: adding AI does not hide weaknesses in documentation. It exposes them. Better examples, clearer wording and more logical content chunks directly improve the quality of generated answers. For teams maintaining internal frameworks, subclass libraries or development standards, AI readiness therefore starts with making existing knowledge easier to understand and retrieve.

Give AI tools, but keep control

Model Context Protocol, or MCP, allows an AI assistant to use predefined tools such as file readers, editors, compilers or database connections. The model does not receive unrestricted access. It can only perform actions exposed through those tools and within the permissions developers define.

That distinction makes AI agents much more relevant to business application development. A read-only query tool cannot suddenly delete records. A Studio agent can edit files and compile code without automatically gaining access to everything outside the workspace. The quality of the agent depends not only on the model, but on the tools, boundaries and feedback loops built around it.

Business meaning belongs above the database

The AI-aware SQL prototype shows why generating valid SQL is not enough. A model may understand tables and columns but still calculate revenue, order counts or profit incorrectly because it does not know the company’s definitions.

A semantic layer solves this by translating technical structures into controlled business concepts. Developers define what revenue means, which dates apply and how KPIs should be calculated. The AI can then answer natural-language questions using those rules instead of guessing from column names. For ISVs and application developers, this is a useful way to rethink how domain knowledge could become an explicit part of the application architecture.

Existing codebases give agents something to learn from

The DataFlex Studio agent can inspect a workspace, edit a view, compile the result, read compiler errors and correct its own changes. It also works particularly well when an existing codebase provides examples of established patterns.

That opens an interesting direction for experienced teams. Subclass layers, naming conventions and recurring application structures are not just implementation details; they become context the agent can learn from. The more consistent the codebase, the more useful an AI assistant may become for focused changes, repetitive additions and eventually broader refactoring work.

Why it matters

The session presents AI as an engineering problem rather than a feature checkbox. Reliable results require relevant context, strong source material, controlled tools, clear permissions and explicit business rules.

For DataFlex developers, that creates several practical areas to explore today: improving internal documentation, standardizing code patterns, defining business terminology more clearly and identifying small, bounded tasks that an agent could safely support.

FlexRoads 2026 On-demand

Access the complete FlexRoads 2026 experience with all 17 session recordings, including technical deep dives, live demonstrations and sessions that go beyond these highlights.

• Go to the FlexRoads event course on the DataFlex Learning Center.
• If you attended FlexRoads 2026, you already have free access when logged in with your DataFlex Account.
• Not an attendee? Purchase on-demand access to all FlexRoads 2026 session recordings for EUR 175 by contacting info@dataaccess.eu.