
AI Agent
Existing application integration
AI can be added to an existing application without replacing its business rules, data or access controls.
Existing application integration
An existing application may hold valuable rules, customer data, integrations and operational knowledge. It need not be replaced to add AI; the suitable approach depends on the system’s condition and the business objective.
AI may summarise a message, suggest a ticket category, prepare a response draft, propose document values or search approved knowledge. These are possible capabilities, not GiSoft deployment claims.
Responsibility layer
Application
Manages access, data, business rules, the current business state and the set of actions approved for execution.
AI integration
Selects the data required for the task, prepares the relevant context, calls the AI provider and validates the structure and correctness of the returned output before further use.
AI model
Generates a suggestion, summary or other requested output based on the context provided. It should not independently perform actions outside the scope defined by the application.
Employee
Reviews and verifies the output wherever required by the process, the level of risk or the organisation’s internal rules.
In a hypothetical Symfony CRM, a customer asks about modernisation and connection to a new ordering system. AI may propose category, priority and follow-up questions. The application selects data and an employee checks the proposal; it is not an automatic diagnosis.
For an invoice, AI may suggest fields, while the application checks required values, duplicates, formats, permissions and business rules. Structured output or confidence does not confirm correctness. A knowledge assistant may cite only documents the user can access; a citation alone does not prove truth.
Timeout, invalid output or duplicate delivery need visible state and process-specific behaviour. Manual fallback and duplicate prevention must be designed and verified. Start with a narrow case and selected users, then measure usefulness, corrections, failures and cost per completed case. Changing AI provider may require contract changes and renewed quality, security and cost assessment.
