Extend the system you already own
Bring the current application and the change you want. You do not need to start from an empty repository or adopt a new stack to use AI Agent Creator. An existing platform-managed agent or workflow can also be the starting point.
The creator inspects the selected workspace before asking about missing facts. It designs the addition while leaving application code and infrastructure unchanged. Implementation starts only in a separate session.
Give the creator usable context
Open the application's repository, including relevant platform configuration when it is maintained there. Point to the current architecture or operating records if they live elsewhere in an approved location.
Explain the intended change and the interfaces that must remain stable. Identify enterprise controls outside the repository, such as shared identity or network services. Use synthetic inputs until customer-data access is agreed.
| Context | What matters for the design |
|---|---|
| Current journey | What the user does today and where the proposed capability changes it. |
| Interfaces and data | Existing API contracts, systems of record, and source owners. |
| Access and approvals | Who may read evidence or authorize a business operation. |
| Delivery and operations | Current environments, release checks, telemetry, and rollback ownership. |
| Constraints | Required stack, compatibility promises, and limits on migration. |
The creator should inspect repository facts before asking you for them. It asks you about enterprise facts it cannot verify. If a control or service is not documented in the repository, the creator must check rather than assume it is absent.
Start with one useful addition
Choose a first journey whose result you can evaluate. For example, an existing service application could gain an evidence-backed case summary while leaving case editing and assignment unchanged.
If follow-up investigation is required, include it explicitly. Otherwise, one bounded model response may be enough. If the system prepares an action, preserve the existing authorization checks and name the approval owner.
Agree on how you will judge the addition against today's behavior. Define what stays unchanged, including response contracts and access boundaries. Specify how the application behaves when AI is unavailable or evidence cannot support an answer.
This example does not require a chat panel or a replacement workflow.
Make reuse and migration explicit
Evaluate the existing stack and model deployments before introducing alternatives. Reuse suitable capabilities. Before proposing a replacement, name the requirement or gap it addresses and the cost of changing the system.
When migration is justified, the architecture package should explain coexistence and compatibility. The implementation plan should identify data movement, cutover, rollback, and eventual decommissioning where applicable.
Name the system of record and the owner of each operation during the transition.
Review the addition as a complete package
The creator produces the same four architecture artifacts used for a new project, but their scope is the agreed change. The specification should distinguish new acceptance criteria from behavior that must be preserved.
Review the data and action boundaries alongside the component changes. Check that the delivery plan includes regression tests and release controls appropriate to the existing application.
Discovery confirmation and complete-package approval remain separate gates. After approval, use the implementation handoff in a new session. If scope or architecture changes, repeat the affected approval step.
Begin with the intended change
/ai-agent-creator Add evidence-backed case summarization and follow-up investigation to this
existing application. Inspect the current stack and architecture records.
Preserve the case APIs and authorization model. Reuse our enterprise services,
compare agentic and simpler execution, and plan a bounded first release with
regression checks and rollback.
Install the plugin and supporting skills without replacing existing project configuration.