Put the small edits into words.

Make the request specific

Name the task and the intended change. A duration, location or task-name adjustment is easier to review than a vague instruction to “fix the production.” If several tasks have similar names, identify the relevant phase or date as well.

In the workspace, the Agent uses the configured model route to propose or carry out supported scheduling actions. Watch the actual change and its receipt. A conversational answer alone is not evidence that the project has been updated.

Read the change in context

After an edit, check the affected task, its dates and nearby work. Look at another view if that makes the consequence clearer. Extending compositing is not the same as confirming the sound team is available or moving the delivery commitment.

The homepage has a deliberately bounded, guided example: choose a request, inspect the preview, apply it and undo it. It demonstrates the interaction without sending a model request or touching your real project. It is not a benchmark of model accuracy or a guarantee that a live request will produce that exact answer.

Know what is connected

The Agent requires an available AI route. An enabled hosted Beta can provide limited access without a visitor key; otherwise, use a supported connection configured in Settings. Hosted availability and allowance are limited, and a separate provider may charge for usage. Manual scheduling remains available when AI is unavailable.

Using the Agent sends your prompt and relevant project context to the configured AI service. Keep sensitive production details out of requests unless that route is appropriate for them. Browser-local storage does not mean an AI conversation never leaves the browser.

• Check the selected connection in Settings.
• Ask for a bounded change and inspect the result.
• Use Undo when needed; keep manual control of the schedule.