Choose the right route.
Use a dependable API where one exists. Where it does not, define the permitted screens, operations and stopping conditions.
Connect AI to the applications you already use. Read screens, prepare changes and check the result, with a person approving consequential actions.
Explore the capabilityOlder applications and supplier portals can leave useful work outside an integration. Computer use provides another route, but the interface, access and failure modes need careful investigation.
Use a dependable API where one exists. Where it does not, define the permitted screens, operations and stopping conditions.
Isolate the session, scope the credentials and check the destination before entering information or committing a change.
A click is not confirmation. Check the resulting state, record the action and stop when the interface behaves unexpectedly.
Prepare a service record in a legacy application, check its fields and pause before submission.
Interfaces change. Build tests for missing controls, unexpected screens and partial completion; do not give an agent unrestricted desktop access.
No. Reliable APIs are usually the better integration route. Computer use is useful when the interface is the practical access point.
Only within an explicitly agreed permission model. Sensitive or consequential actions should have appropriate approval and independent checks.
Technical reference: OpenAI: computer-use guidance (opens in a new tab)