Why Amend Uses Managed AI Models

Photo by Steve A Johnson on Pexels

Amend did not train a foundation model. Mend, the assistant inside Amend, uses managed AI models through Google Cloud and combines them with the platform's workflow, permissions and business context.

That distinction matters. A general model can generate and summarize text, but it does not know which customer a user may access, which estimate is current or whether an action requires approval. The surrounding product has to provide those controls.

The model is one part of the system

When a user asks Mend for help, Amend selects the relevant records the user is allowed to access and supplies the context needed for the task. The system then validates the proposed result and presents consequential actions for review. Activity records help show what was prepared and what the user approved.

Model versions can change after testing. The product should improve without asking customers to follow vendor model names or rebuild their workflows every time the underlying technology changes.

What Amend builds around it

Our work is the contractor workflow: the data model, permissions, connections between customer and job records, approval gates, user experience and safeguards. Those parts determine whether an AI response becomes useful office work or just another block of text.

Using managed models also does not mean customer data becomes public training material. Amend's current controls, subprocessors and data practices are described on the Security page.