AI Agent Tools and Enterprise Integrations
Tools give Digital Employees approved capabilities beyond model reasoning and generation.
Depending on the deployment, tools can include:
- web search and browsers
- files and document repositories
- enterprise applications
- APIs
- native connectors
- databases
- coding environments
- messaging or workflow systems
- MCP servers
- customer-specific services
Tools are capabilities, not jobs
A tool does not define what a Digital Employee is responsible for.
The Digital Employee defines the job. The skill determines why and when a capability is required. The execution layer invokes the approved implementation.
This separation allows a tool to be replaced without rewriting the workforce design.
Request capabilities, not hard-coded systems
Where practical, a skill should request the information or capability it needs rather than hard-code one customer’s system.
For example, a skill may request approved past performance records. The Customer AI Control Plane can route that request to SharePoint for one organization, another document repository for a second organization, or approved uploaded files for a smaller deployment.
That keeps customer infrastructure out of the reusable business method.
Tools and permissions
The existence of a tool does not imply permission to use it.
Tool access can be constrained by the Digital Employee’s role, the directing human’s entitlements, the workflow stage, the requested action, or other customer policy.
This is particularly important for tools that change systems, communicate externally, or retrieve restricted information.
Replaceable execution
As model platforms add native search, browsers, computer use, connectors, or other capabilities, the AI Team can use those mechanisms without redefining the business job.
The architecture treats the capability as replaceable infrastructure.
See Model Context Protocol and AI Teams for one standardized way to expose capabilities.