Does the agent get its own credentials?

It should not need them. Acting with the permissions of the person it works for is what keeps an agent from becoming a way around the access controls the company already set.

What the agent did, in which system, on whose authority, and whether a person approved it. That record is what makes an action defensible after the fact.

Governed Execution

Governed execution is an AI agent taking real action across company systems, but only within defined permissions and approval points.

It names the middle ground between the two positions companies usually start from. One is read-only AI, which is safe and stops short of the work. The other is an autonomous agent with broad credentials, which finishes the work and is difficult to defend to a security team. Governed execution is the agent acting for real, inside boundaries that were decided in advance.

Three things make it work: the agent inherits the acting person’s permissions rather than holding its own elevated ones, consequential actions stop at an approval workflow, and everything it did is recorded well enough to answer for afterwards. With those in place the interesting question stops being whether to let AI act and becomes which actions to gate.

Frequently Asked Questions

Automation follows a path someone defined in advance. Governed execution lets the agent work out the path, and constrains what it is allowed to do along the way.