When software ships its own chatbot, your AI is whatever the vendor built, improving at the vendor's pace, limited to the features the vendor wired up. Bring-your-own-agent inverts that. The software becomes something your own agent operates, and you keep the choice of which agent that is.
What "bring your own agent" means
Your workspace is exposed as a server that speaks the Model Context Protocol, the open standard for connecting agents to tools. Any compatible client can connect: the assistant you use for everything else can now also run your business, with full access to the same tools a person would use, within your permissions.
Why open matters
Three reasons. Choice: you are not married to one vendor's assistant, so you use the agent you already trust. Capability: an open connection exposes every action, not the handful a vendor decided to surface, so the agent can actually finish work rather than nibble at the edges. And the frontier: models improve constantly, and an open protocol lets you point the better one at the same workspace the day it ships, with nothing to rebuild.
Governance comes with it, not against it
Open does not mean unguarded. The agent acts only within the permissions of the person it represents, checked when tools are offered and again when they run. Usage is metered against a budget you control. Every action is recorded, so what the agent did and when stays accountable. You get the flexibility of bringing your own agent and the control of a system that knows exactly what it is allowed to do.
And if you do not have an agent yet
Bring-your-own is the richest path, but it is not a requirement. A workspace also accepts a plain-language request directly, does the work, and replies, so a caller with no agent of their own is never stuck. You can connect your own agent later and lose nothing.
The shape of the next decade of software is open by protocol, operated by the agent you choose. See how to connect yours, or launch a workspace and point your agent at it.