From Bob

Bob's definition of agent-centric.

Must be a protocol
- Must have an openly published specification.
- Any independently-developed software that follow the specification must be able to use the protocol to communicate with other software components that do likewise.
Must be agent-centric, where agents can be individuals or groups, and each agent must be able to control their own identity, interactions, and data, and select their own software to use, as long as the software follows the protocol.
Must either use a common economic vocabulary to communicate, or be able to translate their ingoing and outgoing communications into a common vocabulary.
The protocol itself must be managed as a commons in the Elinor Ostrom sense.
Must be able to provide reliable local consensus among the participating agents in a scope agreed upon by the participants.
- More on this requirement in a comment.

Some people might disagree with this requirement, but

I think, at this stage of the evolution of the general intellect, the protocol needs to work on the World Wide Web. Can provide private spaces as needed, but accessible via a Web browser.

There is no single protocol that satisfies all of those requirements at this time.

Several candidate protocols were mentioned and discussed in this previous thread. None of them is adequate by itself. Yet.

Agent centric computing

From valueflows, see on github.