The future will not be just about singular agents but more about networks/systems of agents where agents can discover and collaborate with other agents.
In the same way that current LLMs support tool calling (whereby a set of tools is made available to the LLM, which it can then use to formulate a more useful response), in the future, agents will be provided access to a network/system of other agents. The agent can then invoke these other agents in order to handle tasks that it can't do itself.