Agent hosting considerations
Now that you have a sense of the broader agentic concepts, let's discuss what it means to host and run these agents. We must think about how and where computations run, how they scale, how they operate, and how they're managed. At the same time, some patterns that we expect to see as agents are more widely applied and adopted. The following diagram shows an example of likely permutations.
Three distinct strategies are represented here. On the left-hand side of the diagram, you see a model where our agents are hosted, scaled, and managed within the environments of each agent provider. These agents are published and consumed as services, operating in what is labeled as an agent as a service (AaaS) model. On the right-hand side is a model where a provider's agents are all hosted in a dedicated customer environment.
In the middle of the diagram is a mixed deployment model that combines these two strategies, hosting some agents locally in the customer's environment and interacting with some agents that are hosted remotely in a provider's environment.
A fourth option (not shown) could be where agents are built as low or no-code services that are scaled and managed by agent infrastructure services. We won't cover these in detail because the architecture and hosting of managed agents is dictated primarily by the organization that owns the services.
You can imagine the range of factors that might influence the adoption of one of these models. Compliance, regulatory, and security constraints, for example, could push someone toward customer-hosted agents. Scale, agility, and efficiency could push organizations more toward the AaaS model.
The key concept here is that agents can and are deployed and hosted in many ways. It's your job to determine how agents can best be applied. The footprint, security, and deployment, among other factors, significantly affects how you approach building and operating agents. Private and public agents, for example, may have different designs and release lifecycles.
