
Greg Sharma
Network Infrastructure Authority
Enterprise network planning has always followed the rhythm of the working day. We estimate demand by headcount, plan capacity around business hours and schedule maintenance when user activity is low. In food manufacturing, production runs around the clock, but many business applications still see their heaviest use during office hours.
AI agents challenge that model. Unlike assistants that answer questions, agents can carry out tasks across multiple systems, triggered by schedules, events or requests. As they move from pilots into daily operations, the network will carry not only the work of people but also work that no person is watching. The most important change is dependence. When automated work relies on a connection, that connection becomes part of the business process.
Cloud-Hosted Agents Still Depend on the Network
Running an agent in the cloud does not remove its dependence on the network. What matters is where the systems and data it needs are located.
An agent working entirely within cloud services may add little load to an office or plant connection. An agent that reaches an ERP system, file share or plant application hosted inside the business depends on every link between those environments. When those links support order processing, scheduling or inventory updates, they become part of how the work gets done. A congested or failed connection can stall a process even when no employee is waiting at a screen.
Quiet Hours Are No Longer Safe to Assume
Agents do not need to follow the office day. They process scheduled work overnight, respond to events and continue tasks after employees have signed off. Industry research projects that enterprise network traffic could grow several times over the next decade as agent adoption spreads, although the impact on any one organization will depend on what it automates and where those systems are hosted.
When automated work relies on a connection, that connection becomes part of the business process.
The more immediate consequence is operational. Evenings and weekends have long been treated as safe windows for firmware upgrades, changes and backups. A connection that looks quiet overnight may now be supporting an important automated process. The question shifts from when the fewest people are online to which business processes depend on a connection and what happens if it is interrupted.
Headcount No Longer Explains Demand
An agent completing a task typically generates far more network activity than a person doing the same work. It gathers information repeatedly, queries several systems and may hand work to other agents. Some workloads also send documents and operational data from company systems to cloud models, placing demand on upload capacity that many branch and broadband circuits are least equipped to provide.
As a result, two sites with similar headcounts can have very different requirements if one hosts systems used by automated workloads across the business. Capacity conversations will therefore need to move from how many people use a site to which workloads run there, where their data sits and when they need to operate.
Where Infrastructure Leaders Should Focus
The right starting point is a clear understanding of the work being automated and the connections it depends on, not a blanket bandwidth upgrade. For each proposed agent, map the systems it will reach, where they are hosted and when the task runs. Then decide what happens if a connection fails. Can the work wait? Will it retry? Or will someone need to step in? This turns a general conversation about AI readiness into a practical discussion about business continuity. From there, use real traffic data from early deployments to add capacity in both directions where the evidence supports it, and provide backup paths for connections that become critical.
Maintenance planning should include automated workloads, not just users. A quiet office does not necessarily mean an inactive business. Teams also need visibility into which agents are reaching which systems and how much capacity they consume. A healthy connection does not guarantee that the business process has completed correctly. Infrastructure, application and security teams should agree early on what each agent may access and who is accountable when something goes wrong.
The Network Becomes Part of the Work
AI agents will not affect every enterprise network in the same way. Some workloads will stay largely within cloud services, while others will depend on connections to offices, plants and company-hosted systems. The leadership task is to understand those dependencies before automated processes become business-critical.
Organizations that do this now will scale automation with confidence. Those that do not may discover their network dependencies for the first time during an outage.