■ HIGH RISK ■ Technology
Substantially, over time. Configuration and monitoring are being automated away from both the software-defined and the AI-analytics side, though cabling, campus deployments, and the moments when the network is down and everything depends on it keep humans employed.
“Software-defined networking and AI monitoring are eating this role from both ends.”
Our AI replacement risk score — how we score jobs
Network admins keep the packets moving: configuring switches and routers, managing VLANs and firewall rules, running the VPN, chasing intermittent Wi-Fi complaints across a building, monitoring bandwidth, and troubleshooting the mysterious slowness that turns out to be a duplex mismatch or a backup job nobody scheduled properly. There's a physical half too — patch panels, cable runs, hardware refreshes, and standing in a wiring closet at 6am because a switch stack won't come back up.
The automatable half is dominant. Software-defined networking replaced device-by-device CLI configuration with intent-based policy applied centrally; cloud networking replaced entire physical estates with Terraform files. Controller platforms handle provisioning, compliance checking, and config drift automatically. AI-driven network operations tools baseline normal traffic, flag anomalies, correlate a user complaint with an access point's channel interference, and often propose or apply the remediation. Firewall rule review, capacity reporting, and documentation — traditionally tedious and error-prone — are things models do quickly and consistently. Meanwhile, the workloads themselves keep migrating to cloud providers, shrinking the estate a company needs to administer at all. Our 65 reflects strong automation pressure tempered by real physical work.
What resists is the interface between the model and the wire. Physical layer faults — a chewed fibre, a failing transceiver, water in a conduit — need a person and a ladder. New site builds, warehouse and factory floor networks, and industrial control environments involve constraints no controller understands. Security architecture decisions, vendor negotiations, and outage response where the monitoring system is itself unreachable stay human. And regulated or air-gapped networks in defence, healthcare, and utilities move slowly by design. The likely endpoint is fewer, more senior network engineers who write automation and design topologies, plus field technicians who touch hardware — with the middle-tier admin who mainly logged into devices squeezed out.
Automatability: our editorial assessment of current and near-term AI capability
Erosion is steady rather than sudden and will bite hardest around 2030. Cloud migration has already shrunk on-premise estates, and AI network operations platforms are now closing the loop from detection to remediation. Expect enterprise network admin headcount to keep falling through the decade while field engineering and industrial networking hold up. Facilities that physically exist — campuses, factories, hospitals — will still need people on site well beyond 2040.
It removes the repetitive part of the job, which is most of it for a mid-tier admin. What SDN doesn't do is design a topology for a new factory, diagnose a failing transceiver, or decide how to segment a network after a breach. The role compresses toward senior engineering and field work, with the device-by-device configuration tier disappearing.
Safer than help desk, riskier than security engineering. Our risk score of 65 reflects a genuine squeeze: fewer roles, higher expectations. People who add automation and security skills tend to do fine; people whose value is knowing a particular vendor's command syntax are the ones being displaced as that syntax gets abstracted away.
Mostly detection and correlation: it learns what normal traffic looks like, spots deviations earlier than threshold alerts, links a user's complaint to a specific access point or link, and increasingly executes known remediations automatically. It's very good at the pattern recognition that used to require an experienced admin staring at graphs, and quite bad at anything requiring physical inspection.
It's one of the strongest available moves. Network knowledge is genuinely valuable in security operations, and security has the useful property of being adversarial — attackers adapt, so the work doesn't stabilize into something automatable. Firewall, segmentation, and traffic analysis experience translate directly, and the demand outlook is considerably better.