■ HIGH RISK ■ Technology
Largely, yes — the tier-one help desk is being hollowed out fast. What remains is the physical and the political: hands on hardware, and the escalations where someone senior is angry and the answer isn't in the knowledge base.
“Have you tried turning it off and on again? AI has — 10,000 times before you even called.”
Our AI replacement risk score — how we score jobs
The job is a queue. Password resets, VPN failures, printer drivers, mailbox permissions, a laptop that won't wake, onboarding a new hire with the right software licenses, imaging machines, and patiently explaining that the file isn't gone, it's in OneDrive. Ticket in, diagnose, resolve or escalate, close with notes. Underneath sits inventory, asset tracking, and the occasional 2pm panic when the conference room display refuses to cooperate ninety seconds before the board call.
Self-service already ate the top of the funnel. Password resets went automated years ago; identity platforms handle provisioning and deprovisioning from HR records; MDM tools push patches and reimage devices remotely. Now LLM agents plugged into the ticket system and the knowledge base handle the conversational tier: they read the error, ask the right clarifying question, execute the fix through an API, and close the ticket with better notes than a tired human writes at 5pm. Endpoint telemetry flags failing disks before the user notices. The economics are brutal — support volume that used to justify a team of six can be handled by two people plus an agent, and vendors price accordingly. Our risk score of 75 reflects a role losing its entire entry tier.
Resistance comes from atoms and from status. Someone must physically swap the laptop, run the cable, deal with the printer that is genuinely broken, and stage hardware for a new office. Someone must sit with the executive whose presentation is failing in front of clients — a moment where the value delivered is reassurance as much as repair. Weird legacy systems, air-gapped environments, and manufacturing floors full of decade-old machines also stay human. But note the shape: fewer roles, more field and infrastructure work, less of the phone-and-chat queue that employed most of this workforce.
Automatability: our editorial assessment of current and near-term AI capability
Already biting. Automated identity management removed the single highest-volume ticket type years ago, and AI service-desk agents are now closing a growing share of the rest without human touch. Expect tier-one help desk headcount to fall steeply through the late 2020s, with outsourced contact-centre style support hit hardest and fastest. Field and infrastructure roles decline more slowly and will persist through the 2030s wherever hardware physically exists.
It's a narrowing door. The role historically worked as an entry point because volume created junior seats, and that volume is exactly what automation absorbed. It can still work if you treat it as a two-year stepping stone and aggressively build cloud, networking, or security skills on the side. Staying at tier one for five years is now a genuine risk.
We won't quote a number we can't source, but directionally: password, access, and common software questions — historically the bulk of ticket volume — are heavily automated already at organizations that have invested. What survives skews toward hardware faults, unusual configurations, and situations where the user can't accurately describe what's wrong.
Field technician and on-site support roles, because they involve physically touching equipment. Also safer: support for specialized industrial, medical, or laboratory systems, and internal support at organizations with heavy on-premise legacy estates. The most exposed are remote chat and phone-based tier-one roles, especially outsourced ones.
Pick one adjacent depth: cloud administration, identity and access management, networking, or security operations. Then get hands-on with the automation platforms your own org uses — scripting, MDM, workflow tools, and AI agent configuration. Being the person who builds and governs the automated support layer is the most direct path out of the queue.