Partly, and the employment data already reflects it. Federal figures put computer support specialists at a 3 percent decline from 2025 to 2035, losing about 24,300 positions from a base of 903,100, with median pay of $62,890. Self-service, automated password resets and chat assistants absorbed a large share of the tickets that used to reach a person, and that process started well before generative AI.
What survives is the part that was always hardest: working out what is actually wrong when the user’s description is incomplete, and dealing with the person as well as the problem.
The short answer
IT support roles are contracting rather than disappearing, and what is emerging is a hybrid model rather than full automation. Tier one work, meaning password resets, access requests, standard software installs and answering documented questions, is heavily automated and continues to shrink. Diagnosis of unusual faults, physical hardware work, security incidents and anything requiring judgement about competing priorities remains human.
The practical consequence for anyone in the field is job evolution rather than job displacement. Support experience leads directly into systems administration, security operations and platform work, and those adjacent fields pay considerably more.
- What AI tools handle now: ticket triage and routing, password and access self-service, knowledge base answers, standard software deployment, and first-line responses to documented questions.
- Where efficiency actually comes from: removing the highest-volume identical requests, not from answering hard questions faster.
- What stays human: diagnosis under uncertainty, physical work, security response, and the customer experience side of a difficult interaction.
- What IT support teams look like now: smaller at tier one, unchanged or larger at the diagnostic and security end.
- Where the human versus AI line sits: not by difficulty, but by whether the problem was described well enough to be looked up.
Five moves that keep you on the right side of it
If you work in support and want to be on the right side of this, five concrete moves matter more than anything else.
1. Measure your own ticket mix. Take a month of your closed tickets and sort them into three piles: resolved by following a documented procedure, resolved by diagnosis, and resolved by talking to someone. The first pile is your exposure, expressed as a percentage. Most people are surprised by how large it is.
2. Automate your own repetitive work. The person who writes the script or the knowledge base article that removes a recurring ticket type becomes the person who improves the service rather than the one whose ticket volume disappeared. This is the single highest-return action available and it is usually unblocked.
3. Move toward diagnosis deliberately. Volunteer for the tickets nobody wants: the intermittent fault, the problem that three people already failed to reproduce, the one where the user’s account of events does not add up. That work is where the durable skill lives.
4. Pick an adjacent specialism. Security operations, identity and access management, endpoint management and cloud administration all recruit heavily from support. Each pays more, each is growing, and each builds on knowledge you already have.
5. Learn the systems you work alongside. Understanding why the automated triage routed a ticket the way it did, and where it routes wrongly, turns you into the person who supervises the tooling rather than the one it replaced.
Common situations and what to do
Common situations people in this position run into, and what to do about them.
Ticket volume is falling and you are worried. Check what is falling. If it is password resets and access requests, that is automation working as intended and your remaining work is more valuable per ticket. If total volume including complex tickets is falling, the team is likely shrinking.
Your organisation deployed a chat assistant and deflection is high. Deflection rates measure tickets not raised, not problems solved. Ask what the reopen rate looks like and how many users escalate after trying it. High deflection with high escalation means problems are being delayed rather than resolved, and the cost of that shows up later as longer resolution times on tickets that finally reach a person.
You are stuck in tier one with no visible route out. This is common and it is usually solvable by making the case with evidence. Bring the automation you built, or the recurring problem you eliminated, rather than asking for a promotion in the abstract.
Every remaining ticket feels harder than it used to. That is the expected consequence of removing the simple ones, and it is worth naming to your manager explicitly. Handling twenty difficult tickets is not the same job as handling forty mixed ones, and any staffing model that treats them as equivalent will burn people out within a year.
Automated diagnosis keeps sending you down the wrong path. Treat the tool’s suggestion as one hypothesis rather than a conclusion. Systems trained on common cases are confidently wrong about uncommon ones, and the tickets reaching a human are disproportionately uncommon by definition.
Why support automated before most office work
The sequence here is worth understanding, because it explains both how far the decline goes and where it stops.
Three properties made support unusually easy to automate, and they arrived in order.
The tickets were categorised. Support has run on ticketing systems for decades, which means there was a labelled record of every request and its resolution. That is training data nobody had to create deliberately, and it is why automated triage worked here earlier than in work with no equivalent record.
The high-volume cases were identical. Password resets are the same request every time. Once self-service handled them, a large fraction of total volume left the human queue in a single step, without any sophisticated technology.
The outcome is verifiable. Either the user can log in or they cannot. Where success is unambiguous, automation can be trusted quickly, and organisations adopt it faster than in work where nobody can tell afterwards whether the output was right.
Now look at what is left. Intermittent faults are not categorised, because each one is different. Physical work has no digital record to learn from. And whether a difficult conversation with a stressed executive went well is not something a system can evaluate. The properties that made the first eighty percent easy are absent from the remainder, which is why the projection is a slow decline rather than an elimination.
How adjacent IT roles compare
For context on where the wider technology sector is heading, three comparisons are useful. Information security analysts are forecast to grow 21 percent at median pay of $129,180, which is the strongest adjacent destination. Network and computer systems administrators decline 4 percent at $99,130, so that route is stable in pay but not in headcount. Computer network architects grow 8 percent at $134,050.
The pattern across all of them is the same one visible in support itself. Work defined by carrying out a known procedure is contracting. Work defined by designing systems, or by being accountable for security and reliability, is growing and pays substantially more.
For how exposure is assessed generally, the Bureau publishes AI exposure categories for 831 occupations and states plainly that exposure “does not imply job loss, productivity gains, automation probability, or wage effects.” Support scores as highly exposed, and the 3 percent decline rather than a collapse is a useful reminder that exposure and outcome are different measures.
Building the judgement to supervise these systems, rather than compete with them, is a transferable skill. If you want a structured route in, explore Coursiv AI lessons and check current plan details on the official site.
What the remaining work actually looks like
Three categories account for most of what still requires a person, and they are worth recognising because they are where to aim.
Diagnosis under uncertainty. The user says the application is slow. It is slow for them and fast for everyone else, monitoring reports everything healthy, and it started on Tuesday. Nothing in that description identifies the problem, and finding it requires forming hypotheses and eliminating them. This is the least automatable part of the job.
Physical and local work. Hardware failures, peripherals, network points, device replacement and anything requiring someone to be in a building. Steady rather than growing, and completely resistant to automation.
The human side. A senior person whose presentation will not display in ten minutes needs the problem fixed and needs to be managed. Someone who has just fallen for a phishing email needs to report it without feeling they will be punished for it, because the alternative is that they do not report it at all. Neither of those is a technical interaction.
That last category is underrated. Support staff are frequently the first to hear about a security incident, and whether they hear about it early depends on whether people find them approachable. No self-service portal replicates that.
It also has a measurable consequence that rarely appears in staffing models. The gap between a phishing click and a report to security is the window in which the incident can be contained cheaply. Organisations where people are comfortable admitting a mistake to a familiar face close that gap in minutes. Organisations that removed the familiar face in favour of a form close it in hours, and the difference in what an incident ends up costing is not small.
Where the decline is likely to stop
The 3 percent figure is a projection rather than a floor, but the shape of what remains suggests why it is a slow decline rather than a steeper one.
Every organisation with physical premises and physical devices needs someone able to attend them. Every organisation subject to security requirements needs someone accountable for endpoint state and access. And every organisation with people in it generates problems that are described badly, which is the condition under which diagnosis is required rather than lookup.
Those three requirements do not scale down to zero as tooling improves. They scale with the size of the organisation and the complexity of what it runs, both of which have been rising. That is the counterweight against automation, and it is why the projection is minus 3 rather than minus 30.
Your next step
Sort one month of your own tickets into the three piles described above and calculate the percentage that was resolved by following a documented procedure. That number is your exposure, measured rather than guessed, and it tells you how urgent the next move is.
Then take the single most repetitive ticket type in that pile and eliminate it, with a script, a self-service option or a properly written article. Doing that once changes how you are seen, and it is the most reliable way anyone moves from handling the queue to improving the service.