Sitting at a computer for eight hours was never the point
The way most of us use computers, as the clerk between messy reality and rigid systems, is a thirty-year-old habit that AI is quietly ending. Computing stays. The job of being middleware does not.
I have been in corporate IT since the 1990s. Lately, the more time I spend with AI, the more convinced I am that the arrangement we have taken for granted since then is coming to an end: people sitting in front of a computer for eight hours a day, translating the world into clicks and keystrokes.
Not computing, networks or data centres. The game. The way humans use the digital world.
That world arrived in offices in the nineties and became the default shape of work. You learned the ritual. You sat down. You opened the console, the ticket system, the vendor GUI, the spreadsheet. You were the interface layer between messy reality and rigid systems. Competence meant knowing the path through the menus, the right incantation, the change window, the group policy, the runbook from 2014 that nobody had updated.
That arrangement is what is finishing.
Bits first, atoms later
Elon Musk has drawn the line more clearly than most of the industry is willing to. Digital work, in his words "just someone at a computer doing something", he expects AI to take over "like lightning", the way computers took over manual calculation, only faster. Work that means shaping atoms lasts longer: welding, electrical work, anything done with hands.
I agree with that cut, and I am not repeating it to borrow a headline. I am repeating it because it matches what I see from inside the work. The physical side of my own patch is the same: the rack that has to be touched, the fibre that has to be pulled.
Most of what corporate IT actually does is information work wearing an infrastructure costume. Write the ticket. Change the policy. Reconcile the CMDB. Approve the access request. Draft the runbook. Map the dependency. Explain the outage. Generate the compliance pack. That is all bits. An agent that can read the same consoles, the same documents, the same logs and the same chat threads can already do a large fraction of it. The remaining hard parts are the ones that punch through into atoms or into other humans: a failed disk in a remote rack, a fibre cut, a vendor who will only talk to a named person, a regulator who wants a signature, a customer whose process lives in someone's head.
So the claim is not that computers disappear. The claim is that the human-as-clerk-of-the-machine era is ending.
Keyboards will not vanish. They will stop being the job.
The next interface is conversation. You state the outcome. The system traverses the tools. Voice is the obvious version of that, and it is already good enough that a lot of typing is optional rather than required. That does not mean every keyboard goes in the bin in 2027. Precision work still likes keys. Quiet rooms, legal records, diffs you actually need to see, and anything you would not want spoken aloud all keep a screen around.
The honest end-state is a world where the keyboard becomes a specialist tool, like a manual gearbox. Still there, but no longer the default way most people spend the day.
Once that is reliable enough, the old digital world is finished even if some people still type. The unit of work stops being "open the app, click the thing, fill the form" and becomes "say what should be true and let an agent make it true."
The IT industry does not vanish. It polarises.
This is the part that gets flattened into "AI takes all the jobs." Headcount in commodity digital operations will compress. A lot of helpdesk, L1/L2, routine sysadmin, report writing and "I configured the thing" work falls. That is not a moral judgement. It is what happens when the scarce skill stops being "I know how to make the box do the thing."
What does not fall:
- Deciding what should exist.
- Owning the consequences when it is wrong.
- Connecting digital systems to atoms, regulation, money and other humans.
- Keeping the physical substrate running: power, cooling, fibre, radios, devices that fail in rain.
- Judging whether the model's output is actually true in this business, this jurisdiction, this customer.
That last one is the job that grows. Models are already better than the median operator at a lot of digital tasks. They are not yet reliably better in the tail: the weird outage, the hostile tenant, the one process that only exists as tribal knowledge. Until verification is cheap, a human still has to stand behind the result.
The 1990s game rewarded people who could hold a large, brittle system in their head and type the right incantation. The next game rewards people who can specify outcomes, instrument them, and accept responsibility when the specification was incomplete. Same industry on the org chart, different personality, different career.
Why this will be slower than the capability curve, and why that does not save the old job
Capability is not adoption.
Banks, hospitals, councils and anyone with a board will not let an agent close a change or move money just because it can. Someone still has to be the accountable human. The digital estate since the nineties is a pile of half-documented APIs, on-prem leftovers, and vendors who charge extra for the "AI-ready" connector. Agents eat clean digital work first. Dirty estates take years.
Those frictions buy time. They do not buy a return to the old arrangement. They create a messy decade where the people who understand both the legacy estate and the agents are unusually valuable. After that, the value sits in the seams: where bits meet atoms, where a confident answer has to become a safe action, where a small business owner still wants a person who will pick up the phone when the model does something stupid.
What "technical competence" means after the ritual ends
It does not mean "I can still SSH."
It means you can describe a business process so an agent can execute it without inventing a step that only existed in a runbook nobody read. It means you can tell when the model is confidently wrong about your environment. It means you can design the human checkpoints that actually catch harm instead of theatre. It means you can still walk into a comms room and know what the blinking light means. It means you can explain the system to a non-technical owner and stand behind it.
The people who only ever learned the ritual of the current tools get replaced. The people who learned how systems fail, and what the business actually needs, get leveraged. That has been the pattern with every previous automation wave. The difference now is speed, and the fact that the thing being automated is cognition sitting on a screen, which is most of white-collar IT.
The game that is ending
The digital world that arrived in offices in the 1990s trained a generation to sit, click and be middleware. That was a historically specific way of working. It felt permanent because it lasted thirty years and paid well.
It was never the point of computing. It was a bottleneck. AI is removing the bottleneck.
What replaces it is not a world without computers. It is computers you talk to, that talk to other computers, and that still need a smaller number of people who understand both the conversation and the atoms underneath it. Smaller industry by headcount. Different industry by skill.
I do not think that is a eulogy. I think it is a job-description change that arrived earlier than most of the profession is willing to admit.