The Job That Stopped Existing
One number moved that nobody in this industry put on a slide, and it describes the first job you hired somebody into.
Start with the Center for an Urban Future, which published a report this month on entry-level technology work. In New York City, entry-level tech job postings are down forty-nine percent since 2022. That is a steeper drop in total jobs posted than any other field they looked at. And they ran a control, which is the part worth holding onto. Across occupations with the highest exposure to artificial intelligence, entry-level postings fell twenty-nine percent over that same period. Across occupations with the least exposure, entry-level postings went up twenty-one percent. Same city, same four years, same economy, opposite directions.
Now back it up with the federal numbers, from the Bureau of Labor Statistics. Computer support specialists — that is the actual job title, the help desk, the person who takes the ticket — nine hundred and three thousand of them working in 2025. The projection through 2035 is a decline of three percent. Twenty-four thousand three hundred jobs, gone. Median pay, sixty-one thousand eight hundred and sixty dollars. And the education BLS lists for it is some college, or a high school diploma plus certifications.
Then look at the rest of that same federal table, the same ten years. Information security analysts, projected up twenty-one percent — forty thousand six hundred jobs added. Software developers, up ten percent, a hundred and seventy-four thousand seven hundred. Data scientists, up thirty-five percent, ninety-five thousand four hundred. Add those three together: the top of the ladder gains three hundred and ten thousand jobs over the same decade the bottom rung loses twenty-four thousand. One source, one data set, and the lines run in opposite directions depending on where you look.
One more, and this one is private hiring rather than projection. SignalFire, a venture capital firm, publishes an annual report on technology talent — that is their own research about their own market, so weigh it accordingly. New graduate hiring at the largest technology companies is down about sixty-five percent against 2019. At early-stage startups, down about seventy-six.
Four counts. Every one of them says the same thing about the bottom of the ladder, and not one of them says it about the top.
So that is what the counting shows. One of those four sources went further, and said why.
Built Out Of Easy Tickets
The Bureau of Labor Statistics does not usually tell you why. On this one, it does. Employment of computer user support specialists is projected to decline, it writes, as organizations continue to implement automated tools, such as chatbots, for troubleshooting. And then it adds a sentence that reads like consolation and isn’t: the automation may free up some of those specialists to handle more complex cases — but fewer are expected to be needed overall.
Here is why it lands exactly there. Fast Company made the argument plainly this month: this technology does not remove jobs. It unbundles tasks. It reaches into a role, pulls out individual pieces of work, and leaves whatever is left still wearing the job title.
So think about what the junior position in your shop actually is. It was never really a job. It was a bundle — the password resets, the printer, the new-hire setup, the ticket with a known answer — assembled on purpose out of the most repeatable and best-documented work in the building. You put those things together because they are the things a person can learn by doing them badly a few times with somebody watching.
And repeatable and well-documented is the same specification as automates first. Not a similar specification. The same one. The criteria that made that work teachable are the criteria that made it the first target. Nobody in this industry sat down to dismantle the apprenticeship. They sat down to clear the easy tickets, and the apprenticeship was made of easy tickets.
Watch it land. TeamViewer shipped a support agent called Tia Troubleshooting — that is TeamViewer describing TeamViewer’s own product — which no longer recommends the fix, it executes it inside the session, with a technician approving each one and the approval logged for audit. The moment where somebody learned by doing is now a moment where somebody senior clicks approve.
And RDE Technologies, an MSP in Oceanside, New York, announced it is running a tool called Vight — again, the vendor’s own release — that takes the recording of every support call, writes the ticket note, posts the time entry to the PSA, and flags the calls where a client sounds frustrated. Founder Nathan Spitalny says it lets them coach from real calls instead of secondhand accounts. That is a genuinely good thing to be able to do, and I want to be careful here — he is describing coaching the technicians he already has. But the capability underneath it is that the lesson inside a service call can now be captured without anybody sitting next to that call to learn it.
So the rung didn’t get harder to climb. It got absorbed — and everything above it still requires somebody to have climbed it.
Which lands somewhere very specific — in a sentence almost every shop in this industry has already written down.
The Engineer Isn’t For Sale
Every MSP business plan written in the last twenty years has the same sentence buried somewhere inside it: we hire at the help desk and grow them into engineers. That sentence is a supply assumption. It has quietly stopped being true, and the obvious workaround is closing at the same time.
The workaround is to skip the growing and buy the engineer. So look at that end of the market. Geoffrey Sanzenbacher at Boston College’s Center for Retirement Research asked whether the careers of older workers are being cut short by AI, and counted how often people over fifty-five left the workforce between 2014 and 2025. Computer programmers: exits up more than twenty-five percent. Accountants and auditors, twenty-two. People who paint things for a living, about two. And after ChatGPT launched, the AI-exposed jobs saw relative increases in people transitioning out of work specifically into unemployment — not into retirement. Sanzenbacher is careful about this, and says the impact of AI on these workers remains an open question. Take the caution. Then notice which occupation posted the largest number on the board.
And the junior you might still hire got more expensive on the way. Research commissioned by Iris Software Group, surveying five hundred and eleven UK human resources leaders and five hundred UK employees — a software company funding research about its own market, so weigh it accordingly — found new starters arriving at or above what the people already sitting there earn. Among employees who found out what the new hire was paid, a quarter were earning exactly the same, and one in six were earning less than the newcomer. And here is the line that matters for this episode: seventy-one percent of people two to five years into their careers said it is hard to feel motivated to train somebody being paid nearly as much as they are. Eighty-four percent of the HR leaders are worried about losing those employees. Twenty percent have a plan.
Both ends, in the same window. Buying seniority is harder because the seniors are leaving. Buying junior is dearer because the floor moved. The engineer you need in 2029 is not currently for sale.
So here is the choice, and it is a pricing decision rather than a hiring one.
Manufacture the engineer. That means deliberately holding back work the AI in your own stack could already close — real tickets, on real clients — and routing them to a person who needs to learn on them. That is not a training line item. That is gross margin you are choosing not to collect, every month, visibly, on purpose. Decide what percentage you can carry and write the number down.
Or buy the engineer. Which means competing on salary against vendors, against integrators, and against your own clients, who are all hiring the same person you are — out of a pool the automation is thinning from both ends.
Pay for seniority in margin now, or pay for it in salary later. One of those you can budget.
And the budget is the part that is already set — by you, in agreements you have already signed.
Because the margin you reserve for training has to come out of a rate you already quoted, and every agreement you signed before this year priced labor you assumed you could hire cheaply. Go through your next three renewals and find the delivery rate that was built on a help desk hire at sixty-two thousand dollars, because that is the number that no longer holds. The shops that move the delivery line now, while they can still explain exactly why, fund the training out of price — the ones that wait fund it out of profit.
What to Consider
Put an actual number on the reserved capacity and carry it as a named line. Take the percentage of delivery hours you intend to hold back from automation, multiply it by your blended rate, and that is the annual cost of manufacturing an engineer — a real figure, not a training budget. Put it in your P&L under its own name before the first month it shows up, because if it arrives as unexplained margin drift you will quietly cancel it in the third month when a ticket runs long and a client asks why.
Sort your accounts into the ones you can train on and the ones you cannot, then price them differently. Training requires a forgiving tolerance, predictable work and a client who will accept a slightly slower resolution on a known-answer ticket — and some of your accounts have none of those. Those accounts should be carrying full senior rates, because they get senior delivery on every ticket by necessity. The ones that can absorb a learner are subsidizing your bench, and they should be priced knowing that, not by accident.
Move the rate at the next renewal while the explanation is still worth something. Right now “we are funding the engineers who will still be here in five years” is a specific, defensible reason almost nobody in your market is giving, and it separates you from the shop down the road quoting the same stack for less. In two or three years, when every provider is short of senior people, that same sentence reads as the industry’s standard excuse and buys you nothing. The reason has a shelf life, and it is shorter than the problem.
Picture the shop that got this right. It is 2029, and it has three engineers who came up inside the building, know every client environment by name, and did not arrive from anybody’s job board. It is not bidding against four competitors on price, because the other three cannot staff the work. That shop did not do anything clever. It just kept paying for the one thing everybody else stopped paying for. And the shops that stopped training are not merely short an engineer. They are the reason the price of one goes up.
If this trend continues, by the 2029 renewal cycle, the shops that held training capacity out of automation will be quoting a premium for engineering continuity and getting it, while the ones that took the full margin in 2026 will be buying that same continuity on the open market at a price somebody else sets.